{
"cells": [
{
"cell_type": "markdown",
"id": "d602e75d",
"metadata": {},
"source": [
"# 0. Import Dependencies"
]
},
{
"cell_type": "code",
"execution_count": 1,
"id": "2d97e5ec",
"metadata": {},
"outputs": [
{
"name": "stderr",
"output_type": "stream",
"text": [
"c:\\Users\\cpras\\anaconda3\\envs\\AItrainer\\lib\\site-packages\\numpy\\_distributor_init.py:30: UserWarning: loaded more than 1 DLL from .libs:\n",
"c:\\Users\\cpras\\anaconda3\\envs\\AItrainer\\lib\\site-packages\\numpy\\.libs\\libopenblas.FB5AE2TYXYH2IJRDKGDGQ3XBKLKTF43H.gfortran-win_amd64.dll\n",
"c:\\Users\\cpras\\anaconda3\\envs\\AItrainer\\lib\\site-packages\\numpy\\.libs\\libopenblas.WCDJNK7YVMPZQ2ME2ZZHJJRJ3JIKNDB7.gfortran-win_amd64.dll\n",
" warnings.warn(\"loaded more than 1 DLL from .libs:\"\n"
]
}
],
"source": [
"import cv2\n",
"import numpy as np\n",
"import os\n",
"from matplotlib import pyplot as plt\n",
"import time\n",
"import mediapipe as mp\n",
"import tensorflow as tf\n",
"import math\n",
"\n",
"from sklearn.model_selection import train_test_split\n",
"from sklearn.metrics import multilabel_confusion_matrix, accuracy_score, classification_report\n",
"from tensorflow.keras.utils import to_categorical\n",
"\n",
"import tensorflow as tf\n",
"from tensorflow.keras import backend as K\n",
"from tensorflow.keras.callbacks import TensorBoard, EarlyStopping, ReduceLROnPlateau, ModelCheckpoint\n",
"\n",
"from tensorflow.keras.models import Sequential, Model\n",
"\n",
"from tensorflow.keras.layers import (LSTM, Dense, Concatenate, Attention, Dropout, Softmax,\n",
" Input, Flatten, Activation, Bidirectional, Permute, multiply, \n",
" ConvLSTM2D, MaxPooling3D, TimeDistributed, Conv2D, MaxPooling2D)\n",
"\n",
"from scipy import stats\n",
"\n",
"# disable some of the tf/keras training warnings \n",
"os.environ['TF_CPP_MIN_LOG_LEVEL'] = \"3\"\n",
"tf.get_logger().setLevel(\"ERROR\")\n",
"tf.autograph.set_verbosity(1)\n",
"\n",
"# suppress untraced functions warning\n",
"import absl.logging\n",
"absl.logging.set_verbosity(absl.logging.ERROR)"
]
},
{
"cell_type": "markdown",
"id": "55f470e2",
"metadata": {},
"source": [
"# 1. Keypoints using MP Pose"
]
},
{
"cell_type": "code",
"execution_count": 2,
"id": "20cde117",
"metadata": {},
"outputs": [],
"source": [
"# Pre-trained pose estimation model from Google Mediapipe\n",
"mp_pose = mp.solutions.pose\n",
"\n",
"# Supported Mediapipe visualization tools\n",
"mp_drawing = mp.solutions.drawing_utils"
]
},
{
"cell_type": "code",
"execution_count": 3,
"id": "716e9f8e",
"metadata": {},
"outputs": [],
"source": [
"def mediapipe_detection(image, model):\n",
" \"\"\"\n",
" This function detects human pose estimation keypoints from webcam footage\n",
" \n",
" \"\"\"\n",
" image = cv2.cvtColor(image, cv2.COLOR_BGR2RGB) # COLOR CONVERSION BGR 2 RGB\n",
" image.flags.writeable = False # Image is no longer writeable\n",
" results = model.process(image) # Make prediction\n",
" image.flags.writeable = True # Image is now writeable \n",
" image = cv2.cvtColor(image, cv2.COLOR_RGB2BGR) # COLOR COVERSION RGB 2 BGR\n",
" return image, results"
]
},
{
"cell_type": "code",
"execution_count": 4,
"id": "9bd7ba58",
"metadata": {},
"outputs": [],
"source": [
"def draw_landmarks(image, results):\n",
" \"\"\"\n",
" This function draws keypoints and landmarks detected by the human pose estimation model\n",
" \n",
" \"\"\"\n",
" mp_drawing.draw_landmarks(image, results.pose_landmarks, mp_pose.POSE_CONNECTIONS,\n",
" mp_drawing.DrawingSpec(color=(245,117,66), thickness=2, circle_radius=2), \n",
" mp_drawing.DrawingSpec(color=(245,66,230), thickness=2, circle_radius=2) \n",
" )"
]
},
{
"cell_type": "code",
"execution_count": 5,
"id": "c0ebe952",
"metadata": {},
"outputs": [],
"source": [
"cap = cv2.VideoCapture(0) # camera object\n",
"HEIGHT = int(cap.get(cv2.CAP_PROP_FRAME_HEIGHT)) # webcam video frame height\n",
"WIDTH = int(cap.get(cv2.CAP_PROP_FRAME_WIDTH)) # webcam video frame width\n",
"FPS = int(cap.get(cv2.CAP_PROP_FPS)) # webcam video fram rate \n",
"\n",
"# Set and test mediapipe model using webcam\n",
"with mp_pose.Pose(min_detection_confidence=0.5, min_tracking_confidence=0.5) as pose:\n",
" while cap.isOpened():\n",
"\n",
" # Read feed\n",
" ret, frame = cap.read()\n",
" \n",
" # Make detection\n",
" image, results = mediapipe_detection(frame, pose)\n",
" \n",
" # Extract landmarks\n",
" try:\n",
" landmarks = results.pose_landmarks.landmark\n",
" except:\n",
" pass\n",
" \n",
" # Render detections\n",
" draw_landmarks(image, results) \n",
" \n",
" # Display frame on screen\n",
" cv2.imshow('OpenCV Feed', image)\n",
" \n",
" # Exit / break out logic\n",
" if cv2.waitKey(10) & 0xFF == ord('q'):\n",
" break\n",
"\n",
" cap.release()\n",
" cv2.destroyAllWindows()"
]
},
{
"cell_type": "markdown",
"id": "f9fb32f3",
"metadata": {},
"source": [
"# 2. Extract Keypoints"
]
},
{
"cell_type": "code",
"execution_count": 6,
"id": "a81823f6",
"metadata": {},
"outputs": [],
"source": [
"# Recollect and organize keypoints from the test\n",
"pose = []\n",
"for res in results.pose_landmarks.landmark:\n",
" test = np.array([res.x, res.y, res.z, res.visibility])\n",
" pose.append(test)"
]
},
{
"cell_type": "code",
"execution_count": 7,
"id": "cd92eee3",
"metadata": {},
"outputs": [],
"source": [
"# 33 landmarks with 4 values (x, y, z, visibility)\n",
"num_landmarks = len(landmarks)\n",
"num_values = len(test)\n",
"num_input_values = num_landmarks*num_values"
]
},
{
"cell_type": "code",
"execution_count": 8,
"id": "9dad5b8c",
"metadata": {},
"outputs": [],
"source": [
"# This is an example of what we would use as an input into our AI models\n",
"pose = np.array([[res.x, res.y, res.z, res.visibility] for res in results.pose_landmarks.landmark]).flatten() if results.pose_landmarks else np.zeros(33*4)"
]
},
{
"cell_type": "code",
"execution_count": 9,
"id": "6f4d1079",
"metadata": {},
"outputs": [],
"source": [
"def extract_keypoints(results):\n",
" \"\"\"\n",
" Processes and organizes the keypoints detected from the pose estimation model \n",
" to be used as inputs for the exercise decoder models\n",
" \n",
" \"\"\"\n",
" pose = np.array([[res.x, res.y, res.z, res.visibility] for res in results.pose_landmarks.landmark]).flatten() if results.pose_landmarks else np.zeros(33*4)\n",
" return pose"
]
},
{
"cell_type": "markdown",
"id": "03b907e8",
"metadata": {},
"source": [
"# 3. Setup Folders for Collection"
]
},
{
"cell_type": "code",
"execution_count": 10,
"id": "ddcaecfd",
"metadata": {},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
"c:\\Users\\cpras\\Documents\\GitHub\\AI_Personal_Trainer\\data\n"
]
}
],
"source": [
"# Path for exported data, numpy arrays\n",
"DATA_PATH = os.path.join(os. getcwd(),'data') \n",
"print(DATA_PATH)\n",
"\n",
"# make directory if it does not exist yet\n",
"if not os.path.exists(DATA_PATH):\n",
" os.makedirs(DATA_PATH)\n",
"\n",
"# Actions/exercises that we try to detect\n",
"actions = np.array(['curl', 'press', 'squat'])\n",
"num_classes = len(actions)\n",
"\n",
"# How many videos worth of data\n",
"no_sequences = 50\n",
"\n",
"# Videos are going to be this many frames in length\n",
"sequence_length = FPS*1\n",
"\n",
"# Folder start\n",
"# Change this to collect more data and not lose previously collected data\n",
"start_folder = 101"
]
},
{
"cell_type": "code",
"execution_count": 62,
"id": "fed6b275",
"metadata": {},
"outputs": [],
"source": [
"# Build folder paths\n",
"for action in actions: \n",
" for sequence in range(start_folder,no_sequences+start_folder):\n",
" try: \n",
" os.makedirs(os.path.join(DATA_PATH, action, str(sequence))) \n",
" except:\n",
" pass"
]
},
{
"cell_type": "markdown",
"id": "7622b573",
"metadata": {},
"source": [
"# 4. Collect Keypoint Values for Training and Testing"
]
},
{
"cell_type": "code",
"execution_count": 11,
"id": "d224561f",
"metadata": {},
"outputs": [],
"source": [
"# Colors associated with each exercise (e.g., curls are denoted by blue, squats are denoted by orange, etc.)\n",
"colors = [(245,117,16), (117,245,16), (16,117,245)]"
]
},
{
"cell_type": "code",
"execution_count": 64,
"id": "41b81490",
"metadata": {},
"outputs": [],
"source": [
"# Collect Training Data\n",
"\n",
"cap = cv2.VideoCapture(0)\n",
"# Set mediapipe model \n",
"with mp_pose.Pose(min_detection_confidence=0.5, min_tracking_confidence=0.5) as pose:\n",
" # Loop through actions\n",
" for idx, action in enumerate(actions):\n",
" # Loop through sequences (i.e., videos)\n",
" for sequence in range(start_folder, start_folder+no_sequences):\n",
" # Loop through video length (i.e, sequence length)\n",
" for frame_num in range(sequence_length):\n",
" # Read feed\n",
" ret, frame = cap.read()\n",
" \n",
" # Make detection\n",
" image, results = mediapipe_detection(frame, pose)\n",
"\n",
" # Extract landmarks\n",
" try:\n",
" landmarks = results.pose_landmarks.landmark\n",
" except:\n",
" pass\n",
" \n",
" # Render detections\n",
" draw_landmarks(image, results) \n",
"\n",
" # Apply visualization logic\n",
" if frame_num == 0: # If first frame in sequence, print that you're starting a new data collection and wait 500 ms\n",
" cv2.putText(image, 'STARTING COLLECTION', (120,200), \n",
" cv2.FONT_HERSHEY_SIMPLEX, 1, (0,255, 0), 4, cv2.LINE_AA)\n",
" \n",
" cv2.putText(image, 'Collecting {} Video # {}'.format(action, sequence), (15,30), \n",
" cv2.FONT_HERSHEY_SIMPLEX, 1, (255, 255, 255), 8, cv2.LINE_AA)\n",
" cv2.putText(image, 'Collecting {} Video # {}'.format(action, sequence), (15,30), \n",
" cv2.FONT_HERSHEY_SIMPLEX, 1, colors[idx], 4, cv2.LINE_AA)\n",
" \n",
" # Show to screen\n",
" cv2.imshow('OpenCV Feed', image)\n",
" cv2.waitKey(500)\n",
" else: \n",
" cv2.putText(image, 'Collecting {} Video # {}'.format(action, sequence), (15,30), \n",
" cv2.FONT_HERSHEY_SIMPLEX, 1, (255, 255, 255), 8, cv2.LINE_AA)\n",
" cv2.putText(image, 'Collecting {} Video # {}'.format(action, sequence), (15,30), \n",
" cv2.FONT_HERSHEY_SIMPLEX, 1, colors[idx], 4, cv2.LINE_AA)\n",
" \n",
" # Show to screen\n",
" cv2.imshow('OpenCV Feed', image)\n",
"\n",
" # Export keypoints (sequence + pose landmarks)\n",
" keypoints = extract_keypoints(results)\n",
" npy_path = os.path.join(DATA_PATH, action, str(sequence), str(frame_num))\n",
" np.save(npy_path, keypoints)\n",
"\n",
" # Break gracefully\n",
" if cv2.waitKey(10) & 0xFF == ord('q'):\n",
" break\n",
" \n",
" cap.release()\n",
" cv2.destroyAllWindows()"
]
},
{
"cell_type": "code",
"execution_count": 12,
"id": "4b016129",
"metadata": {},
"outputs": [],
"source": [
"cap.release()\n",
"cv2.destroyAllWindows()"
]
},
{
"cell_type": "markdown",
"id": "d01ec993",
"metadata": {},
"source": [
"# 5. Preprocess Data and Create Labels/Features"
]
},
{
"cell_type": "code",
"execution_count": 13,
"id": "cad528c8",
"metadata": {},
"outputs": [],
"source": [
"label_map = {label:num for num, label in enumerate(actions)}"
]
},
{
"cell_type": "code",
"execution_count": 14,
"id": "a0add3fb",
"metadata": {},
"outputs": [],
"source": [
"# Load and organize recorded training data\n",
"sequences, labels = [], []\n",
"for action in actions:\n",
" for sequence in np.array(os.listdir(os.path.join(DATA_PATH, action))).astype(int):\n",
" window = []\n",
" for frame_num in range(sequence_length): \n",
" # LSTM input data\n",
" res = np.load(os.path.join(DATA_PATH, action, str(sequence), \"{}.npy\".format(frame_num)))\n",
" window.append(res) \n",
" \n",
" sequences.append(window)\n",
" labels.append(label_map[action])"
]
},
{
"cell_type": "code",
"execution_count": 15,
"id": "ab459ce4",
"metadata": {},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
"(450, 30, 132) (450, 3)\n"
]
}
],
"source": [
"# Make sure first dimensions of arrays match\n",
"X = np.array(sequences)\n",
"y = to_categorical(labels).astype(int)\n",
"print(X.shape, y.shape)"
]
},
{
"cell_type": "code",
"execution_count": 16,
"id": "5ac49993",
"metadata": {},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
"(405, 30, 132) (405, 3)\n"
]
}
],
"source": [
"# Split into training, validation, and testing datasets\n",
"X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=0.10, random_state=1)\n",
"print(X_train.shape, y_train.shape)\n",
"X_train, X_val, y_train, y_val = train_test_split(X_train, y_train, test_size=15/90, random_state=2)"
]
},
{
"cell_type": "markdown",
"id": "e53ae03d",
"metadata": {},
"source": [
"# 6. Build and Train Neural Networks"
]
},
{
"cell_type": "code",
"execution_count": 17,
"id": "912f3153",
"metadata": {},
"outputs": [],
"source": [
"# Callbacks to be used during neural network training \n",
"es_callback = EarlyStopping(monitor='val_loss', min_delta=5e-4, patience=10, verbose=0, mode='min')\n",
"lr_callback = ReduceLROnPlateau(monitor='val_loss', factor=0.2, patience=5, min_lr=0.00001, verbose=0, mode='min')\n",
"chkpt_callback = ModelCheckpoint(filepath=DATA_PATH, monitor='val_loss', verbose=0, save_best_only=True, \n",
" save_weights_only=False, mode='min', save_freq=1)\n",
"\n",
"# Optimizer\n",
"opt = tf.keras.optimizers.Adam(learning_rate=0.01)\n",
"\n",
"# some hyperparamters\n",
"batch_size = 32\n",
"max_epochs = 500"
]
},
{
"cell_type": "markdown",
"id": "c1ca0cad",
"metadata": {},
"source": [
"## 6a. LSTM"
]
},
{
"cell_type": "code",
"execution_count": 18,
"id": "730f987b",
"metadata": {},
"outputs": [],
"source": [
"# Set up Tensorboard logging and callbacks\n",
"NAME = f\"ExerciseRecognition-LSTM-{int(time.time())}\"\n",
"log_dir = os.path.join(os.getcwd(), 'logs', NAME,'')\n",
"tb_callback = TensorBoard(log_dir=log_dir)\n",
"\n",
"callbacks = [tb_callback, es_callback, lr_callback, chkpt_callback]"
]
},
{
"cell_type": "code",
"execution_count": 19,
"id": "ae7595c1",
"metadata": {},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
"Model: \"sequential\"\n",
"_________________________________________________________________\n",
" Layer (type) Output Shape Param # \n",
"=================================================================\n",
" lstm (LSTM) (None, 30, 128) 133632 \n",
" \n",
" lstm_1 (LSTM) (None, 30, 256) 394240 \n",
" \n",
" lstm_2 (LSTM) (None, 128) 197120 \n",
" \n",
" dense (Dense) (None, 128) 16512 \n",
" \n",
" dense_1 (Dense) (None, 64) 8256 \n",
" \n",
" dense_2 (Dense) (None, 3) 195 \n",
" \n",
"=================================================================\n",
"Total params: 749,955\n",
"Trainable params: 749,955\n",
"Non-trainable params: 0\n",
"_________________________________________________________________\n",
"None\n"
]
}
],
"source": [
"lstm = Sequential()\n",
"lstm.add(LSTM(128, return_sequences=True, activation='relu', input_shape=(sequence_length, num_input_values)))\n",
"lstm.add(LSTM(256, return_sequences=True, activation='relu'))\n",
"lstm.add(LSTM(128, return_sequences=False, activation='relu'))\n",
"lstm.add(Dense(128, activation='relu'))\n",
"lstm.add(Dense(64, activation='relu'))\n",
"lstm.add(Dense(actions.shape[0], activation='softmax'))\n",
"print(lstm.summary())"
]
},
{
"cell_type": "code",
"execution_count": 20,
"id": "8a10e698",
"metadata": {},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
"Epoch 1/500\n",
"11/11 [==============================] - 58s 5s/step - loss: 1.1959 - categorical_accuracy: 0.4718 - val_loss: 1.1662 - val_categorical_accuracy: 0.3529 - lr: 0.0010\n",
"Epoch 2/500\n",
"11/11 [==============================] - 54s 5s/step - loss: 0.9662 - categorical_accuracy: 0.3887 - val_loss: 0.7277 - val_categorical_accuracy: 0.5441 - lr: 0.0010\n",
"Epoch 3/500\n",
"11/11 [==============================] - 55s 5s/step - loss: 0.7514 - categorical_accuracy: 0.7092 - val_loss: 0.9325 - val_categorical_accuracy: 0.5882 - lr: 0.0010\n",
"Epoch 4/500\n",
"11/11 [==============================] - 54s 5s/step - loss: 0.8151 - categorical_accuracy: 0.7537 - val_loss: 4.7314 - val_categorical_accuracy: 0.4706 - lr: 0.0010\n",
"Epoch 5/500\n",
"11/11 [==============================] - 55s 5s/step - loss: 0.8428 - categorical_accuracy: 0.7567 - val_loss: 0.6531 - val_categorical_accuracy: 0.9706 - lr: 0.0010\n",
"Epoch 6/500\n",
"11/11 [==============================] - 55s 5s/step - loss: 0.6416 - categorical_accuracy: 0.8427 - val_loss: 0.7113 - val_categorical_accuracy: 0.6765 - lr: 0.0010\n",
"Epoch 7/500\n",
"11/11 [==============================] - 55s 5s/step - loss: 0.7782 - categorical_accuracy: 0.5964 - val_loss: 0.6241 - val_categorical_accuracy: 0.5882 - lr: 0.0010\n",
"Epoch 8/500\n",
"11/11 [==============================] - 62s 6s/step - loss: 0.4477 - categorical_accuracy: 0.8071 - val_loss: 0.3213 - val_categorical_accuracy: 0.9559 - lr: 0.0010\n",
"Epoch 9/500\n",
"11/11 [==============================] - 55s 5s/step - loss: 0.1791 - categorical_accuracy: 0.9733 - val_loss: 2.7620 - val_categorical_accuracy: 0.9412 - lr: 0.0010\n",
"Epoch 10/500\n",
"11/11 [==============================] - 55s 5s/step - loss: 1.7325 - categorical_accuracy: 0.6469 - val_loss: 0.8761 - val_categorical_accuracy: 0.8824 - lr: 0.0010\n",
"Epoch 11/500\n",
"11/11 [==============================] - 55s 5s/step - loss: 0.7803 - categorical_accuracy: 0.6558 - val_loss: 0.6232 - val_categorical_accuracy: 0.7647 - lr: 0.0010\n",
"Epoch 12/500\n",
"11/11 [==============================] - 55s 5s/step - loss: 0.5408 - categorical_accuracy: 0.8783 - val_loss: 0.2358 - val_categorical_accuracy: 0.8971 - lr: 0.0010\n",
"Epoch 13/500\n",
"11/11 [==============================] - 55s 5s/step - loss: 0.2337 - categorical_accuracy: 0.9496 - val_loss: 0.1032 - val_categorical_accuracy: 0.9853 - lr: 0.0010\n",
"Epoch 14/500\n",
"11/11 [==============================] - 55s 5s/step - loss: 1.0003 - categorical_accuracy: 0.7092 - val_loss: 1.2232 - val_categorical_accuracy: 0.2941 - lr: 0.0010\n",
"Epoch 15/500\n",
"11/11 [==============================] - 55s 5s/step - loss: 0.9862 - categorical_accuracy: 0.4896 - val_loss: 0.7796 - val_categorical_accuracy: 0.6029 - lr: 0.0010\n",
"Epoch 16/500\n",
"11/11 [==============================] - 56s 5s/step - loss: 0.7276 - categorical_accuracy: 0.6558 - val_loss: 0.6212 - val_categorical_accuracy: 0.6324 - lr: 0.0010\n",
"Epoch 17/500\n",
"11/11 [==============================] - 55s 5s/step - loss: 0.6449 - categorical_accuracy: 0.6588 - val_loss: 0.5822 - val_categorical_accuracy: 0.6176 - lr: 0.0010\n",
"Epoch 18/500\n",
"11/11 [==============================] - 56s 5s/step - loss: 0.5741 - categorical_accuracy: 0.6736 - val_loss: 0.5254 - val_categorical_accuracy: 0.6324 - lr: 0.0010\n",
"Epoch 19/500\n",
"11/11 [==============================] - 65s 6s/step - loss: 0.5246 - categorical_accuracy: 0.6617 - val_loss: 0.4942 - val_categorical_accuracy: 0.6324 - lr: 2.0000e-04\n",
"Epoch 20/500\n",
"11/11 [==============================] - 54s 5s/step - loss: 0.4960 - categorical_accuracy: 0.6736 - val_loss: 0.4694 - val_categorical_accuracy: 0.6324 - lr: 2.0000e-04\n",
"Epoch 21/500\n",
"11/11 [==============================] - 55s 5s/step - loss: 0.4588 - categorical_accuracy: 0.6766 - val_loss: 0.4269 - val_categorical_accuracy: 0.6471 - lr: 2.0000e-04\n",
"Epoch 22/500\n",
"11/11 [==============================] - 54s 5s/step - loss: 0.4117 - categorical_accuracy: 0.6825 - val_loss: 0.3713 - val_categorical_accuracy: 0.6471 - lr: 2.0000e-04\n",
"Epoch 23/500\n",
"11/11 [==============================] - 62s 6s/step - loss: 0.3329 - categorical_accuracy: 0.7122 - val_loss: 0.2746 - val_categorical_accuracy: 0.9118 - lr: 2.0000e-04\n"
]
},
{
"data": {
"text/plain": [
"<keras.callbacks.History at 0x1d83f889850>"
]
},
"execution_count": 20,
"metadata": {},
"output_type": "execute_result"
}
],
"source": [
"lstm.compile(optimizer=opt, loss='categorical_crossentropy', metrics=['categorical_accuracy'])\n",
"lstm.fit(X_train, y_train, batch_size=batch_size, epochs=max_epochs, validation_data=(X_val, y_val), callbacks=callbacks)"
]
},
{
"cell_type": "markdown",
"id": "8f58c4d8",
"metadata": {},
"source": [
"## 6b. LSTM + Attention"
]
},
{
"cell_type": "code",
"execution_count": 21,
"id": "c6e12666",
"metadata": {},
"outputs": [],
"source": [
"# Set up Tensorboard logging and callbacks\n",
"NAME = f\"ExerciseRecognition-AttnLSTM-{int(time.time())}\"\n",
"log_dir = os.path.join(os.getcwd(), 'logs', NAME,'')\n",
"tb_callback = TensorBoard(log_dir=log_dir)\n",
"\n",
"callbacks = [tb_callback, es_callback, lr_callback, chkpt_callback]"
]
},
{
"cell_type": "code",
"execution_count": 22,
"id": "07591dac",
"metadata": {},
"outputs": [],
"source": [
"def attention_block(inputs, time_steps):\n",
" \"\"\"\n",
" Attention layer for deep neural network\n",
" \n",
" \"\"\"\n",
" # Attention weights\n",
" a = Permute((2, 1))(inputs)\n",
" a = Dense(time_steps, activation='softmax')(a)\n",
" \n",
" # Attention vector\n",
" a_probs = Permute((2, 1), name='attention_vec')(a)\n",
" \n",
" # Luong's multiplicative score\n",
" output_attention_mul = multiply([inputs, a_probs], name='attention_mul') \n",
" \n",
" return output_attention_mul"
]
},
{
"cell_type": "code",
"execution_count": 23,
"id": "c5c2e2e0",
"metadata": {},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
"Model: \"model\"\n",
"__________________________________________________________________________________________________\n",
" Layer (type) Output Shape Param # Connected to \n",
"==================================================================================================\n",
" input_1 (InputLayer) [(None, 30, 132)] 0 [] \n",
" \n",
" bidirectional (Bidirectional) (None, 30, 512) 796672 ['input_1[0][0]'] \n",
" \n",
" permute (Permute) (None, 512, 30) 0 ['bidirectional[0][0]'] \n",
" \n",
" dense_3 (Dense) (None, 512, 30) 930 ['permute[0][0]'] \n",
" \n",
" attention_vec (Permute) (None, 30, 512) 0 ['dense_3[0][0]'] \n",
" \n",
" attention_mul (Multiply) (None, 30, 512) 0 ['bidirectional[0][0]', \n",
" 'attention_vec[0][0]'] \n",
" \n",
" flatten (Flatten) (None, 15360) 0 ['attention_mul[0][0]'] \n",
" \n",
" dense_4 (Dense) (None, 512) 7864832 ['flatten[0][0]'] \n",
" \n",
" dropout (Dropout) (None, 512) 0 ['dense_4[0][0]'] \n",
" \n",
" dense_5 (Dense) (None, 3) 1539 ['dropout[0][0]'] \n",
" \n",
"==================================================================================================\n",
"Total params: 8,663,973\n",
"Trainable params: 8,663,973\n",
"Non-trainable params: 0\n",
"__________________________________________________________________________________________________\n",
"None\n"
]
}
],
"source": [
"HIDDEN_UNITS = 256\n",
"\n",
"# Input\n",
"inputs = Input(shape=(sequence_length, num_input_values))\n",
"\n",
"# Bi-LSTM\n",
"lstm_out = Bidirectional(LSTM(HIDDEN_UNITS, return_sequences=True))(inputs)\n",
"\n",
"# Attention\n",
"attention_mul = attention_block(lstm_out, sequence_length)\n",
"attention_mul = Flatten()(attention_mul)\n",
"\n",
"# Fully Connected Layer\n",
"x = Dense(2*HIDDEN_UNITS, activation='relu')(attention_mul)\n",
"x = Dropout(0.5)(x)\n",
"\n",
"# Output\n",
"x = Dense(actions.shape[0], activation='softmax')(x)\n",
"\n",
"# Bring it all together\n",
"AttnLSTM = Model(inputs=[inputs], outputs=x)\n",
"print(AttnLSTM.summary())"
]
},
{
"cell_type": "code",
"execution_count": 24,
"id": "cf2f988d",
"metadata": {},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
"Epoch 1/500\n",
"11/11 [==============================] - 146s 13s/step - loss: 0.9463 - categorical_accuracy: 0.5134 - val_loss: 0.4818 - val_categorical_accuracy: 0.8529 - lr: 0.0010\n",
"Epoch 2/500\n",
"11/11 [==============================] - 142s 13s/step - loss: 0.3754 - categorical_accuracy: 0.8576 - val_loss: 0.0902 - val_categorical_accuracy: 1.0000 - lr: 0.0010\n",
"Epoch 3/500\n",
"11/11 [==============================] - 148s 14s/step - loss: 0.0666 - categorical_accuracy: 0.9852 - val_loss: 0.0059 - val_categorical_accuracy: 1.0000 - lr: 0.0010\n",
"Epoch 4/500\n",
"11/11 [==============================] - 144s 13s/step - loss: 0.0391 - categorical_accuracy: 0.9852 - val_loss: 6.4974e-04 - val_categorical_accuracy: 1.0000 - lr: 0.0010\n",
"Epoch 5/500\n",
"11/11 [==============================] - 145s 13s/step - loss: 0.0871 - categorical_accuracy: 0.9763 - val_loss: 0.0122 - val_categorical_accuracy: 1.0000 - lr: 0.0010\n",
"Epoch 6/500\n",
"11/11 [==============================] - 145s 13s/step - loss: 0.0815 - categorical_accuracy: 0.9644 - val_loss: 0.0237 - val_categorical_accuracy: 1.0000 - lr: 0.0010\n",
"Epoch 7/500\n",
"11/11 [==============================] - 150s 14s/step - loss: 0.0190 - categorical_accuracy: 0.9911 - val_loss: 0.0328 - val_categorical_accuracy: 0.9853 - lr: 0.0010\n",
"Epoch 8/500\n",
"11/11 [==============================] - 142s 13s/step - loss: 0.0249 - categorical_accuracy: 0.9941 - val_loss: 6.4866e-04 - val_categorical_accuracy: 1.0000 - lr: 0.0010\n",
"Epoch 9/500\n",
"11/11 [==============================] - 144s 13s/step - loss: 0.0101 - categorical_accuracy: 0.9941 - val_loss: 0.0058 - val_categorical_accuracy: 1.0000 - lr: 0.0010\n",
"Epoch 10/500\n",
"11/11 [==============================] - 143s 13s/step - loss: 0.0173 - categorical_accuracy: 0.9941 - val_loss: 0.0012 - val_categorical_accuracy: 1.0000 - lr: 2.0000e-04\n",
"Epoch 11/500\n",
"11/11 [==============================] - 154s 14s/step - loss: 0.0176 - categorical_accuracy: 0.9941 - val_loss: 0.0010 - val_categorical_accuracy: 1.0000 - lr: 2.0000e-04\n",
"Epoch 12/500\n",
"11/11 [==============================] - 144s 13s/step - loss: 0.0040 - categorical_accuracy: 0.9970 - val_loss: 5.7718e-04 - val_categorical_accuracy: 1.0000 - lr: 2.0000e-04\n",
"Epoch 13/500\n",
"11/11 [==============================] - 144s 13s/step - loss: 0.0015 - categorical_accuracy: 1.0000 - val_loss: 5.7380e-04 - val_categorical_accuracy: 1.0000 - lr: 2.0000e-04\n",
"Epoch 14/500\n",
"11/11 [==============================] - 145s 13s/step - loss: 0.0014 - categorical_accuracy: 1.0000 - val_loss: 5.2094e-04 - val_categorical_accuracy: 1.0000 - lr: 2.0000e-04\n",
"Epoch 15/500\n",
"11/11 [==============================] - 149s 14s/step - loss: 0.0015 - categorical_accuracy: 1.0000 - val_loss: 4.4772e-04 - val_categorical_accuracy: 1.0000 - lr: 2.0000e-04\n",
"Epoch 16/500\n",
"11/11 [==============================] - 144s 13s/step - loss: 0.0012 - categorical_accuracy: 1.0000 - val_loss: 3.9085e-04 - val_categorical_accuracy: 1.0000 - lr: 2.0000e-04\n",
"Epoch 17/500\n",
"11/11 [==============================] - 147s 13s/step - loss: 0.0010 - categorical_accuracy: 1.0000 - val_loss: 3.4933e-04 - val_categorical_accuracy: 1.0000 - lr: 2.0000e-04\n",
"Epoch 18/500\n",
"11/11 [==============================] - 144s 13s/step - loss: 8.0251e-04 - categorical_accuracy: 1.0000 - val_loss: 3.1589e-04 - val_categorical_accuracy: 1.0000 - lr: 2.0000e-04\n",
"Epoch 19/500\n",
"11/11 [==============================] - 149s 14s/step - loss: 6.4664e-04 - categorical_accuracy: 1.0000 - val_loss: 2.9034e-04 - val_categorical_accuracy: 1.0000 - lr: 2.0000e-04\n",
"Epoch 20/500\n",
"11/11 [==============================] - 143s 13s/step - loss: 7.9226e-04 - categorical_accuracy: 1.0000 - val_loss: 2.6785e-04 - val_categorical_accuracy: 1.0000 - lr: 2.0000e-04\n",
"Epoch 21/500\n",
"11/11 [==============================] - 144s 13s/step - loss: 6.2462e-04 - categorical_accuracy: 1.0000 - val_loss: 2.4908e-04 - val_categorical_accuracy: 1.0000 - lr: 2.0000e-04\n",
"Epoch 22/500\n",
"11/11 [==============================] - 142s 13s/step - loss: 6.9292e-04 - categorical_accuracy: 1.0000 - val_loss: 2.3473e-04 - val_categorical_accuracy: 1.0000 - lr: 2.0000e-04\n",
"Epoch 23/500\n",
"11/11 [==============================] - 157s 14s/step - loss: 5.5603e-04 - categorical_accuracy: 1.0000 - val_loss: 2.2057e-04 - val_categorical_accuracy: 1.0000 - lr: 2.0000e-04\n",
"Epoch 24/500\n",
"11/11 [==============================] - 143s 13s/step - loss: 4.8737e-04 - categorical_accuracy: 1.0000 - val_loss: 2.0835e-04 - val_categorical_accuracy: 1.0000 - lr: 2.0000e-04\n",
"Epoch 25/500\n",
"11/11 [==============================] - 150s 14s/step - loss: 5.3003e-04 - categorical_accuracy: 1.0000 - val_loss: 2.0614e-04 - val_categorical_accuracy: 1.0000 - lr: 4.0000e-05\n",
"Epoch 26/500\n",
"11/11 [==============================] - 143s 13s/step - loss: 5.3267e-04 - categorical_accuracy: 1.0000 - val_loss: 2.0380e-04 - val_categorical_accuracy: 1.0000 - lr: 4.0000e-05\n",
"Epoch 27/500\n",
"11/11 [==============================] - 147s 13s/step - loss: 5.8821e-04 - categorical_accuracy: 1.0000 - val_loss: 2.0116e-04 - val_categorical_accuracy: 1.0000 - lr: 4.0000e-05\n",
"Epoch 28/500\n",
"11/11 [==============================] - 145s 13s/step - loss: 5.7868e-04 - categorical_accuracy: 1.0000 - val_loss: 1.9871e-04 - val_categorical_accuracy: 1.0000 - lr: 4.0000e-05\n",
"Epoch 29/500\n",
"11/11 [==============================] - 144s 13s/step - loss: 4.5697e-04 - categorical_accuracy: 1.0000 - val_loss: 1.9647e-04 - val_categorical_accuracy: 1.0000 - lr: 4.0000e-05\n",
"Epoch 30/500\n",
"11/11 [==============================] - 144s 13s/step - loss: 5.0632e-04 - categorical_accuracy: 1.0000 - val_loss: 1.9593e-04 - val_categorical_accuracy: 1.0000 - lr: 1.0000e-05\n",
"Epoch 31/500\n",
"11/11 [==============================] - 149s 14s/step - loss: 6.3565e-04 - categorical_accuracy: 1.0000 - val_loss: 1.9525e-04 - val_categorical_accuracy: 1.0000 - lr: 1.0000e-05\n",
"Epoch 32/500\n",
"11/11 [==============================] - 151s 13s/step - loss: 5.2290e-04 - categorical_accuracy: 1.0000 - val_loss: 1.9461e-04 - val_categorical_accuracy: 1.0000 - lr: 1.0000e-05\n",
"Epoch 33/500\n",
"11/11 [==============================] - 147s 13s/step - loss: 5.2975e-04 - categorical_accuracy: 1.0000 - val_loss: 1.9395e-04 - val_categorical_accuracy: 1.0000 - lr: 1.0000e-05\n",
"Epoch 34/500\n",
"11/11 [==============================] - 145s 13s/step - loss: 5.6739e-04 - categorical_accuracy: 1.0000 - val_loss: 1.9334e-04 - val_categorical_accuracy: 1.0000 - lr: 1.0000e-05\n",
"Epoch 35/500\n",
"11/11 [==============================] - 147s 13s/step - loss: 5.2916e-04 - categorical_accuracy: 1.0000 - val_loss: 1.9273e-04 - val_categorical_accuracy: 1.0000 - lr: 1.0000e-05\n",
"Epoch 36/500\n",
"11/11 [==============================] - 156s 14s/step - loss: 6.5789e-04 - categorical_accuracy: 1.0000 - val_loss: 1.9208e-04 - val_categorical_accuracy: 1.0000 - lr: 1.0000e-05\n",
"Epoch 37/500\n",
"11/11 [==============================] - 146s 13s/step - loss: 5.9525e-04 - categorical_accuracy: 1.0000 - val_loss: 1.9143e-04 - val_categorical_accuracy: 1.0000 - lr: 1.0000e-05\n",
"Epoch 38/500\n",
"11/11 [==============================] - 146s 13s/step - loss: 5.2344e-04 - categorical_accuracy: 1.0000 - val_loss: 1.9073e-04 - val_categorical_accuracy: 1.0000 - lr: 1.0000e-05\n",
"Epoch 39/500\n",
"11/11 [==============================] - 145s 13s/step - loss: 5.9891e-04 - categorical_accuracy: 1.0000 - val_loss: 1.9004e-04 - val_categorical_accuracy: 1.0000 - lr: 1.0000e-05\n",
"Epoch 40/500\n",
"11/11 [==============================] - 152s 14s/step - loss: 4.6774e-04 - categorical_accuracy: 1.0000 - val_loss: 1.8929e-04 - val_categorical_accuracy: 1.0000 - lr: 1.0000e-05\n",
"Epoch 41/500\n",
"11/11 [==============================] - 147s 13s/step - loss: 5.1332e-04 - categorical_accuracy: 1.0000 - val_loss: 1.8856e-04 - val_categorical_accuracy: 1.0000 - lr: 1.0000e-05\n",
"Epoch 42/500\n",
"11/11 [==============================] - 148s 13s/step - loss: 5.2163e-04 - categorical_accuracy: 1.0000 - val_loss: 1.8792e-04 - val_categorical_accuracy: 1.0000 - lr: 1.0000e-05\n",
"Epoch 43/500\n",
"11/11 [==============================] - 146s 13s/step - loss: 5.1293e-04 - categorical_accuracy: 1.0000 - val_loss: 1.8728e-04 - val_categorical_accuracy: 1.0000 - lr: 1.0000e-05\n",
"Epoch 44/500\n",
"11/11 [==============================] - 152s 14s/step - loss: 4.9182e-04 - categorical_accuracy: 1.0000 - val_loss: 1.8657e-04 - val_categorical_accuracy: 1.0000 - lr: 1.0000e-05\n",
"Epoch 45/500\n",
"11/11 [==============================] - 146s 13s/step - loss: 4.7899e-04 - categorical_accuracy: 1.0000 - val_loss: 1.8585e-04 - val_categorical_accuracy: 1.0000 - lr: 1.0000e-05\n",
"Epoch 46/500\n",
"11/11 [==============================] - 144s 13s/step - loss: 5.3644e-04 - categorical_accuracy: 1.0000 - val_loss: 1.8514e-04 - val_categorical_accuracy: 1.0000 - lr: 1.0000e-05\n",
"Epoch 47/500\n",
"11/11 [==============================] - 146s 13s/step - loss: 6.9909e-04 - categorical_accuracy: 1.0000 - val_loss: 1.8448e-04 - val_categorical_accuracy: 1.0000 - lr: 1.0000e-05\n",
"Epoch 48/500\n",
"11/11 [==============================] - 155s 14s/step - loss: 4.5525e-04 - categorical_accuracy: 1.0000 - val_loss: 1.8381e-04 - val_categorical_accuracy: 1.0000 - lr: 1.0000e-05\n",
"Epoch 49/500\n",
"11/11 [==============================] - 146s 13s/step - loss: 5.4966e-04 - categorical_accuracy: 1.0000 - val_loss: 1.8309e-04 - val_categorical_accuracy: 1.0000 - lr: 1.0000e-05\n",
"Epoch 50/500\n",
"11/11 [==============================] - 152s 14s/step - loss: 4.9885e-04 - categorical_accuracy: 1.0000 - val_loss: 1.8239e-04 - val_categorical_accuracy: 1.0000 - lr: 1.0000e-05\n",
"Epoch 51/500\n",
"11/11 [==============================] - 146s 13s/step - loss: 4.9701e-04 - categorical_accuracy: 1.0000 - val_loss: 1.8164e-04 - val_categorical_accuracy: 1.0000 - lr: 1.0000e-05\n",
"Epoch 52/500\n",
"11/11 [==============================] - 153s 14s/step - loss: 5.7490e-04 - categorical_accuracy: 1.0000 - val_loss: 1.8081e-04 - val_categorical_accuracy: 1.0000 - lr: 1.0000e-05\n",
"Epoch 53/500\n",
"11/11 [==============================] - 147s 13s/step - loss: 6.1352e-04 - categorical_accuracy: 1.0000 - val_loss: 1.8002e-04 - val_categorical_accuracy: 1.0000 - lr: 1.0000e-05\n",
"Epoch 54/500\n",
"11/11 [==============================] - 148s 13s/step - loss: 5.3162e-04 - categorical_accuracy: 1.0000 - val_loss: 1.7926e-04 - val_categorical_accuracy: 1.0000 - lr: 1.0000e-05\n",
"Epoch 55/500\n",
"11/11 [==============================] - 145s 13s/step - loss: 4.3685e-04 - categorical_accuracy: 1.0000 - val_loss: 1.7852e-04 - val_categorical_accuracy: 1.0000 - lr: 1.0000e-05\n",
"Epoch 56/500\n",
"11/11 [==============================] - 153s 14s/step - loss: 4.6293e-04 - categorical_accuracy: 1.0000 - val_loss: 1.7782e-04 - val_categorical_accuracy: 1.0000 - lr: 1.0000e-05\n",
"Epoch 57/500\n",
"11/11 [==============================] - 145s 13s/step - loss: 4.9126e-04 - categorical_accuracy: 1.0000 - val_loss: 1.7709e-04 - val_categorical_accuracy: 1.0000 - lr: 1.0000e-05\n",
"Epoch 58/500\n",
"11/11 [==============================] - 146s 13s/step - loss: 5.6628e-04 - categorical_accuracy: 1.0000 - val_loss: 1.7628e-04 - val_categorical_accuracy: 1.0000 - lr: 1.0000e-05\n",
"Epoch 59/500\n",
"11/11 [==============================] - 147s 13s/step - loss: 4.5964e-04 - categorical_accuracy: 1.0000 - val_loss: 1.7552e-04 - val_categorical_accuracy: 1.0000 - lr: 1.0000e-05\n",
"Epoch 60/500\n",
"11/11 [==============================] - 157s 14s/step - loss: 6.6674e-04 - categorical_accuracy: 1.0000 - val_loss: 1.7479e-04 - val_categorical_accuracy: 1.0000 - lr: 1.0000e-05\n",
"Epoch 61/500\n",
"11/11 [==============================] - 145s 13s/step - loss: 5.1090e-04 - categorical_accuracy: 1.0000 - val_loss: 1.7403e-04 - val_categorical_accuracy: 1.0000 - lr: 1.0000e-05\n",
"Epoch 62/500\n",
"11/11 [==============================] - 145s 13s/step - loss: 4.9706e-04 - categorical_accuracy: 1.0000 - val_loss: 1.7322e-04 - val_categorical_accuracy: 1.0000 - lr: 1.0000e-05\n",
"Epoch 63/500\n",
"11/11 [==============================] - 147s 13s/step - loss: 5.4192e-04 - categorical_accuracy: 1.0000 - val_loss: 1.7244e-04 - val_categorical_accuracy: 1.0000 - lr: 1.0000e-05\n",
"Epoch 64/500\n",
"11/11 [==============================] - 152s 14s/step - loss: 5.9925e-04 - categorical_accuracy: 1.0000 - val_loss: 1.7167e-04 - val_categorical_accuracy: 1.0000 - lr: 1.0000e-05\n",
"Epoch 65/500\n",
"11/11 [==============================] - 145s 13s/step - loss: 4.5704e-04 - categorical_accuracy: 1.0000 - val_loss: 1.7089e-04 - val_categorical_accuracy: 1.0000 - lr: 1.0000e-05\n",
"Epoch 66/500\n",
"11/11 [==============================] - 148s 13s/step - loss: 4.4658e-04 - categorical_accuracy: 1.0000 - val_loss: 1.7017e-04 - val_categorical_accuracy: 1.0000 - lr: 1.0000e-05\n",
"Epoch 67/500\n",
"11/11 [==============================] - 145s 13s/step - loss: 4.1316e-04 - categorical_accuracy: 1.0000 - val_loss: 1.6946e-04 - val_categorical_accuracy: 1.0000 - lr: 1.0000e-05\n",
"Epoch 68/500\n",
"11/11 [==============================] - 154s 14s/step - loss: 4.0055e-04 - categorical_accuracy: 1.0000 - val_loss: 1.6878e-04 - val_categorical_accuracy: 1.0000 - lr: 1.0000e-05\n",
"Epoch 69/500\n",
"11/11 [==============================] - 160s 15s/step - loss: 5.0363e-04 - categorical_accuracy: 1.0000 - val_loss: 1.6799e-04 - val_categorical_accuracy: 1.0000 - lr: 1.0000e-05\n",
"Epoch 70/500\n",
"11/11 [==============================] - 164s 15s/step - loss: 5.1680e-04 - categorical_accuracy: 1.0000 - val_loss: 1.6720e-04 - val_categorical_accuracy: 1.0000 - lr: 1.0000e-05\n",
"Epoch 71/500\n",
"11/11 [==============================] - 147s 13s/step - loss: 4.6495e-04 - categorical_accuracy: 1.0000 - val_loss: 1.6639e-04 - val_categorical_accuracy: 1.0000 - lr: 1.0000e-05\n",
"Epoch 72/500\n",
"11/11 [==============================] - 158s 14s/step - loss: 3.7927e-04 - categorical_accuracy: 1.0000 - val_loss: 1.6566e-04 - val_categorical_accuracy: 1.0000 - lr: 1.0000e-05\n",
"Epoch 73/500\n",
"11/11 [==============================] - 152s 14s/step - loss: 4.0601e-04 - categorical_accuracy: 1.0000 - val_loss: 1.6494e-04 - val_categorical_accuracy: 1.0000 - lr: 1.0000e-05\n",
"Epoch 74/500\n",
"11/11 [==============================] - 155s 14s/step - loss: 4.0291e-04 - categorical_accuracy: 1.0000 - val_loss: 1.6418e-04 - val_categorical_accuracy: 1.0000 - lr: 1.0000e-05\n",
"Epoch 75/500\n",
"11/11 [==============================] - 146s 13s/step - loss: 4.8252e-04 - categorical_accuracy: 1.0000 - val_loss: 1.6345e-04 - val_categorical_accuracy: 1.0000 - lr: 1.0000e-05\n",
"Epoch 76/500\n",
"11/11 [==============================] - 153s 14s/step - loss: 4.6234e-04 - categorical_accuracy: 1.0000 - val_loss: 1.6267e-04 - val_categorical_accuracy: 1.0000 - lr: 1.0000e-05\n",
"Epoch 77/500\n",
"11/11 [==============================] - 147s 13s/step - loss: 4.7696e-04 - categorical_accuracy: 1.0000 - val_loss: 1.6190e-04 - val_categorical_accuracy: 1.0000 - lr: 1.0000e-05\n",
"Epoch 78/500\n",
"11/11 [==============================] - 147s 13s/step - loss: 4.7757e-04 - categorical_accuracy: 1.0000 - val_loss: 1.6107e-04 - val_categorical_accuracy: 1.0000 - lr: 1.0000e-05\n",
"Epoch 79/500\n",
"11/11 [==============================] - 147s 13s/step - loss: 4.7498e-04 - categorical_accuracy: 1.0000 - val_loss: 1.6023e-04 - val_categorical_accuracy: 1.0000 - lr: 1.0000e-05\n",
"Epoch 80/500\n",
"11/11 [==============================] - 151s 14s/step - loss: 4.5403e-04 - categorical_accuracy: 1.0000 - val_loss: 1.5941e-04 - val_categorical_accuracy: 1.0000 - lr: 1.0000e-05\n",
"Epoch 81/500\n",
"11/11 [==============================] - 145s 13s/step - loss: 4.4490e-04 - categorical_accuracy: 1.0000 - val_loss: 1.5862e-04 - val_categorical_accuracy: 1.0000 - lr: 1.0000e-05\n",
"Epoch 82/500\n",
"11/11 [==============================] - 146s 13s/step - loss: 4.3853e-04 - categorical_accuracy: 1.0000 - val_loss: 1.5785e-04 - val_categorical_accuracy: 1.0000 - lr: 1.0000e-05\n",
"Epoch 83/500\n",
"11/11 [==============================] - 146s 13s/step - loss: 4.5649e-04 - categorical_accuracy: 1.0000 - val_loss: 1.5709e-04 - val_categorical_accuracy: 1.0000 - lr: 1.0000e-05\n",
"Epoch 84/500\n",
"11/11 [==============================] - 157s 14s/step - loss: 4.5282e-04 - categorical_accuracy: 1.0000 - val_loss: 1.5631e-04 - val_categorical_accuracy: 1.0000 - lr: 1.0000e-05\n",
"Epoch 85/500\n",
"11/11 [==============================] - 145s 13s/step - loss: 4.4384e-04 - categorical_accuracy: 1.0000 - val_loss: 1.5551e-04 - val_categorical_accuracy: 1.0000 - lr: 1.0000e-05\n",
"Epoch 86/500\n",
"11/11 [==============================] - 146s 13s/step - loss: 4.5909e-04 - categorical_accuracy: 1.0000 - val_loss: 1.5476e-04 - val_categorical_accuracy: 1.0000 - lr: 1.0000e-05\n",
"Epoch 87/500\n",
"11/11 [==============================] - 146s 13s/step - loss: 4.7136e-04 - categorical_accuracy: 1.0000 - val_loss: 1.5396e-04 - val_categorical_accuracy: 1.0000 - lr: 1.0000e-05\n",
"Epoch 88/500\n",
"11/11 [==============================] - 153s 14s/step - loss: 5.0904e-04 - categorical_accuracy: 1.0000 - val_loss: 1.5321e-04 - val_categorical_accuracy: 1.0000 - lr: 1.0000e-05\n",
"Epoch 89/500\n",
"11/11 [==============================] - 146s 13s/step - loss: 4.3501e-04 - categorical_accuracy: 1.0000 - val_loss: 1.5245e-04 - val_categorical_accuracy: 1.0000 - lr: 1.0000e-05\n",
"Epoch 90/500\n",
"11/11 [==============================] - 150s 14s/step - loss: 3.7642e-04 - categorical_accuracy: 1.0000 - val_loss: 1.5165e-04 - val_categorical_accuracy: 1.0000 - lr: 1.0000e-05\n",
"Epoch 91/500\n",
"11/11 [==============================] - 146s 13s/step - loss: 4.0477e-04 - categorical_accuracy: 1.0000 - val_loss: 1.5089e-04 - val_categorical_accuracy: 1.0000 - lr: 1.0000e-05\n",
"Epoch 92/500\n",
"11/11 [==============================] - 153s 14s/step - loss: 4.2010e-04 - categorical_accuracy: 1.0000 - val_loss: 1.5009e-04 - val_categorical_accuracy: 1.0000 - lr: 1.0000e-05\n",
"Epoch 93/500\n",
"11/11 [==============================] - 147s 13s/step - loss: 4.4686e-04 - categorical_accuracy: 1.0000 - val_loss: 1.4930e-04 - val_categorical_accuracy: 1.0000 - lr: 1.0000e-05\n",
"Epoch 94/500\n",
"11/11 [==============================] - 147s 13s/step - loss: 4.8152e-04 - categorical_accuracy: 1.0000 - val_loss: 1.4845e-04 - val_categorical_accuracy: 1.0000 - lr: 1.0000e-05\n",
"Epoch 95/500\n",
"11/11 [==============================] - 148s 13s/step - loss: 4.2370e-04 - categorical_accuracy: 1.0000 - val_loss: 1.4769e-04 - val_categorical_accuracy: 1.0000 - lr: 1.0000e-05\n",
"Epoch 96/500\n",
"11/11 [==============================] - 157s 14s/step - loss: 3.8550e-04 - categorical_accuracy: 1.0000 - val_loss: 1.4696e-04 - val_categorical_accuracy: 1.0000 - lr: 1.0000e-05\n",
"Epoch 97/500\n",
"11/11 [==============================] - 145s 13s/step - loss: 3.4308e-04 - categorical_accuracy: 1.0000 - val_loss: 1.4615e-04 - val_categorical_accuracy: 1.0000 - lr: 1.0000e-05\n",
"Epoch 98/500\n",
"11/11 [==============================] - 146s 13s/step - loss: 4.1931e-04 - categorical_accuracy: 1.0000 - val_loss: 1.4544e-04 - val_categorical_accuracy: 1.0000 - lr: 1.0000e-05\n",
"Epoch 99/500\n",
"11/11 [==============================] - 152s 14s/step - loss: 3.9290e-04 - categorical_accuracy: 1.0000 - val_loss: 1.4467e-04 - val_categorical_accuracy: 1.0000 - lr: 1.0000e-05\n",
"Epoch 100/500\n",
"11/11 [==============================] - 152s 14s/step - loss: 3.4640e-04 - categorical_accuracy: 1.0000 - val_loss: 1.4395e-04 - val_categorical_accuracy: 1.0000 - lr: 1.0000e-05\n",
"Epoch 101/500\n",
"11/11 [==============================] - 146s 13s/step - loss: 3.9130e-04 - categorical_accuracy: 1.0000 - val_loss: 1.4314e-04 - val_categorical_accuracy: 1.0000 - lr: 1.0000e-05\n",
"Epoch 102/500\n",
"11/11 [==============================] - 152s 14s/step - loss: 3.4727e-04 - categorical_accuracy: 1.0000 - val_loss: 1.4239e-04 - val_categorical_accuracy: 1.0000 - lr: 1.0000e-05\n",
"Epoch 103/500\n",
"11/11 [==============================] - 146s 13s/step - loss: 4.5238e-04 - categorical_accuracy: 1.0000 - val_loss: 1.4166e-04 - val_categorical_accuracy: 1.0000 - lr: 1.0000e-05\n",
"Epoch 104/500\n",
" 1/11 [=>............................] - ETA: 2:14 - loss: 3.6348e-04 - categorical_accuracy: 1.0000"
]
},
{
"ename": "KeyboardInterrupt",
"evalue": "",
"output_type": "error",
"traceback": [
"\u001b[1;31m---------------------------------------------------------------------------\u001b[0m",
"\u001b[1;31mKeyboardInterrupt\u001b[0m Traceback (most recent call last)",
"\u001b[1;32mc:\\Users\\cpras\\Documents\\GitHub\\AI_Personal_Trainer\\ExerciseDecoder.ipynb Cell 35\u001b[0m in \u001b[0;36m<cell line: 2>\u001b[1;34m()\u001b[0m\n\u001b[0;32m <a href='vscode-notebook-cell:/c%3A/Users/cpras/Documents/GitHub/AI_Personal_Trainer/ExerciseDecoder.ipynb#ch0000034?line=0'>1</a>\u001b[0m AttnLSTM\u001b[39m.\u001b[39mcompile(optimizer\u001b[39m=\u001b[39m\u001b[39m'\u001b[39m\u001b[39mAdam\u001b[39m\u001b[39m'\u001b[39m, loss\u001b[39m=\u001b[39m\u001b[39m'\u001b[39m\u001b[39mcategorical_crossentropy\u001b[39m\u001b[39m'\u001b[39m, metrics\u001b[39m=\u001b[39m[\u001b[39m'\u001b[39m\u001b[39mcategorical_accuracy\u001b[39m\u001b[39m'\u001b[39m])\n\u001b[1;32m----> <a href='vscode-notebook-cell:/c%3A/Users/cpras/Documents/GitHub/AI_Personal_Trainer/ExerciseDecoder.ipynb#ch0000034?line=1'>2</a>\u001b[0m AttnLSTM\u001b[39m.\u001b[39;49mfit(X_train, y_train, batch_size\u001b[39m=\u001b[39;49mbatch_size, epochs\u001b[39m=\u001b[39;49mmax_epochs, validation_data\u001b[39m=\u001b[39;49m(X_val, y_val), callbacks\u001b[39m=\u001b[39;49mcallbacks)\n",
"File \u001b[1;32m~\\AppData\\Roaming\\Python\\Python38\\site-packages\\keras\\utils\\traceback_utils.py:64\u001b[0m, in \u001b[0;36mfilter_traceback.<locals>.error_handler\u001b[1;34m(*args, **kwargs)\u001b[0m\n\u001b[0;32m 62\u001b[0m filtered_tb \u001b[39m=\u001b[39m \u001b[39mNone\u001b[39;00m\n\u001b[0;32m 63\u001b[0m \u001b[39mtry\u001b[39;00m:\n\u001b[1;32m---> 64\u001b[0m \u001b[39mreturn\u001b[39;00m fn(\u001b[39m*\u001b[39;49margs, \u001b[39m*\u001b[39;49m\u001b[39m*\u001b[39;49mkwargs)\n\u001b[0;32m 65\u001b[0m \u001b[39mexcept\u001b[39;00m \u001b[39mException\u001b[39;00m \u001b[39mas\u001b[39;00m e: \u001b[39m# pylint: disable=broad-except\u001b[39;00m\n\u001b[0;32m 66\u001b[0m filtered_tb \u001b[39m=\u001b[39m _process_traceback_frames(e\u001b[39m.\u001b[39m__traceback__)\n",
"File \u001b[1;32m~\\AppData\\Roaming\\Python\\Python38\\site-packages\\keras\\engine\\training.py:1414\u001b[0m, in \u001b[0;36mModel.fit\u001b[1;34m(self, x, y, batch_size, epochs, verbose, callbacks, validation_split, validation_data, shuffle, class_weight, sample_weight, initial_epoch, steps_per_epoch, validation_steps, validation_batch_size, validation_freq, max_queue_size, workers, use_multiprocessing)\u001b[0m\n\u001b[0;32m 1412\u001b[0m logs \u001b[39m=\u001b[39m tmp_logs \u001b[39m# No error, now safe to assign to logs.\u001b[39;00m\n\u001b[0;32m 1413\u001b[0m end_step \u001b[39m=\u001b[39m step \u001b[39m+\u001b[39m data_handler\u001b[39m.\u001b[39mstep_increment\n\u001b[1;32m-> 1414\u001b[0m callbacks\u001b[39m.\u001b[39;49mon_train_batch_end(end_step, logs)\n\u001b[0;32m 1415\u001b[0m \u001b[39mif\u001b[39;00m \u001b[39mself\u001b[39m\u001b[39m.\u001b[39mstop_training:\n\u001b[0;32m 1416\u001b[0m \u001b[39mbreak\u001b[39;00m\n",
"File \u001b[1;32m~\\AppData\\Roaming\\Python\\Python38\\site-packages\\keras\\callbacks.py:438\u001b[0m, in \u001b[0;36mCallbackList.on_train_batch_end\u001b[1;34m(self, batch, logs)\u001b[0m\n\u001b[0;32m 431\u001b[0m \u001b[39m\"\"\"Calls the `on_train_batch_end` methods of its callbacks.\u001b[39;00m\n\u001b[0;32m 432\u001b[0m \n\u001b[0;32m 433\u001b[0m \u001b[39mArgs:\u001b[39;00m\n\u001b[0;32m 434\u001b[0m \u001b[39m batch: Integer, index of batch within the current epoch.\u001b[39;00m\n\u001b[0;32m 435\u001b[0m \u001b[39m logs: Dict. Aggregated metric results up until this batch.\u001b[39;00m\n\u001b[0;32m 436\u001b[0m \u001b[39m\"\"\"\u001b[39;00m\n\u001b[0;32m 437\u001b[0m \u001b[39mif\u001b[39;00m \u001b[39mself\u001b[39m\u001b[39m.\u001b[39m_should_call_train_batch_hooks:\n\u001b[1;32m--> 438\u001b[0m \u001b[39mself\u001b[39;49m\u001b[39m.\u001b[39;49m_call_batch_hook(ModeKeys\u001b[39m.\u001b[39;49mTRAIN, \u001b[39m'\u001b[39;49m\u001b[39mend\u001b[39;49m\u001b[39m'\u001b[39;49m, batch, logs\u001b[39m=\u001b[39;49mlogs)\n",
"File \u001b[1;32m~\\AppData\\Roaming\\Python\\Python38\\site-packages\\keras\\callbacks.py:297\u001b[0m, in \u001b[0;36mCallbackList._call_batch_hook\u001b[1;34m(self, mode, hook, batch, logs)\u001b[0m\n\u001b[0;32m 295\u001b[0m \u001b[39mself\u001b[39m\u001b[39m.\u001b[39m_call_batch_begin_hook(mode, batch, logs)\n\u001b[0;32m 296\u001b[0m \u001b[39melif\u001b[39;00m hook \u001b[39m==\u001b[39m \u001b[39m'\u001b[39m\u001b[39mend\u001b[39m\u001b[39m'\u001b[39m:\n\u001b[1;32m--> 297\u001b[0m \u001b[39mself\u001b[39;49m\u001b[39m.\u001b[39;49m_call_batch_end_hook(mode, batch, logs)\n\u001b[0;32m 298\u001b[0m \u001b[39melse\u001b[39;00m:\n\u001b[0;32m 299\u001b[0m \u001b[39mraise\u001b[39;00m \u001b[39mValueError\u001b[39;00m(\n\u001b[0;32m 300\u001b[0m \u001b[39mf\u001b[39m\u001b[39m'\u001b[39m\u001b[39mUnrecognized hook: \u001b[39m\u001b[39m{\u001b[39;00mhook\u001b[39m}\u001b[39;00m\u001b[39m. Expected values are [\u001b[39m\u001b[39m\"\u001b[39m\u001b[39mbegin\u001b[39m\u001b[39m\"\u001b[39m\u001b[39m, \u001b[39m\u001b[39m\"\u001b[39m\u001b[39mend\u001b[39m\u001b[39m\"\u001b[39m\u001b[39m]\u001b[39m\u001b[39m'\u001b[39m)\n",
"File \u001b[1;32m~\\AppData\\Roaming\\Python\\Python38\\site-packages\\keras\\callbacks.py:318\u001b[0m, in \u001b[0;36mCallbackList._call_batch_end_hook\u001b[1;34m(self, mode, batch, logs)\u001b[0m\n\u001b[0;32m 315\u001b[0m batch_time \u001b[39m=\u001b[39m time\u001b[39m.\u001b[39mtime() \u001b[39m-\u001b[39m \u001b[39mself\u001b[39m\u001b[39m.\u001b[39m_batch_start_time\n\u001b[0;32m 316\u001b[0m \u001b[39mself\u001b[39m\u001b[39m.\u001b[39m_batch_times\u001b[39m.\u001b[39mappend(batch_time)\n\u001b[1;32m--> 318\u001b[0m \u001b[39mself\u001b[39;49m\u001b[39m.\u001b[39;49m_call_batch_hook_helper(hook_name, batch, logs)\n\u001b[0;32m 320\u001b[0m \u001b[39mif\u001b[39;00m \u001b[39mlen\u001b[39m(\u001b[39mself\u001b[39m\u001b[39m.\u001b[39m_batch_times) \u001b[39m>\u001b[39m\u001b[39m=\u001b[39m \u001b[39mself\u001b[39m\u001b[39m.\u001b[39m_num_batches_for_timing_check:\n\u001b[0;32m 321\u001b[0m end_hook_name \u001b[39m=\u001b[39m hook_name\n",
"File \u001b[1;32m~\\AppData\\Roaming\\Python\\Python38\\site-packages\\keras\\callbacks.py:356\u001b[0m, in \u001b[0;36mCallbackList._call_batch_hook_helper\u001b[1;34m(self, hook_name, batch, logs)\u001b[0m\n\u001b[0;32m 354\u001b[0m \u001b[39mfor\u001b[39;00m callback \u001b[39min\u001b[39;00m \u001b[39mself\u001b[39m\u001b[39m.\u001b[39mcallbacks:\n\u001b[0;32m 355\u001b[0m hook \u001b[39m=\u001b[39m \u001b[39mgetattr\u001b[39m(callback, hook_name)\n\u001b[1;32m--> 356\u001b[0m hook(batch, logs)\n\u001b[0;32m 358\u001b[0m \u001b[39mif\u001b[39;00m \u001b[39mself\u001b[39m\u001b[39m.\u001b[39m_check_timing:\n\u001b[0;32m 359\u001b[0m \u001b[39mif\u001b[39;00m hook_name \u001b[39mnot\u001b[39;00m \u001b[39min\u001b[39;00m \u001b[39mself\u001b[39m\u001b[39m.\u001b[39m_hook_times:\n",
"File \u001b[1;32m~\\AppData\\Roaming\\Python\\Python38\\site-packages\\keras\\callbacks.py:1380\u001b[0m, in \u001b[0;36mModelCheckpoint.on_train_batch_end\u001b[1;34m(self, batch, logs)\u001b[0m\n\u001b[0;32m 1378\u001b[0m \u001b[39mdef\u001b[39;00m \u001b[39mon_train_batch_end\u001b[39m(\u001b[39mself\u001b[39m, batch, logs\u001b[39m=\u001b[39m\u001b[39mNone\u001b[39;00m):\n\u001b[0;32m 1379\u001b[0m \u001b[39mif\u001b[39;00m \u001b[39mself\u001b[39m\u001b[39m.\u001b[39m_should_save_on_batch(batch):\n\u001b[1;32m-> 1380\u001b[0m \u001b[39mself\u001b[39;49m\u001b[39m.\u001b[39;49m_save_model(epoch\u001b[39m=\u001b[39;49m\u001b[39mself\u001b[39;49m\u001b[39m.\u001b[39;49m_current_epoch, batch\u001b[39m=\u001b[39;49mbatch, logs\u001b[39m=\u001b[39;49mlogs)\n",
"File \u001b[1;32m~\\AppData\\Roaming\\Python\\Python38\\site-packages\\keras\\callbacks.py:1458\u001b[0m, in \u001b[0;36mModelCheckpoint._save_model\u001b[1;34m(self, epoch, batch, logs)\u001b[0m\n\u001b[0;32m 1455\u001b[0m \u001b[39mself\u001b[39m\u001b[39m.\u001b[39mmodel\u001b[39m.\u001b[39msave_weights(\n\u001b[0;32m 1456\u001b[0m filepath, overwrite\u001b[39m=\u001b[39m\u001b[39mTrue\u001b[39;00m, options\u001b[39m=\u001b[39m\u001b[39mself\u001b[39m\u001b[39m.\u001b[39m_options)\n\u001b[0;32m 1457\u001b[0m \u001b[39melse\u001b[39;00m:\n\u001b[1;32m-> 1458\u001b[0m \u001b[39mself\u001b[39;49m\u001b[39m.\u001b[39;49mmodel\u001b[39m.\u001b[39;49msave(filepath, overwrite\u001b[39m=\u001b[39;49m\u001b[39mTrue\u001b[39;49;00m, options\u001b[39m=\u001b[39;49m\u001b[39mself\u001b[39;49m\u001b[39m.\u001b[39;49m_options)\n\u001b[0;32m 1460\u001b[0m \u001b[39mself\u001b[39m\u001b[39m.\u001b[39m_maybe_remove_file()\n\u001b[0;32m 1461\u001b[0m \u001b[39mexcept\u001b[39;00m \u001b[39mIsADirectoryError\u001b[39;00m \u001b[39mas\u001b[39;00m e: \u001b[39m# h5py 3.x\u001b[39;00m\n",
"File \u001b[1;32m~\\AppData\\Roaming\\Python\\Python38\\site-packages\\keras\\utils\\traceback_utils.py:64\u001b[0m, in \u001b[0;36mfilter_traceback.<locals>.error_handler\u001b[1;34m(*args, **kwargs)\u001b[0m\n\u001b[0;32m 62\u001b[0m filtered_tb \u001b[39m=\u001b[39m \u001b[39mNone\u001b[39;00m\n\u001b[0;32m 63\u001b[0m \u001b[39mtry\u001b[39;00m:\n\u001b[1;32m---> 64\u001b[0m \u001b[39mreturn\u001b[39;00m fn(\u001b[39m*\u001b[39;49margs, \u001b[39m*\u001b[39;49m\u001b[39m*\u001b[39;49mkwargs)\n\u001b[0;32m 65\u001b[0m \u001b[39mexcept\u001b[39;00m \u001b[39mException\u001b[39;00m \u001b[39mas\u001b[39;00m e: \u001b[39m# pylint: disable=broad-except\u001b[39;00m\n\u001b[0;32m 66\u001b[0m filtered_tb \u001b[39m=\u001b[39m _process_traceback_frames(e\u001b[39m.\u001b[39m__traceback__)\n",
"File \u001b[1;32m~\\AppData\\Roaming\\Python\\Python38\\site-packages\\keras\\engine\\training.py:2435\u001b[0m, in \u001b[0;36mModel.save\u001b[1;34m(self, filepath, overwrite, include_optimizer, save_format, signatures, options, save_traces)\u001b[0m\n\u001b[0;32m 2393\u001b[0m \u001b[39m\"\"\"Saves the model to Tensorflow SavedModel or a single HDF5 file.\u001b[39;00m\n\u001b[0;32m 2394\u001b[0m \n\u001b[0;32m 2395\u001b[0m \u001b[39mPlease see `tf.keras.models.save_model` or the\u001b[39;00m\n\u001b[1;32m (...)\u001b[0m\n\u001b[0;32m 2432\u001b[0m \u001b[39m```\u001b[39;00m\n\u001b[0;32m 2433\u001b[0m \u001b[39m\"\"\"\u001b[39;00m\n\u001b[0;32m 2434\u001b[0m \u001b[39m# pylint: enable=line-too-long\u001b[39;00m\n\u001b[1;32m-> 2435\u001b[0m save\u001b[39m.\u001b[39;49msave_model(\u001b[39mself\u001b[39;49m, filepath, overwrite, include_optimizer, save_format,\n\u001b[0;32m 2436\u001b[0m signatures, options, save_traces)\n",
"File \u001b[1;32m~\\AppData\\Roaming\\Python\\Python38\\site-packages\\keras\\utils\\traceback_utils.py:64\u001b[0m, in \u001b[0;36mfilter_traceback.<locals>.error_handler\u001b[1;34m(*args, **kwargs)\u001b[0m\n\u001b[0;32m 62\u001b[0m filtered_tb \u001b[39m=\u001b[39m \u001b[39mNone\u001b[39;00m\n\u001b[0;32m 63\u001b[0m \u001b[39mtry\u001b[39;00m:\n\u001b[1;32m---> 64\u001b[0m \u001b[39mreturn\u001b[39;00m fn(\u001b[39m*\u001b[39;49margs, \u001b[39m*\u001b[39;49m\u001b[39m*\u001b[39;49mkwargs)\n\u001b[0;32m 65\u001b[0m \u001b[39mexcept\u001b[39;00m \u001b[39mException\u001b[39;00m \u001b[39mas\u001b[39;00m e: \u001b[39m# pylint: disable=broad-except\u001b[39;00m\n\u001b[0;32m 66\u001b[0m filtered_tb \u001b[39m=\u001b[39m _process_traceback_frames(e\u001b[39m.\u001b[39m__traceback__)\n",
"File \u001b[1;32m~\\AppData\\Roaming\\Python\\Python38\\site-packages\\keras\\saving\\save.py:153\u001b[0m, in \u001b[0;36msave_model\u001b[1;34m(model, filepath, overwrite, include_optimizer, save_format, signatures, options, save_traces)\u001b[0m\n\u001b[0;32m 151\u001b[0m \u001b[39melse\u001b[39;00m:\n\u001b[0;32m 152\u001b[0m \u001b[39mwith\u001b[39;00m generic_utils\u001b[39m.\u001b[39mSharedObjectSavingScope():\n\u001b[1;32m--> 153\u001b[0m saved_model_save\u001b[39m.\u001b[39;49msave(model, filepath, overwrite, include_optimizer,\n\u001b[0;32m 154\u001b[0m signatures, options, save_traces)\n",
"File \u001b[1;32m~\\AppData\\Roaming\\Python\\Python38\\site-packages\\keras\\saving\\saved_model\\save.py:93\u001b[0m, in \u001b[0;36msave\u001b[1;34m(model, filepath, overwrite, include_optimizer, signatures, options, save_traces)\u001b[0m\n\u001b[0;32m 91\u001b[0m \u001b[39mwith\u001b[39;00m backend\u001b[39m.\u001b[39mdeprecated_internal_learning_phase_scope(\u001b[39m0\u001b[39m):\n\u001b[0;32m 92\u001b[0m \u001b[39mwith\u001b[39;00m utils\u001b[39m.\u001b[39mkeras_option_scope(save_traces):\n\u001b[1;32m---> 93\u001b[0m saved_nodes, node_paths \u001b[39m=\u001b[39m save_lib\u001b[39m.\u001b[39;49msave_and_return_nodes(\n\u001b[0;32m 94\u001b[0m model, filepath, signatures, options)\n\u001b[0;32m 96\u001b[0m \u001b[39m# Save all metadata to a separate file in the SavedModel directory.\u001b[39;00m\n\u001b[0;32m 97\u001b[0m metadata \u001b[39m=\u001b[39m generate_keras_metadata(saved_nodes, node_paths)\n",
"File \u001b[1;32m~\\AppData\\Roaming\\Python\\Python38\\site-packages\\tensorflow\\python\\saved_model\\save.py:1325\u001b[0m, in \u001b[0;36msave_and_return_nodes\u001b[1;34m(obj, export_dir, signatures, options, experimental_skip_checkpoint)\u001b[0m\n\u001b[0;32m 1321\u001b[0m saved_model \u001b[39m=\u001b[39m saved_model_pb2\u001b[39m.\u001b[39mSavedModel()\n\u001b[0;32m 1322\u001b[0m meta_graph_def \u001b[39m=\u001b[39m saved_model\u001b[39m.\u001b[39mmeta_graphs\u001b[39m.\u001b[39madd()\n\u001b[0;32m 1324\u001b[0m _, exported_graph, object_saver, asset_info, saved_nodes, node_paths \u001b[39m=\u001b[39m (\n\u001b[1;32m-> 1325\u001b[0m _build_meta_graph(obj, signatures, options, meta_graph_def))\n\u001b[0;32m 1326\u001b[0m saved_model\u001b[39m.\u001b[39msaved_model_schema_version \u001b[39m=\u001b[39m (\n\u001b[0;32m 1327\u001b[0m constants\u001b[39m.\u001b[39mSAVED_MODEL_SCHEMA_VERSION)\n\u001b[0;32m 1329\u001b[0m \u001b[39m# Write the checkpoint, copy assets into the assets directory, and write out\u001b[39;00m\n\u001b[0;32m 1330\u001b[0m \u001b[39m# the SavedModel proto itself.\u001b[39;00m\n",
"File \u001b[1;32m~\\AppData\\Roaming\\Python\\Python38\\site-packages\\tensorflow\\python\\saved_model\\save.py:1491\u001b[0m, in \u001b[0;36m_build_meta_graph\u001b[1;34m(obj, signatures, options, meta_graph_def)\u001b[0m\n\u001b[0;32m 1466\u001b[0m \u001b[39m\"\"\"Creates a MetaGraph under a save context.\u001b[39;00m\n\u001b[0;32m 1467\u001b[0m \n\u001b[0;32m 1468\u001b[0m \u001b[39mArgs:\u001b[39;00m\n\u001b[1;32m (...)\u001b[0m\n\u001b[0;32m 1487\u001b[0m \u001b[39m asset_info: `_AssetInfo` tuple containing external assets in the `obj`.\u001b[39;00m\n\u001b[0;32m 1488\u001b[0m \u001b[39m\"\"\"\u001b[39;00m\n\u001b[0;32m 1490\u001b[0m \u001b[39mwith\u001b[39;00m save_context\u001b[39m.\u001b[39msave_context(options):\n\u001b[1;32m-> 1491\u001b[0m \u001b[39mreturn\u001b[39;00m _build_meta_graph_impl(obj, signatures, options, meta_graph_def)\n",
"File \u001b[1;32m~\\AppData\\Roaming\\Python\\Python38\\site-packages\\tensorflow\\python\\saved_model\\save.py:1443\u001b[0m, in \u001b[0;36m_build_meta_graph_impl\u001b[1;34m(obj, signatures, options, meta_graph_def)\u001b[0m\n\u001b[0;32m 1440\u001b[0m augmented_graph_view\u001b[39m.\u001b[39mset_signature(signature_map, wrapped_functions)\n\u001b[0;32m 1442\u001b[0m \u001b[39m# Use _SaveableView to provide a frozen listing of properties and functions.\u001b[39;00m\n\u001b[1;32m-> 1443\u001b[0m saveable_view \u001b[39m=\u001b[39m _SaveableView(augmented_graph_view, options)\n\u001b[0;32m 1444\u001b[0m object_saver \u001b[39m=\u001b[39m util\u001b[39m.\u001b[39mTrackableSaver(augmented_graph_view)\n\u001b[0;32m 1445\u001b[0m asset_info, exported_graph \u001b[39m=\u001b[39m _fill_meta_graph_def(\n\u001b[0;32m 1446\u001b[0m meta_graph_def, saveable_view, signatures,\n\u001b[0;32m 1447\u001b[0m options\u001b[39m.\u001b[39mnamespace_whitelist, options\u001b[39m.\u001b[39mexperimental_custom_gradients)\n",
"File \u001b[1;32m~\\AppData\\Roaming\\Python\\Python38\\site-packages\\tensorflow\\python\\saved_model\\save.py:229\u001b[0m, in \u001b[0;36m_SaveableView.__init__\u001b[1;34m(self, augmented_graph_view, options)\u001b[0m\n\u001b[0;32m 224\u001b[0m \u001b[39mself\u001b[39m\u001b[39m.\u001b[39maugmented_graph_view \u001b[39m=\u001b[39m augmented_graph_view\n\u001b[0;32m 225\u001b[0m \u001b[39mself\u001b[39m\u001b[39m.\u001b[39m_options \u001b[39m=\u001b[39m options\n\u001b[0;32m 227\u001b[0m (\u001b[39mself\u001b[39m\u001b[39m.\u001b[39m_trackable_objects, \u001b[39mself\u001b[39m\u001b[39m.\u001b[39mnode_paths, \u001b[39mself\u001b[39m\u001b[39m.\u001b[39mnode_ids,\n\u001b[0;32m 228\u001b[0m \u001b[39mself\u001b[39m\u001b[39m.\u001b[39m_slot_variables, \u001b[39mself\u001b[39m\u001b[39m.\u001b[39mobject_names) \u001b[39m=\u001b[39m (\n\u001b[1;32m--> 229\u001b[0m \u001b[39mself\u001b[39;49m\u001b[39m.\u001b[39;49maugmented_graph_view\u001b[39m.\u001b[39;49mobjects_ids_and_slot_variables_and_paths())\n\u001b[0;32m 231\u001b[0m untraced_functions \u001b[39m=\u001b[39m \u001b[39mself\u001b[39m\u001b[39m.\u001b[39maugmented_graph_view\u001b[39m.\u001b[39muntraced_functions\n\u001b[0;32m 232\u001b[0m \u001b[39mif\u001b[39;00m untraced_functions:\n",
"File \u001b[1;32m~\\AppData\\Roaming\\Python\\Python38\\site-packages\\tensorflow\\python\\training\\tracking\\graph_view.py:544\u001b[0m, in \u001b[0;36mObjectGraphView.objects_ids_and_slot_variables_and_paths\u001b[1;34m(self)\u001b[0m\n\u001b[0;32m 532\u001b[0m \u001b[39mdef\u001b[39;00m \u001b[39mobjects_ids_and_slot_variables_and_paths\u001b[39m(\u001b[39mself\u001b[39m):\n\u001b[0;32m 533\u001b[0m \u001b[39m\"\"\"Traverse the object graph and list all accessible objects.\u001b[39;00m\n\u001b[0;32m 534\u001b[0m \n\u001b[0;32m 535\u001b[0m \u001b[39m Looks for `Trackable` objects which are dependencies of\u001b[39;00m\n\u001b[1;32m (...)\u001b[0m\n\u001b[0;32m 542\u001b[0m \u001b[39m object -> node id, slot variables, object_names)\u001b[39;00m\n\u001b[0;32m 543\u001b[0m \u001b[39m \"\"\"\u001b[39;00m\n\u001b[1;32m--> 544\u001b[0m trackable_objects, node_paths \u001b[39m=\u001b[39m \u001b[39mself\u001b[39;49m\u001b[39m.\u001b[39;49m_breadth_first_traversal()\n\u001b[0;32m 545\u001b[0m object_names \u001b[39m=\u001b[39m object_identity\u001b[39m.\u001b[39mObjectIdentityDictionary()\n\u001b[0;32m 546\u001b[0m \u001b[39mfor\u001b[39;00m obj, path \u001b[39min\u001b[39;00m node_paths\u001b[39m.\u001b[39mitems():\n",
"File \u001b[1;32m~\\AppData\\Roaming\\Python\\Python38\\site-packages\\tensorflow\\python\\training\\tracking\\graph_view.py:255\u001b[0m, in \u001b[0;36mObjectGraphView._breadth_first_traversal\u001b[1;34m(self)\u001b[0m\n\u001b[0;32m 253\u001b[0m current_trackable \u001b[39m=\u001b[39m to_visit\u001b[39m.\u001b[39mpopleft()\n\u001b[0;32m 254\u001b[0m bfs_sorted\u001b[39m.\u001b[39mappend(current_trackable)\n\u001b[1;32m--> 255\u001b[0m \u001b[39mfor\u001b[39;00m name, dependency \u001b[39min\u001b[39;00m \u001b[39mself\u001b[39m\u001b[39m.\u001b[39mlist_children(current_trackable):\n\u001b[0;32m 256\u001b[0m \u001b[39mif\u001b[39;00m dependency \u001b[39mnot\u001b[39;00m \u001b[39min\u001b[39;00m node_paths:\n\u001b[0;32m 257\u001b[0m node_paths[dependency] \u001b[39m=\u001b[39m (\n\u001b[0;32m 258\u001b[0m node_paths[current_trackable] \u001b[39m+\u001b[39m (\n\u001b[0;32m 259\u001b[0m base\u001b[39m.\u001b[39mTrackableReference(name, dependency),))\n",
"File \u001b[1;32m~\\AppData\\Roaming\\Python\\Python38\\site-packages\\tensorflow\\python\\saved_model\\save.py:143\u001b[0m, in \u001b[0;36m_AugmentedGraphView.list_children\u001b[1;34m(self, obj)\u001b[0m\n\u001b[0;32m 140\u001b[0m \u001b[39mif\u001b[39;00m obj \u001b[39mnot\u001b[39;00m \u001b[39min\u001b[39;00m \u001b[39mself\u001b[39m\u001b[39m.\u001b[39m_children_cache:\n\u001b[0;32m 141\u001b[0m children \u001b[39m=\u001b[39m \u001b[39mself\u001b[39m\u001b[39m.\u001b[39m_children_cache[obj] \u001b[39m=\u001b[39m {}\n\u001b[1;32m--> 143\u001b[0m \u001b[39mfor\u001b[39;00m name, child \u001b[39min\u001b[39;00m \u001b[39msuper\u001b[39;49m(_AugmentedGraphView, \u001b[39mself\u001b[39;49m)\u001b[39m.\u001b[39;49mlist_children(\n\u001b[0;32m 144\u001b[0m obj,\n\u001b[0;32m 145\u001b[0m save_type\u001b[39m=\u001b[39;49mbase\u001b[39m.\u001b[39;49mSaveType\u001b[39m.\u001b[39;49mSAVEDMODEL,\n\u001b[0;32m 146\u001b[0m cache\u001b[39m=\u001b[39;49m\u001b[39mself\u001b[39;49m\u001b[39m.\u001b[39;49m_serialization_cache):\n\u001b[0;32m 147\u001b[0m \u001b[39mif\u001b[39;00m \u001b[39misinstance\u001b[39m(child, defun\u001b[39m.\u001b[39mConcreteFunction):\n\u001b[0;32m 148\u001b[0m child \u001b[39m=\u001b[39m \u001b[39mself\u001b[39m\u001b[39m.\u001b[39m_maybe_uncache_variable_captures(child)\n",
"File \u001b[1;32m~\\AppData\\Roaming\\Python\\Python38\\site-packages\\tensorflow\\python\\training\\tracking\\graph_view.py:203\u001b[0m, in \u001b[0;36mObjectGraphView.list_children\u001b[1;34m(self, obj, save_type, **kwargs)\u001b[0m\n\u001b[0;32m 200\u001b[0m \u001b[39m# pylint: disable=protected-access\u001b[39;00m\n\u001b[0;32m 201\u001b[0m obj\u001b[39m.\u001b[39m_maybe_initialize_trackable()\n\u001b[0;32m 202\u001b[0m children \u001b[39m=\u001b[39m [base\u001b[39m.\u001b[39mTrackableReference(name, ref) \u001b[39mfor\u001b[39;00m name, ref\n\u001b[1;32m--> 203\u001b[0m \u001b[39min\u001b[39;00m obj\u001b[39m.\u001b[39;49m_trackable_children(save_type, \u001b[39m*\u001b[39;49m\u001b[39m*\u001b[39;49mkwargs)\u001b[39m.\u001b[39mitems()]\n\u001b[0;32m 204\u001b[0m \u001b[39m# pylint: enable=protected-access\u001b[39;00m\n\u001b[0;32m 205\u001b[0m \n\u001b[0;32m 206\u001b[0m \u001b[39m# GraphView objects may define children of the root object that are not\u001b[39;00m\n\u001b[0;32m 207\u001b[0m \u001b[39m# actually attached, e.g. a Checkpoint object's save_counter.\u001b[39;00m\n\u001b[0;32m 208\u001b[0m \u001b[39mif\u001b[39;00m obj \u001b[39mis\u001b[39;00m \u001b[39mself\u001b[39m\u001b[39m.\u001b[39mroot \u001b[39mand\u001b[39;00m \u001b[39mself\u001b[39m\u001b[39m.\u001b[39m_attached_dependencies:\n",
"File \u001b[1;32m~\\AppData\\Roaming\\Python\\Python38\\site-packages\\tensorflow\\python\\training\\tracking\\autotrackable.py:115\u001b[0m, in \u001b[0;36mAutoTrackable._trackable_children\u001b[1;34m(self, save_type, **kwargs)\u001b[0m\n\u001b[0;32m 113\u001b[0m \u001b[39mfor\u001b[39;00m fn \u001b[39min\u001b[39;00m functions\u001b[39m.\u001b[39mvalues():\n\u001b[0;32m 114\u001b[0m \u001b[39mif\u001b[39;00m \u001b[39misinstance\u001b[39m(fn, core_types\u001b[39m.\u001b[39mGenericFunction):\n\u001b[1;32m--> 115\u001b[0m fn\u001b[39m.\u001b[39;49m_list_all_concrete_functions_for_serialization() \u001b[39m# pylint: disable=protected-access\u001b[39;00m\n\u001b[0;32m 117\u001b[0m \u001b[39m# Additional dependencies may have been generated during function tracing\u001b[39;00m\n\u001b[0;32m 118\u001b[0m \u001b[39m# (e.g. captured variables). Make sure we return those too.\u001b[39;00m\n\u001b[0;32m 119\u001b[0m children \u001b[39m=\u001b[39m {}\n",
"File \u001b[1;32m~\\AppData\\Roaming\\Python\\Python38\\site-packages\\tensorflow\\python\\eager\\def_function.py:1184\u001b[0m, in \u001b[0;36mFunction._list_all_concrete_functions_for_serialization\u001b[1;34m(self)\u001b[0m\n\u001b[0;32m 1182\u001b[0m concrete_functions \u001b[39m=\u001b[39m []\n\u001b[0;32m 1183\u001b[0m \u001b[39mfor\u001b[39;00m args, kwargs \u001b[39min\u001b[39;00m seen_signatures:\n\u001b[1;32m-> 1184\u001b[0m concrete_functions\u001b[39m.\u001b[39mappend(\u001b[39mself\u001b[39;49m\u001b[39m.\u001b[39;49mget_concrete_function(\u001b[39m*\u001b[39;49margs, \u001b[39m*\u001b[39;49m\u001b[39m*\u001b[39;49mkwargs))\n\u001b[0;32m 1185\u001b[0m \u001b[39mreturn\u001b[39;00m concrete_functions\n",
"File \u001b[1;32m~\\AppData\\Roaming\\Python\\Python38\\site-packages\\tensorflow\\python\\eager\\def_function.py:1239\u001b[0m, in \u001b[0;36mFunction.get_concrete_function\u001b[1;34m(self, *args, **kwargs)\u001b[0m\n\u001b[0;32m 1237\u001b[0m \u001b[39mdef\u001b[39;00m \u001b[39mget_concrete_function\u001b[39m(\u001b[39mself\u001b[39m, \u001b[39m*\u001b[39margs, \u001b[39m*\u001b[39m\u001b[39m*\u001b[39mkwargs):\n\u001b[0;32m 1238\u001b[0m \u001b[39m# Implements GenericFunction.get_concrete_function.\u001b[39;00m\n\u001b[1;32m-> 1239\u001b[0m concrete \u001b[39m=\u001b[39m \u001b[39mself\u001b[39;49m\u001b[39m.\u001b[39;49m_get_concrete_function_garbage_collected(\u001b[39m*\u001b[39;49margs, \u001b[39m*\u001b[39;49m\u001b[39m*\u001b[39;49mkwargs)\n\u001b[0;32m 1240\u001b[0m concrete\u001b[39m.\u001b[39m_garbage_collector\u001b[39m.\u001b[39mrelease() \u001b[39m# pylint: disable=protected-access\u001b[39;00m\n\u001b[0;32m 1241\u001b[0m \u001b[39mreturn\u001b[39;00m concrete\n",
"File \u001b[1;32m~\\AppData\\Roaming\\Python\\Python38\\site-packages\\tensorflow\\python\\eager\\def_function.py:1230\u001b[0m, in \u001b[0;36mFunction._get_concrete_function_garbage_collected\u001b[1;34m(self, *args, **kwargs)\u001b[0m\n\u001b[0;32m 1225\u001b[0m \u001b[39mreturn\u001b[39;00m \u001b[39mself\u001b[39m\u001b[39m.\u001b[39m_stateless_fn\u001b[39m.\u001b[39m_get_concrete_function_garbage_collected( \u001b[39m# pylint: disable=protected-access\u001b[39;00m\n\u001b[0;32m 1226\u001b[0m \u001b[39m*\u001b[39margs, \u001b[39m*\u001b[39m\u001b[39m*\u001b[39mkwargs)\n\u001b[0;32m 1227\u001b[0m \u001b[39melif\u001b[39;00m \u001b[39mself\u001b[39m\u001b[39m.\u001b[39m_stateful_fn \u001b[39mis\u001b[39;00m \u001b[39mnot\u001b[39;00m \u001b[39mNone\u001b[39;00m:\n\u001b[0;32m 1228\u001b[0m \u001b[39m# In this case we have not created variables on the first call. So we can\u001b[39;00m\n\u001b[0;32m 1229\u001b[0m \u001b[39m# run the first trace but we should fail if variables are created.\u001b[39;00m\n\u001b[1;32m-> 1230\u001b[0m concrete \u001b[39m=\u001b[39m \u001b[39mself\u001b[39;49m\u001b[39m.\u001b[39;49m_stateful_fn\u001b[39m.\u001b[39;49m_get_concrete_function_garbage_collected( \u001b[39m# pylint: disable=protected-access\u001b[39;49;00m\n\u001b[0;32m 1231\u001b[0m \u001b[39m*\u001b[39;49margs, \u001b[39m*\u001b[39;49m\u001b[39m*\u001b[39;49mkwargs)\n\u001b[0;32m 1232\u001b[0m \u001b[39mif\u001b[39;00m \u001b[39mself\u001b[39m\u001b[39m.\u001b[39m_created_variables:\n\u001b[0;32m 1233\u001b[0m \u001b[39mraise\u001b[39;00m \u001b[39mValueError\u001b[39;00m(\u001b[39m\"\u001b[39m\u001b[39mCreating variables on a non-first call to a function\u001b[39m\u001b[39m\"\u001b[39m\n\u001b[0;32m 1234\u001b[0m \u001b[39m\"\u001b[39m\u001b[39m decorated with tf.function.\u001b[39m\u001b[39m\"\u001b[39m)\n",
"File \u001b[1;32m~\\AppData\\Roaming\\Python\\Python38\\site-packages\\tensorflow\\python\\eager\\function.py:2533\u001b[0m, in \u001b[0;36mFunction._get_concrete_function_garbage_collected\u001b[1;34m(self, *args, **kwargs)\u001b[0m\n\u001b[0;32m 2531\u001b[0m args, kwargs \u001b[39m=\u001b[39m \u001b[39mNone\u001b[39;00m, \u001b[39mNone\u001b[39;00m\n\u001b[0;32m 2532\u001b[0m \u001b[39mwith\u001b[39;00m \u001b[39mself\u001b[39m\u001b[39m.\u001b[39m_lock:\n\u001b[1;32m-> 2533\u001b[0m graph_function, _ \u001b[39m=\u001b[39m \u001b[39mself\u001b[39;49m\u001b[39m.\u001b[39;49m_maybe_define_function(args, kwargs)\n\u001b[0;32m 2534\u001b[0m seen_names \u001b[39m=\u001b[39m \u001b[39mset\u001b[39m()\n\u001b[0;32m 2535\u001b[0m captured \u001b[39m=\u001b[39m object_identity\u001b[39m.\u001b[39mObjectIdentitySet(\n\u001b[0;32m 2536\u001b[0m graph_function\u001b[39m.\u001b[39mgraph\u001b[39m.\u001b[39minternal_captures)\n",
"File \u001b[1;32m~\\AppData\\Roaming\\Python\\Python38\\site-packages\\tensorflow\\python\\eager\\function.py:2711\u001b[0m, in \u001b[0;36mFunction._maybe_define_function\u001b[1;34m(self, args, kwargs)\u001b[0m\n\u001b[0;32m 2708\u001b[0m cache_key \u001b[39m=\u001b[39m \u001b[39mself\u001b[39m\u001b[39m.\u001b[39m_function_cache\u001b[39m.\u001b[39mgeneralize(cache_key)\n\u001b[0;32m 2709\u001b[0m (args, kwargs) \u001b[39m=\u001b[39m cache_key\u001b[39m.\u001b[39m_placeholder_value() \u001b[39m# pylint: disable=protected-access\u001b[39;00m\n\u001b[1;32m-> 2711\u001b[0m graph_function \u001b[39m=\u001b[39m \u001b[39mself\u001b[39;49m\u001b[39m.\u001b[39;49m_create_graph_function(args, kwargs)\n\u001b[0;32m 2712\u001b[0m \u001b[39mself\u001b[39m\u001b[39m.\u001b[39m_function_cache\u001b[39m.\u001b[39madd(cache_key, cache_key_deletion_observer,\n\u001b[0;32m 2713\u001b[0m graph_function)\n\u001b[0;32m 2715\u001b[0m \u001b[39mreturn\u001b[39;00m graph_function, filtered_flat_args\n",
"File \u001b[1;32m~\\AppData\\Roaming\\Python\\Python38\\site-packages\\tensorflow\\python\\eager\\function.py:2627\u001b[0m, in \u001b[0;36mFunction._create_graph_function\u001b[1;34m(self, args, kwargs)\u001b[0m\n\u001b[0;32m 2622\u001b[0m missing_arg_names \u001b[39m=\u001b[39m [\n\u001b[0;32m 2623\u001b[0m \u001b[39m\"\u001b[39m\u001b[39m%s\u001b[39;00m\u001b[39m_\u001b[39m\u001b[39m%d\u001b[39;00m\u001b[39m\"\u001b[39m \u001b[39m%\u001b[39m (arg, i) \u001b[39mfor\u001b[39;00m i, arg \u001b[39min\u001b[39;00m \u001b[39menumerate\u001b[39m(missing_arg_names)\n\u001b[0;32m 2624\u001b[0m ]\n\u001b[0;32m 2625\u001b[0m arg_names \u001b[39m=\u001b[39m base_arg_names \u001b[39m+\u001b[39m missing_arg_names\n\u001b[0;32m 2626\u001b[0m graph_function \u001b[39m=\u001b[39m ConcreteFunction(\n\u001b[1;32m-> 2627\u001b[0m func_graph_module\u001b[39m.\u001b[39;49mfunc_graph_from_py_func(\n\u001b[0;32m 2628\u001b[0m \u001b[39mself\u001b[39;49m\u001b[39m.\u001b[39;49m_name,\n\u001b[0;32m 2629\u001b[0m \u001b[39mself\u001b[39;49m\u001b[39m.\u001b[39;49m_python_function,\n\u001b[0;32m 2630\u001b[0m args,\n\u001b[0;32m 2631\u001b[0m kwargs,\n\u001b[0;32m 2632\u001b[0m \u001b[39mself\u001b[39;49m\u001b[39m.\u001b[39;49minput_signature,\n\u001b[0;32m 2633\u001b[0m autograph\u001b[39m=\u001b[39;49m\u001b[39mself\u001b[39;49m\u001b[39m.\u001b[39;49m_autograph,\n\u001b[0;32m 2634\u001b[0m autograph_options\u001b[39m=\u001b[39;49m\u001b[39mself\u001b[39;49m\u001b[39m.\u001b[39;49m_autograph_options,\n\u001b[0;32m 2635\u001b[0m arg_names\u001b[39m=\u001b[39;49marg_names,\n\u001b[0;32m 2636\u001b[0m capture_by_value\u001b[39m=\u001b[39;49m\u001b[39mself\u001b[39;49m\u001b[39m.\u001b[39;49m_capture_by_value),\n\u001b[0;32m 2637\u001b[0m \u001b[39mself\u001b[39m\u001b[39m.\u001b[39m_function_attributes,\n\u001b[0;32m 2638\u001b[0m spec\u001b[39m=\u001b[39m\u001b[39mself\u001b[39m\u001b[39m.\u001b[39mfunction_spec,\n\u001b[0;32m 2639\u001b[0m \u001b[39m# Tell the ConcreteFunction to clean up its graph once it goes out of\u001b[39;00m\n\u001b[0;32m 2640\u001b[0m \u001b[39m# scope. This is not the default behavior since it gets used in some\u001b[39;00m\n\u001b[0;32m 2641\u001b[0m \u001b[39m# places (like Keras) where the FuncGraph lives longer than the\u001b[39;00m\n\u001b[0;32m 2642\u001b[0m \u001b[39m# ConcreteFunction.\u001b[39;00m\n\u001b[0;32m 2643\u001b[0m shared_func_graph\u001b[39m=\u001b[39m\u001b[39mFalse\u001b[39;00m)\n\u001b[0;32m 2644\u001b[0m \u001b[39mreturn\u001b[39;00m graph_function\n",
"File \u001b[1;32m~\\AppData\\Roaming\\Python\\Python38\\site-packages\\tensorflow\\python\\framework\\func_graph.py:1141\u001b[0m, in \u001b[0;36mfunc_graph_from_py_func\u001b[1;34m(name, python_func, args, kwargs, signature, func_graph, autograph, autograph_options, add_control_dependencies, arg_names, op_return_value, collections, capture_by_value, acd_record_initial_resource_uses)\u001b[0m\n\u001b[0;32m 1138\u001b[0m \u001b[39melse\u001b[39;00m:\n\u001b[0;32m 1139\u001b[0m _, original_func \u001b[39m=\u001b[39m tf_decorator\u001b[39m.\u001b[39munwrap(python_func)\n\u001b[1;32m-> 1141\u001b[0m func_outputs \u001b[39m=\u001b[39m python_func(\u001b[39m*\u001b[39;49mfunc_args, \u001b[39m*\u001b[39;49m\u001b[39m*\u001b[39;49mfunc_kwargs)\n\u001b[0;32m 1143\u001b[0m \u001b[39m# invariant: `func_outputs` contains only Tensors, CompositeTensors,\u001b[39;00m\n\u001b[0;32m 1144\u001b[0m \u001b[39m# TensorArrays and `None`s.\u001b[39;00m\n\u001b[0;32m 1145\u001b[0m func_outputs \u001b[39m=\u001b[39m nest\u001b[39m.\u001b[39mmap_structure(\n\u001b[0;32m 1146\u001b[0m convert, func_outputs, expand_composites\u001b[39m=\u001b[39m\u001b[39mTrue\u001b[39;00m)\n",
"File \u001b[1;32m~\\AppData\\Roaming\\Python\\Python38\\site-packages\\tensorflow\\python\\eager\\def_function.py:677\u001b[0m, in \u001b[0;36mFunction._defun_with_scope.<locals>.wrapped_fn\u001b[1;34m(*args, **kwds)\u001b[0m\n\u001b[0;32m 673\u001b[0m \u001b[39mwith\u001b[39;00m default_graph\u001b[39m.\u001b[39m_variable_creator_scope(scope, priority\u001b[39m=\u001b[39m\u001b[39m50\u001b[39m): \u001b[39m# pylint: disable=protected-access\u001b[39;00m\n\u001b[0;32m 674\u001b[0m \u001b[39m# __wrapped__ allows AutoGraph to swap in a converted function. We give\u001b[39;00m\n\u001b[0;32m 675\u001b[0m \u001b[39m# the function a weak reference to itself to avoid a reference cycle.\u001b[39;00m\n\u001b[0;32m 676\u001b[0m \u001b[39mwith\u001b[39;00m OptionalXlaContext(compile_with_xla):\n\u001b[1;32m--> 677\u001b[0m out \u001b[39m=\u001b[39m weak_wrapped_fn()\u001b[39m.\u001b[39;49m__wrapped__(\u001b[39m*\u001b[39;49margs, \u001b[39m*\u001b[39;49m\u001b[39m*\u001b[39;49mkwds)\n\u001b[0;32m 678\u001b[0m \u001b[39mreturn\u001b[39;00m out\n",
"File \u001b[1;32m~\\AppData\\Roaming\\Python\\Python38\\site-packages\\keras\\saving\\saved_model\\save_impl.py:572\u001b[0m, in \u001b[0;36mlayer_call_wrapper.<locals>.wrapper\u001b[1;34m(*args, **kwargs)\u001b[0m\n\u001b[0;32m 567\u001b[0m \u001b[39mwith\u001b[39;00m base_layer_utils\u001b[39m.\u001b[39mcall_context()\u001b[39m.\u001b[39menter(\n\u001b[0;32m 568\u001b[0m layer, inputs\u001b[39m=\u001b[39minputs, build_graph\u001b[39m=\u001b[39m\u001b[39mFalse\u001b[39;00m, training\u001b[39m=\u001b[39mtraining,\n\u001b[0;32m 569\u001b[0m saving\u001b[39m=\u001b[39m\u001b[39mTrue\u001b[39;00m):\n\u001b[0;32m 570\u001b[0m \u001b[39mwith\u001b[39;00m autocast_variable\u001b[39m.\u001b[39menable_auto_cast_variables(\n\u001b[0;32m 571\u001b[0m layer\u001b[39m.\u001b[39m_compute_dtype_object): \u001b[39m# pylint: disable=protected-access\u001b[39;00m\n\u001b[1;32m--> 572\u001b[0m ret \u001b[39m=\u001b[39m method(\u001b[39m*\u001b[39;49margs, \u001b[39m*\u001b[39;49m\u001b[39m*\u001b[39;49mkwargs)\n\u001b[0;32m 573\u001b[0m _restore_layer_losses(original_losses)\n\u001b[0;32m 574\u001b[0m \u001b[39mreturn\u001b[39;00m ret\n",
"File \u001b[1;32m~\\AppData\\Roaming\\Python\\Python38\\site-packages\\keras\\saving\\saved_model\\utils.py:168\u001b[0m, in \u001b[0;36mmaybe_add_training_arg.<locals>.wrap_with_training_arg\u001b[1;34m(*args, **kwargs)\u001b[0m\n\u001b[0;32m 165\u001b[0m set_training_arg(training, training_arg_index, args, kwargs)\n\u001b[0;32m 166\u001b[0m \u001b[39mreturn\u001b[39;00m wrapped_call(\u001b[39m*\u001b[39margs, \u001b[39m*\u001b[39m\u001b[39m*\u001b[39mkwargs)\n\u001b[1;32m--> 168\u001b[0m \u001b[39mreturn\u001b[39;00m control_flow_util\u001b[39m.\u001b[39;49msmart_cond(\n\u001b[0;32m 169\u001b[0m training, \u001b[39mlambda\u001b[39;49;00m: replace_training_and_call(\u001b[39mTrue\u001b[39;49;00m),\n\u001b[0;32m 170\u001b[0m \u001b[39mlambda\u001b[39;49;00m: replace_training_and_call(\u001b[39mFalse\u001b[39;49;00m))\n",
"File \u001b[1;32m~\\AppData\\Roaming\\Python\\Python38\\site-packages\\keras\\utils\\control_flow_util.py:105\u001b[0m, in \u001b[0;36msmart_cond\u001b[1;34m(pred, true_fn, false_fn, name)\u001b[0m\n\u001b[0;32m 102\u001b[0m \u001b[39mif\u001b[39;00m \u001b[39misinstance\u001b[39m(pred, tf\u001b[39m.\u001b[39mVariable):\n\u001b[0;32m 103\u001b[0m \u001b[39mreturn\u001b[39;00m tf\u001b[39m.\u001b[39mcond(\n\u001b[0;32m 104\u001b[0m pred, true_fn\u001b[39m=\u001b[39mtrue_fn, false_fn\u001b[39m=\u001b[39mfalse_fn, name\u001b[39m=\u001b[39mname)\n\u001b[1;32m--> 105\u001b[0m \u001b[39mreturn\u001b[39;00m tf\u001b[39m.\u001b[39;49m__internal__\u001b[39m.\u001b[39;49msmart_cond\u001b[39m.\u001b[39;49msmart_cond(\n\u001b[0;32m 106\u001b[0m pred, true_fn\u001b[39m=\u001b[39;49mtrue_fn, false_fn\u001b[39m=\u001b[39;49mfalse_fn, name\u001b[39m=\u001b[39;49mname)\n",
"File \u001b[1;32m~\\AppData\\Roaming\\Python\\Python38\\site-packages\\tensorflow\\python\\framework\\smart_cond.py:53\u001b[0m, in \u001b[0;36msmart_cond\u001b[1;34m(pred, true_fn, false_fn, name)\u001b[0m\n\u001b[0;32m 51\u001b[0m \u001b[39mif\u001b[39;00m pred_value \u001b[39mis\u001b[39;00m \u001b[39mnot\u001b[39;00m \u001b[39mNone\u001b[39;00m:\n\u001b[0;32m 52\u001b[0m \u001b[39mif\u001b[39;00m pred_value:\n\u001b[1;32m---> 53\u001b[0m \u001b[39mreturn\u001b[39;00m true_fn()\n\u001b[0;32m 54\u001b[0m \u001b[39melse\u001b[39;00m:\n\u001b[0;32m 55\u001b[0m \u001b[39mreturn\u001b[39;00m false_fn()\n",
"File \u001b[1;32m~\\AppData\\Roaming\\Python\\Python38\\site-packages\\keras\\saving\\saved_model\\utils.py:169\u001b[0m, in \u001b[0;36mmaybe_add_training_arg.<locals>.wrap_with_training_arg.<locals>.<lambda>\u001b[1;34m()\u001b[0m\n\u001b[0;32m 165\u001b[0m set_training_arg(training, training_arg_index, args, kwargs)\n\u001b[0;32m 166\u001b[0m \u001b[39mreturn\u001b[39;00m wrapped_call(\u001b[39m*\u001b[39margs, \u001b[39m*\u001b[39m\u001b[39m*\u001b[39mkwargs)\n\u001b[0;32m 168\u001b[0m \u001b[39mreturn\u001b[39;00m control_flow_util\u001b[39m.\u001b[39msmart_cond(\n\u001b[1;32m--> 169\u001b[0m training, \u001b[39mlambda\u001b[39;00m: replace_training_and_call(\u001b[39mTrue\u001b[39;49;00m),\n\u001b[0;32m 170\u001b[0m \u001b[39mlambda\u001b[39;00m: replace_training_and_call(\u001b[39mFalse\u001b[39;00m))\n",
"File \u001b[1;32m~\\AppData\\Roaming\\Python\\Python38\\site-packages\\keras\\saving\\saved_model\\utils.py:166\u001b[0m, in \u001b[0;36mmaybe_add_training_arg.<locals>.wrap_with_training_arg.<locals>.replace_training_and_call\u001b[1;34m(training)\u001b[0m\n\u001b[0;32m 164\u001b[0m \u001b[39mdef\u001b[39;00m \u001b[39mreplace_training_and_call\u001b[39m(training):\n\u001b[0;32m 165\u001b[0m set_training_arg(training, training_arg_index, args, kwargs)\n\u001b[1;32m--> 166\u001b[0m \u001b[39mreturn\u001b[39;00m wrapped_call(\u001b[39m*\u001b[39;49margs, \u001b[39m*\u001b[39;49m\u001b[39m*\u001b[39;49mkwargs)\n",
"File \u001b[1;32m~\\AppData\\Roaming\\Python\\Python38\\site-packages\\keras\\saving\\saved_model\\save_impl.py:634\u001b[0m, in \u001b[0;36m_wrap_call_and_conditional_losses.<locals>.call_and_return_conditional_losses\u001b[1;34m(*args, **kwargs)\u001b[0m\n\u001b[0;32m 632\u001b[0m \u001b[39mdef\u001b[39;00m \u001b[39mcall_and_return_conditional_losses\u001b[39m(\u001b[39m*\u001b[39margs, \u001b[39m*\u001b[39m\u001b[39m*\u001b[39mkwargs):\n\u001b[0;32m 633\u001b[0m \u001b[39m\"\"\"Returns layer (call_output, conditional losses) tuple.\"\"\"\u001b[39;00m\n\u001b[1;32m--> 634\u001b[0m call_output \u001b[39m=\u001b[39m layer_call(\u001b[39m*\u001b[39;49margs, \u001b[39m*\u001b[39;49m\u001b[39m*\u001b[39;49mkwargs)\n\u001b[0;32m 635\u001b[0m \u001b[39mif\u001b[39;00m version_utils\u001b[39m.\u001b[39mis_v1_layer_or_model(layer):\n\u001b[0;32m 636\u001b[0m conditional_losses \u001b[39m=\u001b[39m layer\u001b[39m.\u001b[39mget_losses_for(\n\u001b[0;32m 637\u001b[0m _filtered_inputs([args, kwargs]))\n",
"File \u001b[1;32m~\\AppData\\Roaming\\Python\\Python38\\site-packages\\keras\\layers\\rnn\\bidirectional.py:366\u001b[0m, in \u001b[0;36mBidirectional.call\u001b[1;34m(self, inputs, training, mask, initial_state, constants)\u001b[0m\n\u001b[0;32m 362\u001b[0m forward_state, backward_state \u001b[39m=\u001b[39m \u001b[39mNone\u001b[39;00m, \u001b[39mNone\u001b[39;00m\n\u001b[0;32m 364\u001b[0m y \u001b[39m=\u001b[39m \u001b[39mself\u001b[39m\u001b[39m.\u001b[39mforward_layer(forward_inputs,\n\u001b[0;32m 365\u001b[0m initial_state\u001b[39m=\u001b[39mforward_state, \u001b[39m*\u001b[39m\u001b[39m*\u001b[39mkwargs)\n\u001b[1;32m--> 366\u001b[0m y_rev \u001b[39m=\u001b[39m \u001b[39mself\u001b[39;49m\u001b[39m.\u001b[39;49mbackward_layer(backward_inputs,\n\u001b[0;32m 367\u001b[0m initial_state\u001b[39m=\u001b[39;49mbackward_state, \u001b[39m*\u001b[39;49m\u001b[39m*\u001b[39;49mkwargs)\n\u001b[0;32m 368\u001b[0m \u001b[39melse\u001b[39;00m:\n\u001b[0;32m 369\u001b[0m y \u001b[39m=\u001b[39m \u001b[39mself\u001b[39m\u001b[39m.\u001b[39mforward_layer(inputs, \u001b[39m*\u001b[39m\u001b[39m*\u001b[39mkwargs)\n",
"File \u001b[1;32m~\\AppData\\Roaming\\Python\\Python38\\site-packages\\keras\\layers\\rnn\\base_rnn.py:515\u001b[0m, in \u001b[0;36mRNN.__call__\u001b[1;34m(self, inputs, initial_state, constants, **kwargs)\u001b[0m\n\u001b[0;32m 511\u001b[0m inputs, initial_state, constants \u001b[39m=\u001b[39m rnn_utils\u001b[39m.\u001b[39mstandardize_args(\n\u001b[0;32m 512\u001b[0m inputs, initial_state, constants, \u001b[39mself\u001b[39m\u001b[39m.\u001b[39m_num_constants)\n\u001b[0;32m 514\u001b[0m \u001b[39mif\u001b[39;00m initial_state \u001b[39mis\u001b[39;00m \u001b[39mNone\u001b[39;00m \u001b[39mand\u001b[39;00m constants \u001b[39mis\u001b[39;00m \u001b[39mNone\u001b[39;00m:\n\u001b[1;32m--> 515\u001b[0m \u001b[39mreturn\u001b[39;00m \u001b[39msuper\u001b[39;49m(RNN, \u001b[39mself\u001b[39;49m)\u001b[39m.\u001b[39;49m\u001b[39m__call__\u001b[39;49m(inputs, \u001b[39m*\u001b[39;49m\u001b[39m*\u001b[39;49mkwargs)\n\u001b[0;32m 517\u001b[0m \u001b[39m# If any of `initial_state` or `constants` are specified and are Keras\u001b[39;00m\n\u001b[0;32m 518\u001b[0m \u001b[39m# tensors, then add them to the inputs and temporarily modify the\u001b[39;00m\n\u001b[0;32m 519\u001b[0m \u001b[39m# input_spec to include them.\u001b[39;00m\n\u001b[0;32m 521\u001b[0m additional_inputs \u001b[39m=\u001b[39m []\n",
"File \u001b[1;32m~\\AppData\\Roaming\\Python\\Python38\\site-packages\\keras\\utils\\traceback_utils.py:64\u001b[0m, in \u001b[0;36mfilter_traceback.<locals>.error_handler\u001b[1;34m(*args, **kwargs)\u001b[0m\n\u001b[0;32m 62\u001b[0m filtered_tb \u001b[39m=\u001b[39m \u001b[39mNone\u001b[39;00m\n\u001b[0;32m 63\u001b[0m \u001b[39mtry\u001b[39;00m:\n\u001b[1;32m---> 64\u001b[0m \u001b[39mreturn\u001b[39;00m fn(\u001b[39m*\u001b[39;49margs, \u001b[39m*\u001b[39;49m\u001b[39m*\u001b[39;49mkwargs)\n\u001b[0;32m 65\u001b[0m \u001b[39mexcept\u001b[39;00m \u001b[39mException\u001b[39;00m \u001b[39mas\u001b[39;00m e: \u001b[39m# pylint: disable=broad-except\u001b[39;00m\n\u001b[0;32m 66\u001b[0m filtered_tb \u001b[39m=\u001b[39m _process_traceback_frames(e\u001b[39m.\u001b[39m__traceback__)\n",
"File \u001b[1;32m~\\AppData\\Roaming\\Python\\Python38\\site-packages\\keras\\engine\\base_layer.py:1014\u001b[0m, in \u001b[0;36mLayer.__call__\u001b[1;34m(self, *args, **kwargs)\u001b[0m\n\u001b[0;32m 1010\u001b[0m inputs \u001b[39m=\u001b[39m \u001b[39mself\u001b[39m\u001b[39m.\u001b[39m_maybe_cast_inputs(inputs, input_list)\n\u001b[0;32m 1012\u001b[0m \u001b[39mwith\u001b[39;00m autocast_variable\u001b[39m.\u001b[39menable_auto_cast_variables(\n\u001b[0;32m 1013\u001b[0m \u001b[39mself\u001b[39m\u001b[39m.\u001b[39m_compute_dtype_object):\n\u001b[1;32m-> 1014\u001b[0m outputs \u001b[39m=\u001b[39m call_fn(inputs, \u001b[39m*\u001b[39;49margs, \u001b[39m*\u001b[39;49m\u001b[39m*\u001b[39;49mkwargs)\n\u001b[0;32m 1016\u001b[0m \u001b[39mif\u001b[39;00m \u001b[39mself\u001b[39m\u001b[39m.\u001b[39m_activity_regularizer:\n\u001b[0;32m 1017\u001b[0m \u001b[39mself\u001b[39m\u001b[39m.\u001b[39m_handle_activity_regularization(inputs, outputs)\n",
"File \u001b[1;32m~\\AppData\\Roaming\\Python\\Python38\\site-packages\\keras\\utils\\traceback_utils.py:92\u001b[0m, in \u001b[0;36minject_argument_info_in_traceback.<locals>.error_handler\u001b[1;34m(*args, **kwargs)\u001b[0m\n\u001b[0;32m 90\u001b[0m bound_signature \u001b[39m=\u001b[39m \u001b[39mNone\u001b[39;00m\n\u001b[0;32m 91\u001b[0m \u001b[39mtry\u001b[39;00m:\n\u001b[1;32m---> 92\u001b[0m \u001b[39mreturn\u001b[39;00m fn(\u001b[39m*\u001b[39;49margs, \u001b[39m*\u001b[39;49m\u001b[39m*\u001b[39;49mkwargs)\n\u001b[0;32m 93\u001b[0m \u001b[39mexcept\u001b[39;00m \u001b[39mException\u001b[39;00m \u001b[39mas\u001b[39;00m e: \u001b[39m# pylint: disable=broad-except\u001b[39;00m\n\u001b[0;32m 94\u001b[0m \u001b[39mif\u001b[39;00m \u001b[39mhasattr\u001b[39m(e, \u001b[39m'\u001b[39m\u001b[39m_keras_call_info_injected\u001b[39m\u001b[39m'\u001b[39m):\n\u001b[0;32m 95\u001b[0m \u001b[39m# Only inject info for the innermost failing call\u001b[39;00m\n",
"File \u001b[1;32m~\\AppData\\Roaming\\Python\\Python38\\site-packages\\keras\\layers\\rnn\\lstm.py:673\u001b[0m, in \u001b[0;36mLSTM.call\u001b[1;34m(self, inputs, mask, training, initial_state)\u001b[0m\n\u001b[0;32m 669\u001b[0m last_output, outputs, new_h, new_c, runtime \u001b[39m=\u001b[39m standard_lstm(\n\u001b[0;32m 670\u001b[0m \u001b[39m*\u001b[39m\u001b[39m*\u001b[39mnormal_lstm_kwargs)\n\u001b[0;32m 671\u001b[0m \u001b[39melse\u001b[39;00m:\n\u001b[0;32m 672\u001b[0m (last_output, outputs, new_h, new_c,\n\u001b[1;32m--> 673\u001b[0m runtime) \u001b[39m=\u001b[39m lstm_with_backend_selection(\u001b[39m*\u001b[39;49m\u001b[39m*\u001b[39;49mnormal_lstm_kwargs)\n\u001b[0;32m 675\u001b[0m states \u001b[39m=\u001b[39m [new_h, new_c]\n\u001b[0;32m 677\u001b[0m \u001b[39mif\u001b[39;00m \u001b[39mself\u001b[39m\u001b[39m.\u001b[39mstateful:\n",
"File \u001b[1;32m~\\AppData\\Roaming\\Python\\Python38\\site-packages\\keras\\layers\\rnn\\lstm.py:1183\u001b[0m, in \u001b[0;36mlstm_with_backend_selection\u001b[1;34m(inputs, init_h, init_c, kernel, recurrent_kernel, bias, mask, time_major, go_backwards, sequence_lengths, zero_output_for_mask, return_sequences)\u001b[0m\n\u001b[0;32m 1177\u001b[0m defun_gpu_lstm \u001b[39m=\u001b[39m gru_lstm_utils\u001b[39m.\u001b[39mgenerate_defun_backend(\n\u001b[0;32m 1178\u001b[0m api_name, gru_lstm_utils\u001b[39m.\u001b[39mGPU_DEVICE_NAME, gpu_lstm_with_fallback,\n\u001b[0;32m 1179\u001b[0m supportive_attribute)\n\u001b[0;32m 1181\u001b[0m \u001b[39m# Call the normal LSTM impl and register the cuDNN impl function. The\u001b[39;00m\n\u001b[0;32m 1182\u001b[0m \u001b[39m# grappler will kick in during session execution to optimize the graph.\u001b[39;00m\n\u001b[1;32m-> 1183\u001b[0m last_output, outputs, new_h, new_c, runtime \u001b[39m=\u001b[39m defun_standard_lstm(\u001b[39m*\u001b[39;49m\u001b[39m*\u001b[39;49mparams)\n\u001b[0;32m 1184\u001b[0m gru_lstm_utils\u001b[39m.\u001b[39mfunction_register(defun_gpu_lstm, \u001b[39m*\u001b[39m\u001b[39m*\u001b[39mparams)\n\u001b[0;32m 1186\u001b[0m \u001b[39mreturn\u001b[39;00m last_output, outputs, new_h, new_c, runtime\n",
"File \u001b[1;32m~\\AppData\\Roaming\\Python\\Python38\\site-packages\\tensorflow\\python\\eager\\function.py:2452\u001b[0m, in \u001b[0;36mFunction.__call__\u001b[1;34m(self, *args, **kwargs)\u001b[0m\n\u001b[0;32m 2449\u001b[0m \u001b[39m\"\"\"Calls a graph function specialized to the inputs.\"\"\"\u001b[39;00m\n\u001b[0;32m 2450\u001b[0m \u001b[39mwith\u001b[39;00m \u001b[39mself\u001b[39m\u001b[39m.\u001b[39m_lock:\n\u001b[0;32m 2451\u001b[0m (graph_function,\n\u001b[1;32m-> 2452\u001b[0m filtered_flat_args) \u001b[39m=\u001b[39m \u001b[39mself\u001b[39;49m\u001b[39m.\u001b[39;49m_maybe_define_function(args, kwargs)\n\u001b[0;32m 2453\u001b[0m \u001b[39mreturn\u001b[39;00m graph_function\u001b[39m.\u001b[39m_call_flat(\n\u001b[0;32m 2454\u001b[0m filtered_flat_args, captured_inputs\u001b[39m=\u001b[39mgraph_function\u001b[39m.\u001b[39mcaptured_inputs)\n",
"File \u001b[1;32m~\\AppData\\Roaming\\Python\\Python38\\site-packages\\tensorflow\\python\\eager\\function.py:2711\u001b[0m, in \u001b[0;36mFunction._maybe_define_function\u001b[1;34m(self, args, kwargs)\u001b[0m\n\u001b[0;32m 2708\u001b[0m cache_key \u001b[39m=\u001b[39m \u001b[39mself\u001b[39m\u001b[39m.\u001b[39m_function_cache\u001b[39m.\u001b[39mgeneralize(cache_key)\n\u001b[0;32m 2709\u001b[0m (args, kwargs) \u001b[39m=\u001b[39m cache_key\u001b[39m.\u001b[39m_placeholder_value() \u001b[39m# pylint: disable=protected-access\u001b[39;00m\n\u001b[1;32m-> 2711\u001b[0m graph_function \u001b[39m=\u001b[39m \u001b[39mself\u001b[39;49m\u001b[39m.\u001b[39;49m_create_graph_function(args, kwargs)\n\u001b[0;32m 2712\u001b[0m \u001b[39mself\u001b[39m\u001b[39m.\u001b[39m_function_cache\u001b[39m.\u001b[39madd(cache_key, cache_key_deletion_observer,\n\u001b[0;32m 2713\u001b[0m graph_function)\n\u001b[0;32m 2715\u001b[0m \u001b[39mreturn\u001b[39;00m graph_function, filtered_flat_args\n",
"File \u001b[1;32m~\\AppData\\Roaming\\Python\\Python38\\site-packages\\tensorflow\\python\\eager\\function.py:2627\u001b[0m, in \u001b[0;36mFunction._create_graph_function\u001b[1;34m(self, args, kwargs)\u001b[0m\n\u001b[0;32m 2622\u001b[0m missing_arg_names \u001b[39m=\u001b[39m [\n\u001b[0;32m 2623\u001b[0m \u001b[39m\"\u001b[39m\u001b[39m%s\u001b[39;00m\u001b[39m_\u001b[39m\u001b[39m%d\u001b[39;00m\u001b[39m\"\u001b[39m \u001b[39m%\u001b[39m (arg, i) \u001b[39mfor\u001b[39;00m i, arg \u001b[39min\u001b[39;00m \u001b[39menumerate\u001b[39m(missing_arg_names)\n\u001b[0;32m 2624\u001b[0m ]\n\u001b[0;32m 2625\u001b[0m arg_names \u001b[39m=\u001b[39m base_arg_names \u001b[39m+\u001b[39m missing_arg_names\n\u001b[0;32m 2626\u001b[0m graph_function \u001b[39m=\u001b[39m ConcreteFunction(\n\u001b[1;32m-> 2627\u001b[0m func_graph_module\u001b[39m.\u001b[39;49mfunc_graph_from_py_func(\n\u001b[0;32m 2628\u001b[0m \u001b[39mself\u001b[39;49m\u001b[39m.\u001b[39;49m_name,\n\u001b[0;32m 2629\u001b[0m \u001b[39mself\u001b[39;49m\u001b[39m.\u001b[39;49m_python_function,\n\u001b[0;32m 2630\u001b[0m args,\n\u001b[0;32m 2631\u001b[0m kwargs,\n\u001b[0;32m 2632\u001b[0m \u001b[39mself\u001b[39;49m\u001b[39m.\u001b[39;49minput_signature,\n\u001b[0;32m 2633\u001b[0m autograph\u001b[39m=\u001b[39;49m\u001b[39mself\u001b[39;49m\u001b[39m.\u001b[39;49m_autograph,\n\u001b[0;32m 2634\u001b[0m autograph_options\u001b[39m=\u001b[39;49m\u001b[39mself\u001b[39;49m\u001b[39m.\u001b[39;49m_autograph_options,\n\u001b[0;32m 2635\u001b[0m arg_names\u001b[39m=\u001b[39;49marg_names,\n\u001b[0;32m 2636\u001b[0m capture_by_value\u001b[39m=\u001b[39;49m\u001b[39mself\u001b[39;49m\u001b[39m.\u001b[39;49m_capture_by_value),\n\u001b[0;32m 2637\u001b[0m \u001b[39mself\u001b[39m\u001b[39m.\u001b[39m_function_attributes,\n\u001b[0;32m 2638\u001b[0m spec\u001b[39m=\u001b[39m\u001b[39mself\u001b[39m\u001b[39m.\u001b[39mfunction_spec,\n\u001b[0;32m 2639\u001b[0m \u001b[39m# Tell the ConcreteFunction to clean up its graph once it goes out of\u001b[39;00m\n\u001b[0;32m 2640\u001b[0m \u001b[39m# scope. This is not the default behavior since it gets used in some\u001b[39;00m\n\u001b[0;32m 2641\u001b[0m \u001b[39m# places (like Keras) where the FuncGraph lives longer than the\u001b[39;00m\n\u001b[0;32m 2642\u001b[0m \u001b[39m# ConcreteFunction.\u001b[39;00m\n\u001b[0;32m 2643\u001b[0m shared_func_graph\u001b[39m=\u001b[39m\u001b[39mFalse\u001b[39;00m)\n\u001b[0;32m 2644\u001b[0m \u001b[39mreturn\u001b[39;00m graph_function\n",
"File \u001b[1;32m~\\AppData\\Roaming\\Python\\Python38\\site-packages\\tensorflow\\python\\framework\\func_graph.py:1141\u001b[0m, in \u001b[0;36mfunc_graph_from_py_func\u001b[1;34m(name, python_func, args, kwargs, signature, func_graph, autograph, autograph_options, add_control_dependencies, arg_names, op_return_value, collections, capture_by_value, acd_record_initial_resource_uses)\u001b[0m\n\u001b[0;32m 1138\u001b[0m \u001b[39melse\u001b[39;00m:\n\u001b[0;32m 1139\u001b[0m _, original_func \u001b[39m=\u001b[39m tf_decorator\u001b[39m.\u001b[39munwrap(python_func)\n\u001b[1;32m-> 1141\u001b[0m func_outputs \u001b[39m=\u001b[39m python_func(\u001b[39m*\u001b[39;49mfunc_args, \u001b[39m*\u001b[39;49m\u001b[39m*\u001b[39;49mfunc_kwargs)\n\u001b[0;32m 1143\u001b[0m \u001b[39m# invariant: `func_outputs` contains only Tensors, CompositeTensors,\u001b[39;00m\n\u001b[0;32m 1144\u001b[0m \u001b[39m# TensorArrays and `None`s.\u001b[39;00m\n\u001b[0;32m 1145\u001b[0m func_outputs \u001b[39m=\u001b[39m nest\u001b[39m.\u001b[39mmap_structure(\n\u001b[0;32m 1146\u001b[0m convert, func_outputs, expand_composites\u001b[39m=\u001b[39m\u001b[39mTrue\u001b[39;00m)\n",
"File \u001b[1;32m~\\AppData\\Roaming\\Python\\Python38\\site-packages\\keras\\layers\\rnn\\lstm.py:891\u001b[0m, in \u001b[0;36mstandard_lstm\u001b[1;34m(inputs, init_h, init_c, kernel, recurrent_kernel, bias, mask, time_major, go_backwards, sequence_lengths, zero_output_for_mask, return_sequences)\u001b[0m\n\u001b[0;32m 888\u001b[0m h \u001b[39m=\u001b[39m o \u001b[39m*\u001b[39m tf\u001b[39m.\u001b[39mtanh(c)\n\u001b[0;32m 889\u001b[0m \u001b[39mreturn\u001b[39;00m h, [h, c]\n\u001b[1;32m--> 891\u001b[0m last_output, outputs, new_states \u001b[39m=\u001b[39m backend\u001b[39m.\u001b[39;49mrnn(\n\u001b[0;32m 892\u001b[0m step,\n\u001b[0;32m 893\u001b[0m inputs, [init_h, init_c],\n\u001b[0;32m 894\u001b[0m constants\u001b[39m=\u001b[39;49m\u001b[39mNone\u001b[39;49;00m,\n\u001b[0;32m 895\u001b[0m unroll\u001b[39m=\u001b[39;49m\u001b[39mFalse\u001b[39;49;00m,\n\u001b[0;32m 896\u001b[0m time_major\u001b[39m=\u001b[39;49mtime_major,\n\u001b[0;32m 897\u001b[0m mask\u001b[39m=\u001b[39;49mmask,\n\u001b[0;32m 898\u001b[0m go_backwards\u001b[39m=\u001b[39;49mgo_backwards,\n\u001b[0;32m 899\u001b[0m input_length\u001b[39m=\u001b[39;49m(sequence_lengths\n\u001b[0;32m 900\u001b[0m \u001b[39mif\u001b[39;49;00m sequence_lengths \u001b[39mis\u001b[39;49;00m \u001b[39mnot\u001b[39;49;00m \u001b[39mNone\u001b[39;49;00m \u001b[39melse\u001b[39;49;00m timesteps),\n\u001b[0;32m 901\u001b[0m zero_output_for_mask\u001b[39m=\u001b[39;49mzero_output_for_mask,\n\u001b[0;32m 902\u001b[0m return_all_outputs\u001b[39m=\u001b[39;49mreturn_sequences)\n\u001b[0;32m 903\u001b[0m \u001b[39mreturn\u001b[39;00m (last_output, outputs, new_states[\u001b[39m0\u001b[39m], new_states[\u001b[39m1\u001b[39m],\n\u001b[0;32m 904\u001b[0m gru_lstm_utils\u001b[39m.\u001b[39mruntime(gru_lstm_utils\u001b[39m.\u001b[39mRUNTIME_CPU))\n",
"File \u001b[1;32m~\\AppData\\Roaming\\Python\\Python38\\site-packages\\tensorflow\\python\\util\\traceback_utils.py:150\u001b[0m, in \u001b[0;36mfilter_traceback.<locals>.error_handler\u001b[1;34m(*args, **kwargs)\u001b[0m\n\u001b[0;32m 148\u001b[0m filtered_tb \u001b[39m=\u001b[39m \u001b[39mNone\u001b[39;00m\n\u001b[0;32m 149\u001b[0m \u001b[39mtry\u001b[39;00m:\n\u001b[1;32m--> 150\u001b[0m \u001b[39mreturn\u001b[39;00m fn(\u001b[39m*\u001b[39;49margs, \u001b[39m*\u001b[39;49m\u001b[39m*\u001b[39;49mkwargs)\n\u001b[0;32m 151\u001b[0m \u001b[39mexcept\u001b[39;00m \u001b[39mException\u001b[39;00m \u001b[39mas\u001b[39;00m e:\n\u001b[0;32m 152\u001b[0m filtered_tb \u001b[39m=\u001b[39m _process_traceback_frames(e\u001b[39m.\u001b[39m__traceback__)\n",
"File \u001b[1;32m~\\AppData\\Roaming\\Python\\Python38\\site-packages\\tensorflow\\python\\util\\dispatch.py:1082\u001b[0m, in \u001b[0;36madd_dispatch_support.<locals>.decorator.<locals>.op_dispatch_handler\u001b[1;34m(*args, **kwargs)\u001b[0m\n\u001b[0;32m 1080\u001b[0m \u001b[39m# Fallback dispatch system (dispatch v1):\u001b[39;00m\n\u001b[0;32m 1081\u001b[0m \u001b[39mtry\u001b[39;00m:\n\u001b[1;32m-> 1082\u001b[0m \u001b[39mreturn\u001b[39;00m dispatch_target(\u001b[39m*\u001b[39;49margs, \u001b[39m*\u001b[39;49m\u001b[39m*\u001b[39;49mkwargs)\n\u001b[0;32m 1083\u001b[0m \u001b[39mexcept\u001b[39;00m (\u001b[39mTypeError\u001b[39;00m, \u001b[39mValueError\u001b[39;00m):\n\u001b[0;32m 1084\u001b[0m \u001b[39m# Note: convert_to_eager_tensor currently raises a ValueError, not a\u001b[39;00m\n\u001b[0;32m 1085\u001b[0m \u001b[39m# TypeError, when given unexpected types. So we need to catch both.\u001b[39;00m\n\u001b[0;32m 1086\u001b[0m result \u001b[39m=\u001b[39m dispatch(op_dispatch_handler, args, kwargs)\n",
"File \u001b[1;32m~\\AppData\\Roaming\\Python\\Python38\\site-packages\\keras\\backend.py:4624\u001b[0m, in \u001b[0;36mrnn\u001b[1;34m(step_function, inputs, initial_states, go_backwards, mask, constants, unroll, input_length, time_major, zero_output_for_mask, return_all_outputs)\u001b[0m\n\u001b[0;32m 4620\u001b[0m input_time_zero \u001b[39m=\u001b[39m tf\u001b[39m.\u001b[39mnest\u001b[39m.\u001b[39mpack_sequence_as(inputs,\n\u001b[0;32m 4621\u001b[0m [inp[\u001b[39m0\u001b[39m] \u001b[39mfor\u001b[39;00m inp \u001b[39min\u001b[39;00m flatted_inputs])\n\u001b[0;32m 4622\u001b[0m \u001b[39m# output_time_zero is used to determine the cell output shape and its dtype.\u001b[39;00m\n\u001b[0;32m 4623\u001b[0m \u001b[39m# the value is discarded.\u001b[39;00m\n\u001b[1;32m-> 4624\u001b[0m output_time_zero, _ \u001b[39m=\u001b[39m step_function(\n\u001b[0;32m 4625\u001b[0m input_time_zero, \u001b[39mtuple\u001b[39;49m(initial_states) \u001b[39m+\u001b[39;49m \u001b[39mtuple\u001b[39;49m(constants))\n\u001b[0;32m 4627\u001b[0m output_ta_size \u001b[39m=\u001b[39m time_steps_t \u001b[39mif\u001b[39;00m return_all_outputs \u001b[39melse\u001b[39;00m \u001b[39m1\u001b[39m\n\u001b[0;32m 4628\u001b[0m output_ta \u001b[39m=\u001b[39m \u001b[39mtuple\u001b[39m(\n\u001b[0;32m 4629\u001b[0m tf\u001b[39m.\u001b[39mTensorArray(\n\u001b[0;32m 4630\u001b[0m dtype\u001b[39m=\u001b[39mout\u001b[39m.\u001b[39mdtype,\n\u001b[1;32m (...)\u001b[0m\n\u001b[0;32m 4633\u001b[0m tensor_array_name\u001b[39m=\u001b[39m\u001b[39m'\u001b[39m\u001b[39moutput_ta_\u001b[39m\u001b[39m%s\u001b[39;00m\u001b[39m'\u001b[39m \u001b[39m%\u001b[39m i)\n\u001b[0;32m 4634\u001b[0m \u001b[39mfor\u001b[39;00m i, out \u001b[39min\u001b[39;00m \u001b[39menumerate\u001b[39m(tf\u001b[39m.\u001b[39mnest\u001b[39m.\u001b[39mflatten(output_time_zero)))\n",
"File \u001b[1;32m~\\AppData\\Roaming\\Python\\Python38\\site-packages\\keras\\layers\\rnn\\lstm.py:878\u001b[0m, in \u001b[0;36mstandard_lstm.<locals>.step\u001b[1;34m(cell_inputs, cell_states)\u001b[0m\n\u001b[0;32m 875\u001b[0m c_tm1 \u001b[39m=\u001b[39m cell_states[\u001b[39m1\u001b[39m] \u001b[39m# previous carry state\u001b[39;00m\n\u001b[0;32m 877\u001b[0m z \u001b[39m=\u001b[39m backend\u001b[39m.\u001b[39mdot(cell_inputs, kernel)\n\u001b[1;32m--> 878\u001b[0m z \u001b[39m+\u001b[39m\u001b[39m=\u001b[39m backend\u001b[39m.\u001b[39;49mdot(h_tm1, recurrent_kernel)\n\u001b[0;32m 879\u001b[0m z \u001b[39m=\u001b[39m backend\u001b[39m.\u001b[39mbias_add(z, bias)\n\u001b[0;32m 881\u001b[0m z0, z1, z2, z3 \u001b[39m=\u001b[39m tf\u001b[39m.\u001b[39msplit(z, \u001b[39m4\u001b[39m, axis\u001b[39m=\u001b[39m\u001b[39m1\u001b[39m)\n",
"File \u001b[1;32m~\\AppData\\Roaming\\Python\\Python38\\site-packages\\tensorflow\\python\\util\\traceback_utils.py:150\u001b[0m, in \u001b[0;36mfilter_traceback.<locals>.error_handler\u001b[1;34m(*args, **kwargs)\u001b[0m\n\u001b[0;32m 148\u001b[0m filtered_tb \u001b[39m=\u001b[39m \u001b[39mNone\u001b[39;00m\n\u001b[0;32m 149\u001b[0m \u001b[39mtry\u001b[39;00m:\n\u001b[1;32m--> 150\u001b[0m \u001b[39mreturn\u001b[39;00m fn(\u001b[39m*\u001b[39;49margs, \u001b[39m*\u001b[39;49m\u001b[39m*\u001b[39;49mkwargs)\n\u001b[0;32m 151\u001b[0m \u001b[39mexcept\u001b[39;00m \u001b[39mException\u001b[39;00m \u001b[39mas\u001b[39;00m e:\n\u001b[0;32m 152\u001b[0m filtered_tb \u001b[39m=\u001b[39m _process_traceback_frames(e\u001b[39m.\u001b[39m__traceback__)\n",
"File \u001b[1;32m~\\AppData\\Roaming\\Python\\Python38\\site-packages\\tensorflow\\python\\util\\dispatch.py:1082\u001b[0m, in \u001b[0;36madd_dispatch_support.<locals>.decorator.<locals>.op_dispatch_handler\u001b[1;34m(*args, **kwargs)\u001b[0m\n\u001b[0;32m 1080\u001b[0m \u001b[39m# Fallback dispatch system (dispatch v1):\u001b[39;00m\n\u001b[0;32m 1081\u001b[0m \u001b[39mtry\u001b[39;00m:\n\u001b[1;32m-> 1082\u001b[0m \u001b[39mreturn\u001b[39;00m dispatch_target(\u001b[39m*\u001b[39;49margs, \u001b[39m*\u001b[39;49m\u001b[39m*\u001b[39;49mkwargs)\n\u001b[0;32m 1083\u001b[0m \u001b[39mexcept\u001b[39;00m (\u001b[39mTypeError\u001b[39;00m, \u001b[39mValueError\u001b[39;00m):\n\u001b[0;32m 1084\u001b[0m \u001b[39m# Note: convert_to_eager_tensor currently raises a ValueError, not a\u001b[39;00m\n\u001b[0;32m 1085\u001b[0m \u001b[39m# TypeError, when given unexpected types. So we need to catch both.\u001b[39;00m\n\u001b[0;32m 1086\u001b[0m result \u001b[39m=\u001b[39m dispatch(op_dispatch_handler, args, kwargs)\n",
"File \u001b[1;32m~\\AppData\\Roaming\\Python\\Python38\\site-packages\\keras\\backend.py:2223\u001b[0m, in \u001b[0;36mdot\u001b[1;34m(x, y)\u001b[0m\n\u001b[0;32m 2221\u001b[0m out \u001b[39m=\u001b[39m tf\u001b[39m.\u001b[39msparse\u001b[39m.\u001b[39msparse_dense_matmul(x, y)\n\u001b[0;32m 2222\u001b[0m \u001b[39melse\u001b[39;00m:\n\u001b[1;32m-> 2223\u001b[0m out \u001b[39m=\u001b[39m tf\u001b[39m.\u001b[39;49mmatmul(x, y)\n\u001b[0;32m 2224\u001b[0m \u001b[39mreturn\u001b[39;00m out\n",
"File \u001b[1;32m~\\AppData\\Roaming\\Python\\Python38\\site-packages\\tensorflow\\python\\util\\traceback_utils.py:150\u001b[0m, in \u001b[0;36mfilter_traceback.<locals>.error_handler\u001b[1;34m(*args, **kwargs)\u001b[0m\n\u001b[0;32m 148\u001b[0m filtered_tb \u001b[39m=\u001b[39m \u001b[39mNone\u001b[39;00m\n\u001b[0;32m 149\u001b[0m \u001b[39mtry\u001b[39;00m:\n\u001b[1;32m--> 150\u001b[0m \u001b[39mreturn\u001b[39;00m fn(\u001b[39m*\u001b[39;49margs, \u001b[39m*\u001b[39;49m\u001b[39m*\u001b[39;49mkwargs)\n\u001b[0;32m 151\u001b[0m \u001b[39mexcept\u001b[39;00m \u001b[39mException\u001b[39;00m \u001b[39mas\u001b[39;00m e:\n\u001b[0;32m 152\u001b[0m filtered_tb \u001b[39m=\u001b[39m _process_traceback_frames(e\u001b[39m.\u001b[39m__traceback__)\n",
"File \u001b[1;32m~\\AppData\\Roaming\\Python\\Python38\\site-packages\\tensorflow\\python\\util\\dispatch.py:1082\u001b[0m, in \u001b[0;36madd_dispatch_support.<locals>.decorator.<locals>.op_dispatch_handler\u001b[1;34m(*args, **kwargs)\u001b[0m\n\u001b[0;32m 1080\u001b[0m \u001b[39m# Fallback dispatch system (dispatch v1):\u001b[39;00m\n\u001b[0;32m 1081\u001b[0m \u001b[39mtry\u001b[39;00m:\n\u001b[1;32m-> 1082\u001b[0m \u001b[39mreturn\u001b[39;00m dispatch_target(\u001b[39m*\u001b[39;49margs, \u001b[39m*\u001b[39;49m\u001b[39m*\u001b[39;49mkwargs)\n\u001b[0;32m 1083\u001b[0m \u001b[39mexcept\u001b[39;00m (\u001b[39mTypeError\u001b[39;00m, \u001b[39mValueError\u001b[39;00m):\n\u001b[0;32m 1084\u001b[0m \u001b[39m# Note: convert_to_eager_tensor currently raises a ValueError, not a\u001b[39;00m\n\u001b[0;32m 1085\u001b[0m \u001b[39m# TypeError, when given unexpected types. So we need to catch both.\u001b[39;00m\n\u001b[0;32m 1086\u001b[0m result \u001b[39m=\u001b[39m dispatch(op_dispatch_handler, args, kwargs)\n",
"File \u001b[1;32m~\\AppData\\Roaming\\Python\\Python38\\site-packages\\tensorflow\\python\\ops\\math_ops.py:3713\u001b[0m, in \u001b[0;36mmatmul\u001b[1;34m(a, b, transpose_a, transpose_b, adjoint_a, adjoint_b, a_is_sparse, b_is_sparse, output_type, name)\u001b[0m\n\u001b[0;32m 3710\u001b[0m \u001b[39mreturn\u001b[39;00m gen_math_ops\u001b[39m.\u001b[39mbatch_mat_mul_v3(\n\u001b[0;32m 3711\u001b[0m a, b, adj_x\u001b[39m=\u001b[39madjoint_a, adj_y\u001b[39m=\u001b[39madjoint_b, Tout\u001b[39m=\u001b[39moutput_type, name\u001b[39m=\u001b[39mname)\n\u001b[0;32m 3712\u001b[0m \u001b[39melse\u001b[39;00m:\n\u001b[1;32m-> 3713\u001b[0m \u001b[39mreturn\u001b[39;00m gen_math_ops\u001b[39m.\u001b[39;49mmat_mul(\n\u001b[0;32m 3714\u001b[0m a, b, transpose_a\u001b[39m=\u001b[39;49mtranspose_a, transpose_b\u001b[39m=\u001b[39;49mtranspose_b, name\u001b[39m=\u001b[39;49mname)\n",
"File \u001b[1;32m~\\AppData\\Roaming\\Python\\Python38\\site-packages\\tensorflow\\python\\ops\\gen_math_ops.py:6033\u001b[0m, in \u001b[0;36mmat_mul\u001b[1;34m(a, b, transpose_a, transpose_b, name)\u001b[0m\n\u001b[0;32m 6031\u001b[0m transpose_b \u001b[39m=\u001b[39m \u001b[39mFalse\u001b[39;00m\n\u001b[0;32m 6032\u001b[0m transpose_b \u001b[39m=\u001b[39m _execute\u001b[39m.\u001b[39mmake_bool(transpose_b, \u001b[39m\"\u001b[39m\u001b[39mtranspose_b\u001b[39m\u001b[39m\"\u001b[39m)\n\u001b[1;32m-> 6033\u001b[0m _, _, _op, _outputs \u001b[39m=\u001b[39m _op_def_library\u001b[39m.\u001b[39;49m_apply_op_helper(\n\u001b[0;32m 6034\u001b[0m \u001b[39m\"\u001b[39;49m\u001b[39mMatMul\u001b[39;49m\u001b[39m\"\u001b[39;49m, a\u001b[39m=\u001b[39;49ma, b\u001b[39m=\u001b[39;49mb, transpose_a\u001b[39m=\u001b[39;49mtranspose_a, transpose_b\u001b[39m=\u001b[39;49mtranspose_b,\n\u001b[0;32m 6035\u001b[0m name\u001b[39m=\u001b[39;49mname)\n\u001b[0;32m 6036\u001b[0m _result \u001b[39m=\u001b[39m _outputs[:]\n\u001b[0;32m 6037\u001b[0m \u001b[39mif\u001b[39;00m _execute\u001b[39m.\u001b[39mmust_record_gradient():\n",
"File \u001b[1;32m~\\AppData\\Roaming\\Python\\Python38\\site-packages\\tensorflow\\python\\framework\\op_def_library.py:797\u001b[0m, in \u001b[0;36m_apply_op_helper\u001b[1;34m(op_type_name, name, **keywords)\u001b[0m\n\u001b[0;32m 792\u001b[0m must_colocate_inputs \u001b[39m=\u001b[39m [val \u001b[39mfor\u001b[39;00m arg, val \u001b[39min\u001b[39;00m \u001b[39mzip\u001b[39m(op_def\u001b[39m.\u001b[39minput_arg, inputs)\n\u001b[0;32m 793\u001b[0m \u001b[39mif\u001b[39;00m arg\u001b[39m.\u001b[39mis_ref]\n\u001b[0;32m 794\u001b[0m \u001b[39mwith\u001b[39;00m _MaybeColocateWith(must_colocate_inputs):\n\u001b[0;32m 795\u001b[0m \u001b[39m# Add Op to graph\u001b[39;00m\n\u001b[0;32m 796\u001b[0m \u001b[39m# pylint: disable=protected-access\u001b[39;00m\n\u001b[1;32m--> 797\u001b[0m op \u001b[39m=\u001b[39m g\u001b[39m.\u001b[39;49m_create_op_internal(op_type_name, inputs, dtypes\u001b[39m=\u001b[39;49m\u001b[39mNone\u001b[39;49;00m,\n\u001b[0;32m 798\u001b[0m name\u001b[39m=\u001b[39;49mscope, input_types\u001b[39m=\u001b[39;49minput_types,\n\u001b[0;32m 799\u001b[0m attrs\u001b[39m=\u001b[39;49mattr_protos, op_def\u001b[39m=\u001b[39;49mop_def)\n\u001b[0;32m 801\u001b[0m \u001b[39m# `outputs` is returned as a separate return value so that the output\u001b[39;00m\n\u001b[0;32m 802\u001b[0m \u001b[39m# tensors can the `op` per se can be decoupled so that the\u001b[39;00m\n\u001b[0;32m 803\u001b[0m \u001b[39m# `op_callbacks` can function properly. See framework/op_callbacks.py\u001b[39;00m\n\u001b[0;32m 804\u001b[0m \u001b[39m# for more details.\u001b[39;00m\n\u001b[0;32m 805\u001b[0m outputs \u001b[39m=\u001b[39m op\u001b[39m.\u001b[39moutputs\n",
"File \u001b[1;32m~\\AppData\\Roaming\\Python\\Python38\\site-packages\\tensorflow\\python\\framework\\func_graph.py:694\u001b[0m, in \u001b[0;36mFuncGraph._create_op_internal\u001b[1;34m(self, op_type, inputs, dtypes, input_types, name, attrs, op_def, compute_device)\u001b[0m\n\u001b[0;32m 692\u001b[0m inp \u001b[39m=\u001b[39m \u001b[39mself\u001b[39m\u001b[39m.\u001b[39mcapture(inp)\n\u001b[0;32m 693\u001b[0m captured_inputs\u001b[39m.\u001b[39mappend(inp)\n\u001b[1;32m--> 694\u001b[0m \u001b[39mreturn\u001b[39;00m \u001b[39msuper\u001b[39;49m(FuncGraph, \u001b[39mself\u001b[39;49m)\u001b[39m.\u001b[39;49m_create_op_internal( \u001b[39m# pylint: disable=protected-access\u001b[39;49;00m\n\u001b[0;32m 695\u001b[0m op_type, captured_inputs, dtypes, input_types, name, attrs, op_def,\n\u001b[0;32m 696\u001b[0m compute_device)\n",
"File \u001b[1;32m~\\AppData\\Roaming\\Python\\Python38\\site-packages\\tensorflow\\python\\framework\\ops.py:3754\u001b[0m, in \u001b[0;36mGraph._create_op_internal\u001b[1;34m(self, op_type, inputs, dtypes, input_types, name, attrs, op_def, compute_device)\u001b[0m\n\u001b[0;32m 3751\u001b[0m \u001b[39m# _create_op_helper mutates the new Operation. `_mutation_lock` ensures a\u001b[39;00m\n\u001b[0;32m 3752\u001b[0m \u001b[39m# Session.run call cannot occur between creating and mutating the op.\u001b[39;00m\n\u001b[0;32m 3753\u001b[0m \u001b[39mwith\u001b[39;00m \u001b[39mself\u001b[39m\u001b[39m.\u001b[39m_mutation_lock():\n\u001b[1;32m-> 3754\u001b[0m ret \u001b[39m=\u001b[39m Operation(\n\u001b[0;32m 3755\u001b[0m node_def,\n\u001b[0;32m 3756\u001b[0m \u001b[39mself\u001b[39;49m,\n\u001b[0;32m 3757\u001b[0m inputs\u001b[39m=\u001b[39;49minputs,\n\u001b[0;32m 3758\u001b[0m output_types\u001b[39m=\u001b[39;49mdtypes,\n\u001b[0;32m 3759\u001b[0m control_inputs\u001b[39m=\u001b[39;49mcontrol_inputs,\n\u001b[0;32m 3760\u001b[0m input_types\u001b[39m=\u001b[39;49minput_types,\n\u001b[0;32m 3761\u001b[0m original_op\u001b[39m=\u001b[39;49m\u001b[39mself\u001b[39;49m\u001b[39m.\u001b[39;49m_default_original_op,\n\u001b[0;32m 3762\u001b[0m op_def\u001b[39m=\u001b[39;49mop_def)\n\u001b[0;32m 3763\u001b[0m \u001b[39mself\u001b[39m\u001b[39m.\u001b[39m_create_op_helper(ret, compute_device\u001b[39m=\u001b[39mcompute_device)\n\u001b[0;32m 3764\u001b[0m \u001b[39mreturn\u001b[39;00m ret\n",
"File \u001b[1;32m~\\AppData\\Roaming\\Python\\Python38\\site-packages\\tensorflow\\python\\framework\\ops.py:2129\u001b[0m, in \u001b[0;36mOperation.__init__\u001b[1;34m(self, node_def, g, inputs, output_types, control_inputs, input_types, original_op, op_def)\u001b[0m\n\u001b[0;32m 2127\u001b[0m \u001b[39mif\u001b[39;00m op_def \u001b[39mis\u001b[39;00m \u001b[39mNone\u001b[39;00m:\n\u001b[0;32m 2128\u001b[0m op_def \u001b[39m=\u001b[39m \u001b[39mself\u001b[39m\u001b[39m.\u001b[39m_graph\u001b[39m.\u001b[39m_get_op_def(node_def\u001b[39m.\u001b[39mop)\n\u001b[1;32m-> 2129\u001b[0m \u001b[39mself\u001b[39m\u001b[39m.\u001b[39m_c_op \u001b[39m=\u001b[39m _create_c_op(\u001b[39mself\u001b[39;49m\u001b[39m.\u001b[39;49m_graph, node_def, inputs,\n\u001b[0;32m 2130\u001b[0m control_input_ops, op_def)\n\u001b[0;32m 2131\u001b[0m name \u001b[39m=\u001b[39m compat\u001b[39m.\u001b[39mas_str(node_def\u001b[39m.\u001b[39mname)\n\u001b[0;32m 2133\u001b[0m \u001b[39mself\u001b[39m\u001b[39m.\u001b[39m_traceback \u001b[39m=\u001b[39m tf_stack\u001b[39m.\u001b[39mextract_stack_for_node(\u001b[39mself\u001b[39m\u001b[39m.\u001b[39m_c_op)\n",
"File \u001b[1;32m~\\AppData\\Roaming\\Python\\Python38\\site-packages\\tensorflow\\python\\util\\traceback_utils.py:150\u001b[0m, in \u001b[0;36mfilter_traceback.<locals>.error_handler\u001b[1;34m(*args, **kwargs)\u001b[0m\n\u001b[0;32m 148\u001b[0m filtered_tb \u001b[39m=\u001b[39m \u001b[39mNone\u001b[39;00m\n\u001b[0;32m 149\u001b[0m \u001b[39mtry\u001b[39;00m:\n\u001b[1;32m--> 150\u001b[0m \u001b[39mreturn\u001b[39;00m fn(\u001b[39m*\u001b[39;49margs, \u001b[39m*\u001b[39;49m\u001b[39m*\u001b[39;49mkwargs)\n\u001b[0;32m 151\u001b[0m \u001b[39mexcept\u001b[39;00m \u001b[39mException\u001b[39;00m \u001b[39mas\u001b[39;00m e:\n\u001b[0;32m 152\u001b[0m filtered_tb \u001b[39m=\u001b[39m _process_traceback_frames(e\u001b[39m.\u001b[39m__traceback__)\n",
"File \u001b[1;32m~\\AppData\\Roaming\\Python\\Python38\\site-packages\\tensorflow\\python\\framework\\ops.py:1960\u001b[0m, in \u001b[0;36m_create_c_op\u001b[1;34m(graph, node_def, inputs, control_inputs, op_def)\u001b[0m\n\u001b[0;32m 1956\u001b[0m pywrap_tf_session\u001b[39m.\u001b[39mTF_SetAttrValueProto(op_desc, compat\u001b[39m.\u001b[39mas_str(name),\n\u001b[0;32m 1957\u001b[0m serialized)\n\u001b[0;32m 1959\u001b[0m \u001b[39mtry\u001b[39;00m:\n\u001b[1;32m-> 1960\u001b[0m c_op \u001b[39m=\u001b[39m pywrap_tf_session\u001b[39m.\u001b[39;49mTF_FinishOperation(op_desc)\n\u001b[0;32m 1961\u001b[0m \u001b[39mexcept\u001b[39;00m errors\u001b[39m.\u001b[39mInvalidArgumentError \u001b[39mas\u001b[39;00m e:\n\u001b[0;32m 1962\u001b[0m \u001b[39m# Convert to ValueError for backwards compatibility.\u001b[39;00m\n\u001b[0;32m 1963\u001b[0m \u001b[39mraise\u001b[39;00m \u001b[39mValueError\u001b[39;00m(e\u001b[39m.\u001b[39mmessage)\n",
"\u001b[1;31mKeyboardInterrupt\u001b[0m: "
]
}
],
"source": [
"AttnLSTM.compile(optimizer=opt, loss='categorical_crossentropy', metrics=['categorical_accuracy'])\n",
"AttnLSTM.fit(X_train, y_train, batch_size=batch_size, epochs=max_epochs, validation_data=(X_val, y_val), callbacks=callbacks)"
]
},
{
"cell_type": "code",
"execution_count": 26,
"id": "b89f67cf",
"metadata": {},
"outputs": [],
"source": [
"# Model map\n",
"models = {\n",
" 'LSTM': lstm, \n",
" 'LSTM_Attention_128HUs': AttnLSTM, \n",
"}"
]
},
{
"cell_type": "markdown",
"id": "a928f612",
"metadata": {},
"source": [
"# 7a. Save Weights"
]
},
{
"cell_type": "code",
"execution_count": 27,
"id": "0a7647ad",
"metadata": {},
"outputs": [],
"source": [
"for model_name, model in models.items():\n",
" save_dir = os.path.join(os.getcwd(), f\"{model_name}.h5\")\n",
" model.save(save_dir)"
]
},
{
"cell_type": "markdown",
"id": "13fecf26",
"metadata": {},
"source": [
"# 7b. Load Weights"
]
},
{
"cell_type": "code",
"execution_count": 28,
"id": "ed0114a6",
"metadata": {},
"outputs": [],
"source": [
"# Run model rebuild before doing this\n",
"for model_name, model in models.items():\n",
" load_dir = os.path.join(os.getcwd(), f\"{model_name}.h5\")\n",
" model.load_weights(load_dir)"
]
},
{
"cell_type": "markdown",
"id": "1b7747c6",
"metadata": {},
"source": [
"# 8. Make Predictions"
]
},
{
"cell_type": "code",
"execution_count": 29,
"id": "2101a592",
"metadata": {},
"outputs": [],
"source": [
"for model in models.values():\n",
" res = model.predict(X_test, verbose=0) "
]
},
{
"cell_type": "markdown",
"id": "6b36c98e",
"metadata": {},
"source": [
"# 9. Evaluations using Confusion Matrix and Accuracy"
]
},
{
"cell_type": "code",
"execution_count": 30,
"id": "ecf242d9",
"metadata": {},
"outputs": [],
"source": [
"eval_results = {}\n",
"eval_results['confusion matrix'] = None\n",
"eval_results['accuracy'] = None\n",
"eval_results['precision'] = None\n",
"eval_results['recall'] = None\n",
"eval_results['f1 score'] = None\n",
"\n",
"confusion_matrices = {}\n",
"classification_accuracies = {} \n",
"precisions = {}\n",
"recalls = {}\n",
"f1_scores = {} "
]
},
{
"cell_type": "markdown",
"id": "74d6778f",
"metadata": {},
"source": [
"## 9a. Confusion Matrices"
]
},
{
"cell_type": "code",
"execution_count": 31,
"id": "fccbb90f",
"metadata": {},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
"LSTM confusion matrix: \n",
"[[[27 5]\n",
" [ 1 12]]\n",
"\n",
" [[26 0]\n",
" [ 5 14]]\n",
"\n",
" [[31 1]\n",
" [ 0 13]]]\n",
"LSTM_Attention_128HUs confusion matrix: \n",
"[[[32 0]\n",
" [ 0 13]]\n",
"\n",
" [[26 0]\n",
" [ 0 19]]\n",
"\n",
" [[32 0]\n",
" [ 0 13]]]\n"
]
}
],
"source": [
"for model_name, model in models.items():\n",
" yhat = model.predict(X_test, verbose=0)\n",
" \n",
" # Get list of classification predictions\n",
" ytrue = np.argmax(y_test, axis=1).tolist()\n",
" yhat = np.argmax(yhat, axis=1).tolist()\n",
" \n",
" # Confusion matrix\n",
" confusion_matrices[model_name] = multilabel_confusion_matrix(ytrue, yhat)\n",
" print(f\"{model_name} confusion matrix: {os.linesep}{confusion_matrices[model_name]}\")\n",
"\n",
"# Collect results \n",
"eval_results['confusion matrix'] = confusion_matrices"
]
},
{
"cell_type": "markdown",
"id": "b76c6dc5",
"metadata": {},
"source": [
"## 9b. Accuracy"
]
},
{
"cell_type": "code",
"execution_count": 41,
"id": "e36146f5",
"metadata": {},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
"LSTM classification accuracy = 86.667%\n",
"LSTM_Attention_128HUs classification accuracy = 100.0%\n"
]
}
],
"source": [
"for model_name, model in models.items():\n",
" yhat = model.predict(X_test, verbose=0)\n",
" \n",
" # Get list of classification predictions\n",
" ytrue = np.argmax(y_test, axis=1).tolist()\n",
" yhat = np.argmax(yhat, axis=1).tolist()\n",
" \n",
" # Model accuracy\n",
" classification_accuracies[model_name] = accuracy_score(ytrue, yhat) \n",
" print(f\"{model_name} classification accuracy = {round(classification_accuracies[model_name]*100,3)}%\")\n",
"\n",
"# Collect results \n",
"eval_results['accuracy'] = classification_accuracies"
]
},
{
"cell_type": "markdown",
"id": "33efa73a",
"metadata": {},
"source": [
"## 9c. Precision, Recall, and F1 Score"
]
},
{
"cell_type": "code",
"execution_count": 43,
"id": "35067c48",
"metadata": {},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
"LSTM weighted average precision = 0.894\n",
"LSTM weighted average recall = 0.867\n",
"LSTM weighted average f1-score = 0.868\n",
"\n",
"LSTM_Attention_128HUs weighted average precision = 1.0\n",
"LSTM_Attention_128HUs weighted average recall = 1.0\n",
"LSTM_Attention_128HUs weighted average f1-score = 1.0\n",
"\n"
]
}
],
"source": [
"for model_name, model in models.items():\n",
" yhat = model.predict(X_test, verbose=0)\n",
" \n",
" # Get list of classification predictions\n",
" ytrue = np.argmax(y_test, axis=1).tolist()\n",
" yhat = np.argmax(yhat, axis=1).tolist()\n",
" \n",
" # Precision, recall, and f1 score\n",
" report = classification_report(ytrue, yhat, target_names=actions, output_dict=True)\n",
" \n",
" precisions[model_name] = report['weighted avg']['precision']\n",
" recalls[model_name] = report['weighted avg']['recall']\n",
" f1_scores[model_name] = report['weighted avg']['f1-score'] \n",
" \n",
" print(f\"{model_name} weighted average precision = {round(precisions[model_name],3)}\")\n",
" print(f\"{model_name} weighted average recall = {round(recalls[model_name],3)}\")\n",
" print(f\"{model_name} weighted average f1-score = {round(f1_scores[model_name],3)}\\n\")\n",
"\n",
"# Collect results \n",
"eval_results['precision'] = precisions\n",
"eval_results['recall'] = recalls\n",
"eval_results['f1 score'] = f1_scores"
]
},
{
"cell_type": "markdown",
"id": "e5d39476",
"metadata": {},
"source": [
"# 10. Choose Model to Test in Real Time"
]
},
{
"cell_type": "code",
"execution_count": 44,
"id": "d72d0605",
"metadata": {},
"outputs": [],
"source": [
"model = AttnLSTM\n",
"model_name = 'AttnLSTM'"
]
},
{
"cell_type": "markdown",
"id": "9f0015ce",
"metadata": {},
"source": [
"# 11. Calculate Joint Angles & Count Reps"
]
},
{
"cell_type": "code",
"execution_count": 45,
"id": "f172932f",
"metadata": {},
"outputs": [],
"source": [
"def calculate_angle(a,b,c):\n",
" \"\"\"\n",
" Computes 3D joint angle inferred by 3 keypoints and their relative positions to one another\n",
" \n",
" \"\"\"\n",
" a = np.array(a) # First\n",
" b = np.array(b) # Mid\n",
" c = np.array(c) # End\n",
" \n",
" radians = np.arctan2(c[1]-b[1], c[0]-b[0]) - np.arctan2(a[1]-b[1], a[0]-b[0])\n",
" angle = np.abs(radians*180.0/np.pi)\n",
" \n",
" if angle >180.0:\n",
" angle = 360-angle\n",
" \n",
" return angle "
]
},
{
"cell_type": "code",
"execution_count": 46,
"id": "26f357fe",
"metadata": {},
"outputs": [],
"source": [
"def get_coordinates(landmarks, mp_pose, side, joint):\n",
" \"\"\"\n",
" Retrieves x and y coordinates of a particular keypoint from the pose estimation model\n",
" \n",
" Args:\n",
" landmarks: processed keypoints from the pose estimation model\n",
" mp_pose: Mediapipe pose estimation model\n",
" side: 'left' or 'right'. Denotes the side of the body of the landmark of interest.\n",
" joint: 'shoulder', 'elbow', 'wrist', 'hip', 'knee', or 'ankle'. Denotes which body joint is associated with the landmark of interest.\n",
" \n",
" \"\"\"\n",
" coord = getattr(mp_pose.PoseLandmark,side.upper()+\"_\"+joint.upper())\n",
" x_coord_val = landmarks[coord.value].x\n",
" y_coord_val = landmarks[coord.value].y\n",
" return [x_coord_val, y_coord_val] "
]
},
{
"cell_type": "code",
"execution_count": 47,
"id": "f11273cf",
"metadata": {},
"outputs": [],
"source": [
"def viz_joint_angle(image, angle, joint):\n",
" \"\"\"\n",
" Displays the joint angle value near the joint within the image frame\n",
" \n",
" \"\"\"\n",
" cv2.putText(image, str(int(angle)), \n",
" tuple(np.multiply(joint, [640, 480]).astype(int)), \n",
" cv2.FONT_HERSHEY_SIMPLEX, 0.5, (255, 255, 255), 2, cv2.LINE_AA\n",
" )\n",
" return"
]
},
{
"cell_type": "code",
"execution_count": 48,
"id": "b64050d3",
"metadata": {},
"outputs": [],
"source": [
"def count_reps(image, current_action, landmarks, mp_pose):\n",
" \"\"\"\n",
" Counts repetitions of each exercise. Global count and stage (i.e., state) variables are updated within this function.\n",
" \n",
" \"\"\"\n",
"\n",
" global curl_counter, press_counter, squat_counter, curl_stage, press_stage, squat_stage\n",
" \n",
" if current_action == 'curl':\n",
" # Get coords\n",
" shoulder = get_coordinates(landmarks, mp_pose, 'left', 'shoulder')\n",
" elbow = get_coordinates(landmarks, mp_pose, 'left', 'elbow')\n",
" wrist = get_coordinates(landmarks, mp_pose, 'left', 'wrist')\n",
" \n",
" # calculate elbow angle\n",
" angle = calculate_angle(shoulder, elbow, wrist)\n",
" \n",
" # curl counter logic\n",
" if angle < 30:\n",
" curl_stage = \"up\" \n",
" if angle > 140 and curl_stage =='up':\n",
" curl_stage=\"down\" \n",
" curl_counter +=1\n",
" press_stage = None\n",
" squat_stage = None\n",
" \n",
" # Viz joint angle\n",
" viz_joint_angle(image, angle, elbow)\n",
" \n",
" elif current_action == 'press':\n",
" \n",
" # Get coords\n",
" shoulder = get_coordinates(landmarks, mp_pose, 'left', 'shoulder')\n",
" elbow = get_coordinates(landmarks, mp_pose, 'left', 'elbow')\n",
" wrist = get_coordinates(landmarks, mp_pose, 'left', 'wrist')\n",
"\n",
" # Calculate elbow angle\n",
" elbow_angle = calculate_angle(shoulder, elbow, wrist)\n",
" \n",
" # Compute distances between joints\n",
" shoulder2elbow_dist = abs(math.dist(shoulder,elbow))\n",
" shoulder2wrist_dist = abs(math.dist(shoulder,wrist))\n",
" \n",
" # Press counter logic\n",
" if (elbow_angle > 130) and (shoulder2elbow_dist < shoulder2wrist_dist):\n",
" press_stage = \"up\"\n",
" if (elbow_angle < 50) and (shoulder2elbow_dist > shoulder2wrist_dist) and (press_stage =='up'):\n",
" press_stage='down'\n",
" press_counter += 1\n",
" curl_stage = None\n",
" squat_stage = None\n",
" \n",
" # Viz joint angle\n",
" viz_joint_angle(image, elbow_angle, elbow)\n",
" \n",
" elif current_action == 'squat':\n",
" # Get coords\n",
" # left side\n",
" left_shoulder = get_coordinates(landmarks, mp_pose, 'left', 'shoulder')\n",
" left_hip = get_coordinates(landmarks, mp_pose, 'left', 'hip')\n",
" left_knee = get_coordinates(landmarks, mp_pose, 'left', 'knee')\n",
" left_ankle = get_coordinates(landmarks, mp_pose, 'left', 'ankle')\n",
" # right side\n",
" right_shoulder = get_coordinates(landmarks, mp_pose, 'right', 'shoulder')\n",
" right_hip = get_coordinates(landmarks, mp_pose, 'right', 'hip')\n",
" right_knee = get_coordinates(landmarks, mp_pose, 'right', 'knee')\n",
" right_ankle = get_coordinates(landmarks, mp_pose, 'right', 'ankle')\n",
" \n",
" # Calculate knee angles\n",
" left_knee_angle = calculate_angle(left_hip, left_knee, left_ankle)\n",
" right_knee_angle = calculate_angle(right_hip, right_knee, right_ankle)\n",
" \n",
" # Calculate hip angles\n",
" left_hip_angle = calculate_angle(left_shoulder, left_hip, left_knee)\n",
" right_hip_angle = calculate_angle(right_shoulder, right_hip, right_knee)\n",
" \n",
" # Squat counter logic\n",
" thr = 165\n",
" if (left_knee_angle < thr) and (right_knee_angle < thr) and (left_hip_angle < thr) and (right_hip_angle < thr):\n",
" squat_stage = \"down\"\n",
" if (left_knee_angle > thr) and (right_knee_angle > thr) and (left_hip_angle > thr) and (right_hip_angle > thr) and (squat_stage =='down'):\n",
" squat_stage='up'\n",
" squat_counter += 1\n",
" curl_stage = None\n",
" press_stage = None\n",
" \n",
" # Viz joint angles\n",
" viz_joint_angle(image, left_knee_angle, left_knee)\n",
" viz_joint_angle(image, left_hip_angle, left_hip)\n",
" \n",
" else:\n",
" pass"
]
},
{
"cell_type": "markdown",
"id": "a5116ef6",
"metadata": {},
"source": [
"# 12. Test in Real Time"
]
},
{
"cell_type": "code",
"execution_count": 49,
"id": "4775b75e",
"metadata": {},
"outputs": [],
"source": [
"def prob_viz(res, actions, input_frame, colors):\n",
" \"\"\"\n",
" This function displays the model prediction probability distribution over the set of exercise classes\n",
" as a horizontal bar graph\n",
" \n",
" \"\"\"\n",
" output_frame = input_frame.copy()\n",
" for num, prob in enumerate(res): \n",
" cv2.rectangle(output_frame, (0,60+num*40), (int(prob*100), 90+num*40), colors[num], -1)\n",
" cv2.putText(output_frame, actions[num], (0, 85+num*40), cv2.FONT_HERSHEY_SIMPLEX, 1, (255,255,255), 2, cv2.LINE_AA)\n",
" \n",
" return output_frame"
]
},
{
"cell_type": "code",
"execution_count": 50,
"id": "6332bf1a",
"metadata": {},
"outputs": [],
"source": [
"# 1. New detection variables\n",
"sequence = []\n",
"predictions = []\n",
"res = []\n",
"threshold = 0.5 # minimum confidence to classify as an action/exercise\n",
"current_action = ''\n",
"\n",
"# Rep counter logic variables\n",
"curl_counter = 0\n",
"press_counter = 0\n",
"squat_counter = 0\n",
"curl_stage = None\n",
"press_stage = None\n",
"squat_stage = None\n",
"\n",
"# Camera object\n",
"cap = cv2.VideoCapture(0)\n",
"\n",
"# Video writer object that saves a video of the real time test\n",
"fourcc = cv2.VideoWriter_fourcc('M','J','P','G') # video compression format\n",
"HEIGHT = int(cap.get(cv2.CAP_PROP_FRAME_HEIGHT)) # webcam video frame height\n",
"WIDTH = int(cap.get(cv2.CAP_PROP_FRAME_WIDTH)) # webcam video frame width\n",
"FPS = int(cap.get(cv2.CAP_PROP_FPS)) # webcam video fram rate \n",
"\n",
"video_name = os.path.join(os.getcwd(),f\"{model_name}_real_time_test.avi\")\n",
"out = cv2.VideoWriter(video_name, cv2.VideoWriter_fourcc(*\"MJPG\"), FPS, (WIDTH,HEIGHT))\n",
"\n",
"# Set mediapipe model \n",
"with mp_pose.Pose(min_detection_confidence=0.5, min_tracking_confidence=0.5) as pose:\n",
" while cap.isOpened():\n",
"\n",
" # Read feed\n",
" ret, frame = cap.read()\n",
"\n",
" # Make detection\n",
" image, results = mediapipe_detection(frame, pose)\n",
" \n",
" # Draw landmarks\n",
" draw_landmarks(image, results)\n",
" \n",
" # 2. Prediction logic\n",
" keypoints = extract_keypoints(results) \n",
" sequence.append(keypoints) \n",
" sequence = sequence[-sequence_length:]\n",
" \n",
" if len(sequence) == sequence_length:\n",
" res = model.predict(np.expand_dims(sequence, axis=0), verbose=0)[0] \n",
" predictions.append(np.argmax(res))\n",
" current_action = actions[np.argmax(res)]\n",
" confidence = np.max(res)\n",
" \n",
" #3. Viz logic\n",
" # Erase current action variable if no probability is above threshold\n",
" if confidence < threshold:\n",
" current_action = ''\n",
"\n",
" # Viz probabilities\n",
" image = prob_viz(res, actions, image, colors)\n",
" \n",
" # Count reps\n",
" try:\n",
" landmarks = results.pose_landmarks.landmark\n",
" count_reps(\n",
" image, current_action, landmarks, mp_pose)\n",
" except:\n",
" pass\n",
"\n",
" # Display graphical information\n",
" cv2.rectangle(image, (0,0), (640, 40), colors[np.argmax(res)], -1)\n",
" cv2.putText(image, 'curl ' + str(curl_counter), (3,30), \n",
" cv2.FONT_HERSHEY_SIMPLEX, 1, (255, 255, 255), 2, cv2.LINE_AA)\n",
" cv2.putText(image, 'press ' + str(press_counter), (240,30), \n",
" cv2.FONT_HERSHEY_SIMPLEX, 1, (255, 255, 255), 2, cv2.LINE_AA)\n",
" cv2.putText(image, 'squat ' + str(squat_counter), (490,30), \n",
" cv2.FONT_HERSHEY_SIMPLEX, 1, (255, 255, 255), 2, cv2.LINE_AA)\n",
" \n",
" # Show to screen\n",
" cv2.imshow('OpenCV Feed', image)\n",
" \n",
" # Write to video file\n",
" if ret == True:\n",
" out.write(image)\n",
"\n",
" # Break gracefully\n",
" if cv2.waitKey(10) & 0xFF == ord('q'):\n",
" break\n",
" cap.release()\n",
" out.release()\n",
" cv2.destroyAllWindows()"
]
},
{
"cell_type": "code",
"execution_count": 51,
"id": "af9980a1",
"metadata": {},
"outputs": [],
"source": [
"cap.release()\n",
"out.release()\n",
"cv2.destroyAllWindows()"
]
}
],
"metadata": {
"kernelspec": {
"display_name": "Python 3.8.13 ('AItrainer')",
"language": "python",
"name": "python3"
},
"language_info": {
"codemirror_mode": {
"name": "ipython",
"version": 3
},
"file_extension": ".py",
"mimetype": "text/x-python",
"name": "python",
"nbconvert_exporter": "python",
"pygments_lexer": "ipython3",
"version": "3.8.13"
},
"vscode": {
"interpreter": {
"hash": "80aa1d3f3a8cfb37a38c47373cc49a39149184c5fa770d709389b1b8782c1d85"
}
}
},
"nbformat": 4,
"nbformat_minor": 5
}