chore: update
Browse files
app.py
CHANGED
|
@@ -1,5 +1,6 @@
|
|
| 1 |
-
import
|
| 2 |
import shutil
|
|
|
|
| 3 |
from pathlib import Path
|
| 4 |
from time import time
|
| 5 |
from typing import List, Tuple, Union
|
|
@@ -7,33 +8,35 @@ from typing import List, Tuple, Union
|
|
| 7 |
import gradio as gr
|
| 8 |
import numpy as np
|
| 9 |
import pandas as pd
|
| 10 |
-
from
|
| 11 |
-
from
|
| 12 |
-
from sklearn.model_selection import train_test_split
|
| 13 |
|
| 14 |
-
from concrete.ml.common.serialization.loaders import load
|
| 15 |
from concrete.ml.deployment import FHEModelClient, FHEModelDev, FHEModelServer
|
| 16 |
from concrete.ml.sklearn import XGBClassifier as ConcreteXGBoostClassifier
|
| 17 |
|
| 18 |
-
|
| 19 |
-
|
| 20 |
-
import subprocess
|
| 21 |
-
|
| 22 |
-
from preprocessing import ( # pylint: disable=wrong-import-position, no-name-in-module
|
| 23 |
-
map_prediction,
|
| 24 |
-
pretty_print,
|
| 25 |
-
)
|
| 26 |
-
from symptoms_categories import SYMPTOMS_LIST
|
| 27 |
|
| 28 |
-
|
| 29 |
-
# This repository's directory
|
| 30 |
REPO_DIR = Path(__file__).parent
|
|
|
|
|
|
|
|
|
|
|
|
|
| 31 |
|
| 32 |
-
print(f"{REPO_DIR=}")
|
| 33 |
# subprocess.Popen(["uvicorn", "server:app"], cwd=REPO_DIR)
|
| 34 |
# time.sleep(3)
|
| 35 |
|
| 36 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 37 |
def load_data():
|
| 38 |
# Load data
|
| 39 |
df_train = pd.read_csv("./data/Training_preprocessed.csv")
|
|
@@ -61,75 +64,8 @@ def load_model(X_train, y_train):
|
|
| 61 |
return classifier, circuit
|
| 62 |
|
| 63 |
|
| 64 |
-
def key_gen():
|
| 65 |
-
|
| 66 |
-
# Key serialization
|
| 67 |
-
user_id = np.random.randint(0, 2**32)
|
| 68 |
-
|
| 69 |
-
client = FHEModelClient(path_dir=path_to_model, key_dir=f".fhe_keys/{user_id}")
|
| 70 |
-
client.load()
|
| 71 |
-
|
| 72 |
-
# The client first need to create the private and evaluation keys.
|
| 73 |
-
|
| 74 |
-
client.generate_private_and_evaluation_keys()
|
| 75 |
-
|
| 76 |
-
# Get the serialized evaluation keys
|
| 77 |
-
serialized_evaluation_keys = client.get_serialized_evaluation_keys()
|
| 78 |
-
assert isinstance(serialized_evaluation_keys, bytes)
|
| 79 |
-
|
| 80 |
-
np.save(f".fhe_keys/{user_id}/eval_key.npy", serialized_evaluation_keys)
|
| 81 |
-
|
| 82 |
-
serialized_evaluation_keys_shorten = list(serialized_evaluation_keys)[:200]
|
| 83 |
-
serialized_evaluation_keys_shorten_hex = "".join(
|
| 84 |
-
f"{i:02x}" for i in serialized_evaluation_keys_shorten
|
| 85 |
-
)
|
| 86 |
-
# Evaluation keys can be quite large files but only have to be shared once with the server.
|
| 87 |
-
|
| 88 |
-
# Check the size of the evaluation keys (in MB)
|
| 89 |
-
return [
|
| 90 |
-
serialized_evaluation_keys_shorten_hex,
|
| 91 |
-
user_id,
|
| 92 |
-
f"{len(serialized_evaluation_keys) / (10**6):.2f} MB",
|
| 93 |
-
]
|
| 94 |
-
|
| 95 |
-
|
| 96 |
-
def encode_quantize_encrypt(user_symptoms, user_id):
|
| 97 |
-
# check if the key has been generated
|
| 98 |
-
client = FHEModelClient(path_dir=path_to_model, key_dir=f".fhe_keys/{user_id}")
|
| 99 |
-
client.load()
|
| 100 |
-
|
| 101 |
-
user_symptoms = np.fromstring(user_symptoms[2:-2], dtype=int, sep=".").reshape(1, -1)
|
| 102 |
-
|
| 103 |
-
quant_user_symptoms = client.model.quantize_input(user_symptoms)
|
| 104 |
-
encrypted_quantized_user_symptoms = client.quantize_encrypt_serialize(user_symptoms)
|
| 105 |
-
|
| 106 |
-
# print(client.model.predict(vect_x, fhe="simulate"), client.model.predict(vect_x, fhe="execute"))
|
| 107 |
-
# pred_s = client.model.fhe_circuit.simulate(quant_vect)
|
| 108 |
-
# pred_fhe = client.model.fhe_circuit.encrypt_run_decrypt(quant_vect) #
|
| 109 |
-
# non alpha -> \X1124, base64 ou en exa
|
| 110 |
-
|
| 111 |
-
# Compute size
|
| 112 |
-
|
| 113 |
-
np.save(f".fhe_keys/{user_id}/encrypted_quant_vect.npy", encrypted_quantized_user_symptoms)
|
| 114 |
-
|
| 115 |
-
encrypted_quantized_encoding_shorten = list(encrypted_quantized_user_symptoms)[:200]
|
| 116 |
-
encrypted_quantized_encoding_shorten_hex = "".join(
|
| 117 |
-
f"{i:02x}" for i in encrypted_quantized_encoding_shorten
|
| 118 |
-
)
|
| 119 |
-
|
| 120 |
-
return user_symptoms, quant_user_symptoms, encrypted_quantized_encoding_shorten_hex
|
| 121 |
-
|
| 122 |
-
|
| 123 |
-
def decrypt_prediction(encrypted_quantized_vect, user_id):
|
| 124 |
-
fhe_api = FHEModelClient(path_dir=path_to_model, key_dir=f".fhe_keys/{user_id}")
|
| 125 |
-
fhe_api.load()
|
| 126 |
-
fhe_api.generate_private_and_evaluation_keys(force=False)
|
| 127 |
-
predictions = fhe_api.deserialize_decrypt_dequantize(encrypted_quantized_vect)
|
| 128 |
-
return predictions
|
| 129 |
-
|
| 130 |
-
|
| 131 |
def get_user_vect_symptoms_from_checkboxgroup(*user_symptoms) -> np.array:
|
| 132 |
-
symptoms_vector = {key: 0 for key in
|
| 133 |
|
| 134 |
for symptom_box in user_symptoms:
|
| 135 |
for pretty_symptom in symptom_box:
|
|
@@ -148,7 +84,7 @@ def get_user_vect_symptoms_from_checkboxgroup(*user_symptoms) -> np.array:
|
|
| 148 |
return user_symptoms_vect
|
| 149 |
|
| 150 |
|
| 151 |
-
def
|
| 152 |
|
| 153 |
user_symptom_vector = df_test[df_test["prognosis"] == disease].iloc[0].values
|
| 154 |
|
|
@@ -165,45 +101,40 @@ def get_user_symptoms_from_default_disease(disease):
|
|
| 165 |
return pretty_print(columns_with_1)
|
| 166 |
|
| 167 |
|
| 168 |
-
def
|
| 169 |
-
|
| 170 |
-
|
| 171 |
-
|
| 172 |
-
|
| 173 |
-
|
| 174 |
-
|
| 175 |
-
|
| 176 |
-
)
|
| 177 |
-
|
| 178 |
-
|
| 179 |
-
|
| 180 |
-
|
| 181 |
-
|
| 182 |
-
|
| 183 |
-
|
| 184 |
-
if not any(lst for lst in selected_symptoms if lst) and (
|
| 185 |
-
selected_default_disease is None
|
| 186 |
-
or (selected_default_disease is not None and len(selected_default_disease) < 1)
|
| 187 |
):
|
| 188 |
return {
|
| 189 |
-
|
| 190 |
-
error_box: gr.update(
|
| 191 |
visible=True, value="Enter a default disease or select your own symptoms"
|
| 192 |
),
|
| 193 |
}
|
| 194 |
# Case 1: The user has checked his own symptoms
|
| 195 |
if any(lst for lst in selected_symptoms if lst):
|
| 196 |
return {
|
|
|
|
| 197 |
user_vector_textbox: get_user_vect_symptoms_from_checkboxgroup(*selected_symptoms),
|
| 198 |
}
|
| 199 |
|
| 200 |
# Case 2: The user has selected a default disease
|
| 201 |
if selected_default_disease is not None and len(selected_default_disease) > 0:
|
| 202 |
return {
|
| 203 |
-
user_vector_textbox:
|
| 204 |
-
|
| 205 |
-
),
|
| 206 |
-
error_box: gr.update(visible=False),
|
| 207 |
**{
|
| 208 |
box: get_user_symptoms_from_default_disease(selected_default_disease)
|
| 209 |
for box in check_boxes
|
|
@@ -211,24 +142,166 @@ def get_user_symptoms_vector_btn(selected_default_disease, *selected_symptoms):
|
|
| 211 |
}
|
| 212 |
|
| 213 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 214 |
def clear_all_btn():
|
| 215 |
return {
|
|
|
|
| 216 |
user_id_textbox: None,
|
| 217 |
eval_key_textbox: None,
|
| 218 |
-
|
| 219 |
user_vector_textbox: None,
|
| 220 |
-
|
| 221 |
-
|
|
|
|
|
|
|
|
|
|
| 222 |
**{box: None for box in check_boxes},
|
| 223 |
}
|
| 224 |
|
| 225 |
|
| 226 |
if __name__ == "__main__":
|
| 227 |
print("Starting demo ...")
|
|
|
|
| 228 |
|
| 229 |
(df_train, X_train, X_test), (df_test, y_train, y_test) = load_data()
|
| 230 |
|
| 231 |
-
|
| 232 |
|
| 233 |
# Load the model
|
| 234 |
with open("ConcreteXGBoostClassifier.pkl", "r", encoding="utf-8") as file:
|
|
@@ -285,6 +358,8 @@ if __name__ == "__main__":
|
|
| 285 |
)
|
| 286 |
check_boxes.append(check_box)
|
| 287 |
|
|
|
|
|
|
|
| 288 |
# User symptom vector
|
| 289 |
with gr.Row():
|
| 290 |
user_vector_textbox = gr.Textbox(
|
|
@@ -292,7 +367,6 @@ if __name__ == "__main__":
|
|
| 292 |
interactive=False,
|
| 293 |
max_lines=100,
|
| 294 |
)
|
| 295 |
-
error_box = gr.Textbox(label="Error", visible=False)
|
| 296 |
|
| 297 |
with gr.Row():
|
| 298 |
# Submit botton
|
|
@@ -300,20 +374,22 @@ if __name__ == "__main__":
|
|
| 300 |
submit_button = gr.Button("Submit")
|
| 301 |
# Clear botton
|
| 302 |
with gr.Column():
|
| 303 |
-
clear_button = gr.Button("Clear"
|
| 304 |
|
| 305 |
# Click submit botton
|
| 306 |
|
| 307 |
submit_button.click(
|
| 308 |
-
fn=
|
| 309 |
inputs=[box_default, *check_boxes],
|
| 310 |
-
outputs=[user_vector_textbox,
|
| 311 |
)
|
| 312 |
|
| 313 |
gr.Markdown("# Step 2: Generate the keys")
|
| 314 |
gr.Markdown("Client side")
|
| 315 |
|
| 316 |
-
|
|
|
|
|
|
|
| 317 |
|
| 318 |
with gr.Row():
|
| 319 |
# User ID
|
|
@@ -338,25 +414,18 @@ if __name__ == "__main__":
|
|
| 338 |
interactive=False,
|
| 339 |
)
|
| 340 |
|
| 341 |
-
|
| 342 |
-
|
| 343 |
-
|
| 344 |
-
|
| 345 |
-
outputs=[
|
| 346 |
-
user_id_textbox,
|
| 347 |
-
user_vector_textbox,
|
| 348 |
-
eval_key_textbox,
|
| 349 |
-
eval_key_len_textbox,
|
| 350 |
-
box_default,
|
| 351 |
-
error_box,
|
| 352 |
-
*check_boxes,
|
| 353 |
-
],
|
| 354 |
)
|
| 355 |
|
| 356 |
gr.Markdown("# Step 3: Encode the message with the private key")
|
| 357 |
gr.Markdown("Client side")
|
| 358 |
|
| 359 |
-
|
|
|
|
|
|
|
| 360 |
|
| 361 |
with gr.Row():
|
| 362 |
|
|
@@ -377,10 +446,10 @@ if __name__ == "__main__":
|
|
| 377 |
label="Encrypted vector:", max_lines=4, interactive=False
|
| 378 |
)
|
| 379 |
|
| 380 |
-
|
| 381 |
-
|
| 382 |
inputs=[user_vector_textbox, user_id_textbox],
|
| 383 |
-
outputs=[vect_textbox, quant_vect_textbox, encrypted_vect_textbox],
|
| 384 |
)
|
| 385 |
|
| 386 |
gr.Markdown("# Step 4: Run the FHE evaluation")
|
|
@@ -396,10 +465,27 @@ if __name__ == "__main__":
|
|
| 396 |
label="Encrypted vector:", max_lines=4, interactive=False
|
| 397 |
)
|
| 398 |
|
| 399 |
-
decrypt_target_botton.click(
|
| 400 |
-
|
| 401 |
-
|
| 402 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 403 |
)
|
| 404 |
|
| 405 |
demo.launch()
|
|
|
|
| 1 |
+
import os
|
| 2 |
import shutil
|
| 3 |
+
import subprocess
|
| 4 |
from pathlib import Path
|
| 5 |
from time import time
|
| 6 |
from typing import List, Tuple, Union
|
|
|
|
| 8 |
import gradio as gr
|
| 9 |
import numpy as np
|
| 10 |
import pandas as pd
|
| 11 |
+
from preprocessing import pretty_print
|
| 12 |
+
from symptoms_categories import SYMPTOMS_LIST
|
|
|
|
| 13 |
|
| 14 |
+
from concrete.ml.common.serialization.loaders import load
|
| 15 |
from concrete.ml.deployment import FHEModelClient, FHEModelDev, FHEModelServer
|
| 16 |
from concrete.ml.sklearn import XGBClassifier as ConcreteXGBoostClassifier
|
| 17 |
|
| 18 |
+
INPUT_BROWSER_LIMIT = 635
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 19 |
|
| 20 |
+
# This repository's main necessary folders
|
|
|
|
| 21 |
REPO_DIR = Path(__file__).parent
|
| 22 |
+
MODEL_PATH = REPO_DIR / "client_folder"
|
| 23 |
+
KEYS_PATH = REPO_DIR / ".fhe_keys"
|
| 24 |
+
CLIENT_PATH = MODEL_PATH / "client.zip"
|
| 25 |
+
SERVER_PATH = MODEL_PATH / "server.zip"
|
| 26 |
|
|
|
|
| 27 |
# subprocess.Popen(["uvicorn", "server:app"], cwd=REPO_DIR)
|
| 28 |
# time.sleep(3)
|
| 29 |
|
| 30 |
|
| 31 |
+
def clean_directory():
|
| 32 |
+
target_dir = ".fhe_keys"
|
| 33 |
+
if os.path.exists(target_dir) and os.path.isdir(target_dir):
|
| 34 |
+
shutil.rmtree(target_dir)
|
| 35 |
+
print("The .fhe_keys directory and its contents have been successfully removed.")
|
| 36 |
+
else:
|
| 37 |
+
print("The .keys directory does not exist.")
|
| 38 |
+
|
| 39 |
+
|
| 40 |
def load_data():
|
| 41 |
# Load data
|
| 42 |
df_train = pd.read_csv("./data/Training_preprocessed.csv")
|
|
|
|
| 64 |
return classifier, circuit
|
| 65 |
|
| 66 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 67 |
def get_user_vect_symptoms_from_checkboxgroup(*user_symptoms) -> np.array:
|
| 68 |
+
symptoms_vector = {key: 0 for key in VALID_COLUMNS}
|
| 69 |
|
| 70 |
for symptom_box in user_symptoms:
|
| 71 |
for pretty_symptom in symptom_box:
|
|
|
|
| 84 |
return user_symptoms_vect
|
| 85 |
|
| 86 |
|
| 87 |
+
def get_user_vector_from_default_disease(disease):
|
| 88 |
|
| 89 |
user_symptom_vector = df_test[df_test["prognosis"] == disease].iloc[0].values
|
| 90 |
|
|
|
|
| 101 |
return pretty_print(columns_with_1)
|
| 102 |
|
| 103 |
|
| 104 |
+
def get_user_symptoms_vector_fn(selected_default_disease, *selected_symptoms):
|
| 105 |
+
|
| 106 |
+
# Display an error box, if:
|
| 107 |
+
# 1. The user has already selected a default disease and added more symptoms, or
|
| 108 |
+
# 2. The the user has not selected a default disease or symptoms
|
| 109 |
+
if (
|
| 110 |
+
any(lst for lst in selected_symptoms if lst)
|
| 111 |
+
and (selected_default_disease is not None and len(selected_default_disease) > 0)
|
| 112 |
+
and set(pretty_print(selected_symptoms))
|
| 113 |
+
- set(get_user_symptoms_from_default_disease(selected_default_disease))
|
| 114 |
+
) or (
|
| 115 |
+
not any(lst for lst in selected_symptoms if lst)
|
| 116 |
+
and (
|
| 117 |
+
selected_default_disease is None
|
| 118 |
+
or (selected_default_disease is not None and len(selected_default_disease) < 1)
|
| 119 |
+
)
|
|
|
|
|
|
|
|
|
|
| 120 |
):
|
| 121 |
return {
|
| 122 |
+
error_box_1: gr.update(
|
|
|
|
| 123 |
visible=True, value="Enter a default disease or select your own symptoms"
|
| 124 |
),
|
| 125 |
}
|
| 126 |
# Case 1: The user has checked his own symptoms
|
| 127 |
if any(lst for lst in selected_symptoms if lst):
|
| 128 |
return {
|
| 129 |
+
error_box_1: gr.update(visible=False),
|
| 130 |
user_vector_textbox: get_user_vect_symptoms_from_checkboxgroup(*selected_symptoms),
|
| 131 |
}
|
| 132 |
|
| 133 |
# Case 2: The user has selected a default disease
|
| 134 |
if selected_default_disease is not None and len(selected_default_disease) > 0:
|
| 135 |
return {
|
| 136 |
+
user_vector_textbox: get_user_vector_from_default_disease(selected_default_disease),
|
| 137 |
+
error_box_1: gr.update(visible=False),
|
|
|
|
|
|
|
| 138 |
**{
|
| 139 |
box: get_user_symptoms_from_default_disease(selected_default_disease)
|
| 140 |
for box in check_boxes
|
|
|
|
| 142 |
}
|
| 143 |
|
| 144 |
|
| 145 |
+
def key_gen_fn(user_symptoms):
|
| 146 |
+
|
| 147 |
+
print("Cleaning directory ...")
|
| 148 |
+
clean_directory()
|
| 149 |
+
|
| 150 |
+
if user_symptoms is None or (user_symptoms is not None and len(user_symptoms) < 1):
|
| 151 |
+
print("Please submit your symptoms first")
|
| 152 |
+
return {
|
| 153 |
+
error_box_2: gr.update(visible=True, value="Please submit your symptoms first"),
|
| 154 |
+
}
|
| 155 |
+
|
| 156 |
+
# Key serialization
|
| 157 |
+
user_id = np.random.randint(0, 2**32)
|
| 158 |
+
|
| 159 |
+
client = FHEModelClient(path_dir=MODEL_PATH, key_dir=KEYS_PATH / f"{user_id}")
|
| 160 |
+
client.load()
|
| 161 |
+
|
| 162 |
+
# The client first need to create the private and evaluation keys.
|
| 163 |
+
|
| 164 |
+
client.generate_private_and_evaluation_keys()
|
| 165 |
+
|
| 166 |
+
# Get the serialized evaluation keys
|
| 167 |
+
serialized_evaluation_keys = client.get_serialized_evaluation_keys()
|
| 168 |
+
assert isinstance(serialized_evaluation_keys, bytes)
|
| 169 |
+
|
| 170 |
+
# np.save(f".fhe_keys/{user_id}/eval_key.npy", serialized_evaluation_keys)
|
| 171 |
+
evaluation_key_path = KEYS_PATH / f"{user_id}/evaluation_key"
|
| 172 |
+
with evaluation_key_path.open("wb") as evaluation_key_file:
|
| 173 |
+
evaluation_key_file.write(serialized_evaluation_keys)
|
| 174 |
+
|
| 175 |
+
serialized_evaluation_keys_shorten_hex = serialized_evaluation_keys.hex()[:INPUT_BROWSER_LIMIT]
|
| 176 |
+
|
| 177 |
+
return {
|
| 178 |
+
error_box_2: gr.update(visible=False),
|
| 179 |
+
eval_key_textbox: serialized_evaluation_keys_shorten_hex,
|
| 180 |
+
user_id_textbox: user_id,
|
| 181 |
+
eval_key_len_textbox: f"{len(serialized_evaluation_keys) / (10**6):.2f} MB",
|
| 182 |
+
}
|
| 183 |
+
|
| 184 |
+
|
| 185 |
+
def encrypt_fn(user_symptoms, user_id):
|
| 186 |
+
|
| 187 |
+
if not user_symptoms or not user_symptoms:
|
| 188 |
+
return {
|
| 189 |
+
error_box_3: gr.update(
|
| 190 |
+
visible=True, value="Please ensure that the evaluation key has been generated!"
|
| 191 |
+
)
|
| 192 |
+
}
|
| 193 |
+
|
| 194 |
+
# Retrieve the client API
|
| 195 |
+
|
| 196 |
+
client = FHEModelClient(path_dir=MODEL_PATH, key_dir=KEYS_PATH / f"{user_id}")
|
| 197 |
+
client.load()
|
| 198 |
+
|
| 199 |
+
user_symptoms = np.fromstring(user_symptoms[2:-2], dtype=int, sep=".").reshape(1, -1)
|
| 200 |
+
|
| 201 |
+
quant_user_symptoms = client.model.quantize_input(user_symptoms)
|
| 202 |
+
encrypted_quantized_user_symptoms = client.quantize_encrypt_serialize(user_symptoms)
|
| 203 |
+
|
| 204 |
+
encrypted_input_path = KEYS_PATH / f"{user_id}/encrypted_symptoms"
|
| 205 |
+
|
| 206 |
+
with encrypted_input_path.open("wb") as f:
|
| 207 |
+
f.write(encrypted_quantized_user_symptoms)
|
| 208 |
+
|
| 209 |
+
# print(client.model.predict(vect_x, fhe="simulate"), client.model.predict(vect_x, fhe="execute"))
|
| 210 |
+
# pred_s = client.model.fhe_circuit.simulate(quant_vect)
|
| 211 |
+
# pred_fhe = client.model.fhe_circuit.encrypt_run_decrypt(quant_vect) #
|
| 212 |
+
# non alpha -> \X1124, base64 ou en exa
|
| 213 |
+
|
| 214 |
+
# Compute size
|
| 215 |
+
|
| 216 |
+
# np.save(f".fhe_keys/{user_id}/encrypted_quant_vect.npy", encrypted_quantized_user_symptoms)
|
| 217 |
+
|
| 218 |
+
encrypted_quantized_user_symptoms_shorten_hex = encrypted_quantized_user_symptoms.hex()[
|
| 219 |
+
:INPUT_BROWSER_LIMIT
|
| 220 |
+
]
|
| 221 |
+
|
| 222 |
+
return {
|
| 223 |
+
error_box_3: gr.update(visible=False),
|
| 224 |
+
vect_textbox: user_symptoms,
|
| 225 |
+
quant_vect_textbox: quant_user_symptoms,
|
| 226 |
+
encrypted_vect_textbox: encrypted_quantized_user_symptoms_shorten_hex,
|
| 227 |
+
}
|
| 228 |
+
|
| 229 |
+
|
| 230 |
+
# def send_input(user_id, user_symptoms):
|
| 231 |
+
# """Send the encrypted input image as well as the evaluation key to the server.
|
| 232 |
+
|
| 233 |
+
# Args:
|
| 234 |
+
# user_id (int): The current user's ID.
|
| 235 |
+
# filter_name (str): The current filter to consider.
|
| 236 |
+
# """
|
| 237 |
+
# # Get the evaluation key path
|
| 238 |
+
|
| 239 |
+
|
| 240 |
+
# evaluation_key_path = get_client_file_path("evaluation_key", user_id, filter_name)
|
| 241 |
+
|
| 242 |
+
# if user_id == "" or not evaluation_key_path.is_file():
|
| 243 |
+
# raise gr.Error("Please generate the private key first.")
|
| 244 |
+
|
| 245 |
+
# encrypted_input_path = get_client_file_path("encrypted_image", user_id, filter_name)
|
| 246 |
+
# encrypted_symptoms_path = KEYS_PATH / f"{user_id}" / "encrypted_symtoms"
|
| 247 |
+
|
| 248 |
+
# if not encrypted_input_path.is_file():
|
| 249 |
+
# raise gr.Error("Please generate the private key and then encrypt an image first.")
|
| 250 |
+
|
| 251 |
+
# # Define the data and files to post
|
| 252 |
+
# data = {
|
| 253 |
+
# "user_id": user_id,
|
| 254 |
+
# "filter": filter_name,
|
| 255 |
+
# }
|
| 256 |
+
|
| 257 |
+
# files = [
|
| 258 |
+
# ("files", open(encrypted_input_path, "rb")),
|
| 259 |
+
# ("files", open(evaluation_key_path, "rb")),
|
| 260 |
+
# ]
|
| 261 |
+
|
| 262 |
+
# # Send the encrypted input image and evaluation key to the server
|
| 263 |
+
# url = SERVER_URL + "send_input"
|
| 264 |
+
# with requests.post(
|
| 265 |
+
# url=url,
|
| 266 |
+
# data=data,
|
| 267 |
+
# files=files,
|
| 268 |
+
# ) as response:
|
| 269 |
+
# return response.ok
|
| 270 |
+
|
| 271 |
+
|
| 272 |
+
# def decrypt_prediction(encrypted_quantized_vect, user_id):
|
| 273 |
+
# fhe_api = FHEModelClient(path_dir=REPO_DIR, key_dir=f".fhe_keys/{user_id}")
|
| 274 |
+
# fhe_api.load()
|
| 275 |
+
# fhe_api.generate_private_and_evaluation_keys(force=False)
|
| 276 |
+
# predictions = fhe_api.deserialize_decrypt_dequantize(encrypted_quantized_vect)
|
| 277 |
+
# return predictions
|
| 278 |
+
|
| 279 |
+
|
| 280 |
+
|
| 281 |
+
|
| 282 |
def clear_all_btn():
|
| 283 |
return {
|
| 284 |
+
box_default: None,
|
| 285 |
user_id_textbox: None,
|
| 286 |
eval_key_textbox: None,
|
| 287 |
+
quant_vect_textbox: None,
|
| 288 |
user_vector_textbox: None,
|
| 289 |
+
eval_key_len_textbox: None,
|
| 290 |
+
encrypted_vect_textbox: None,
|
| 291 |
+
error_box_1: gr.update(visible=False),
|
| 292 |
+
error_box_2: gr.update(visible=False),
|
| 293 |
+
error_box_3: gr.update(visible=False),
|
| 294 |
**{box: None for box in check_boxes},
|
| 295 |
}
|
| 296 |
|
| 297 |
|
| 298 |
if __name__ == "__main__":
|
| 299 |
print("Starting demo ...")
|
| 300 |
+
|
| 301 |
|
| 302 |
(df_train, X_train, X_test), (df_test, y_train, y_test) = load_data()
|
| 303 |
|
| 304 |
+
VALID_COLUMNS = X_train.columns.to_list()
|
| 305 |
|
| 306 |
# Load the model
|
| 307 |
with open("ConcreteXGBoostClassifier.pkl", "r", encoding="utf-8") as file:
|
|
|
|
| 358 |
)
|
| 359 |
check_boxes.append(check_box)
|
| 360 |
|
| 361 |
+
error_box_1 = gr.Textbox(label="Error", visible=False)
|
| 362 |
+
|
| 363 |
# User symptom vector
|
| 364 |
with gr.Row():
|
| 365 |
user_vector_textbox = gr.Textbox(
|
|
|
|
| 367 |
interactive=False,
|
| 368 |
max_lines=100,
|
| 369 |
)
|
|
|
|
| 370 |
|
| 371 |
with gr.Row():
|
| 372 |
# Submit botton
|
|
|
|
| 374 |
submit_button = gr.Button("Submit")
|
| 375 |
# Clear botton
|
| 376 |
with gr.Column():
|
| 377 |
+
clear_button = gr.Button("Clear")
|
| 378 |
|
| 379 |
# Click submit botton
|
| 380 |
|
| 381 |
submit_button.click(
|
| 382 |
+
fn=get_user_symptoms_vector_fn,
|
| 383 |
inputs=[box_default, *check_boxes],
|
| 384 |
+
outputs=[user_vector_textbox, error_box_1, *check_boxes],
|
| 385 |
)
|
| 386 |
|
| 387 |
gr.Markdown("# Step 2: Generate the keys")
|
| 388 |
gr.Markdown("Client side")
|
| 389 |
|
| 390 |
+
gen_key_btn = gr.Button("Generate the keys and send public part to server")
|
| 391 |
+
|
| 392 |
+
error_box_2 = gr.Textbox(label="Error", visible=False)
|
| 393 |
|
| 394 |
with gr.Row():
|
| 395 |
# User ID
|
|
|
|
| 414 |
interactive=False,
|
| 415 |
)
|
| 416 |
|
| 417 |
+
gen_key_btn.click(
|
| 418 |
+
key_gen_fn,
|
| 419 |
+
inputs=user_vector_textbox,
|
| 420 |
+
outputs=[eval_key_textbox, user_id_textbox, eval_key_len_textbox, error_box_2],
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 421 |
)
|
| 422 |
|
| 423 |
gr.Markdown("# Step 3: Encode the message with the private key")
|
| 424 |
gr.Markdown("Client side")
|
| 425 |
|
| 426 |
+
encrypt_btn = gr.Button("Encode the message with the private key and send it to the server")
|
| 427 |
+
|
| 428 |
+
error_box_3 = gr.Textbox(label="Error", visible=False)
|
| 429 |
|
| 430 |
with gr.Row():
|
| 431 |
|
|
|
|
| 446 |
label="Encrypted vector:", max_lines=4, interactive=False
|
| 447 |
)
|
| 448 |
|
| 449 |
+
encrypt_btn.click(
|
| 450 |
+
encrypt_fn,
|
| 451 |
inputs=[user_vector_textbox, user_id_textbox],
|
| 452 |
+
outputs=[vect_textbox, quant_vect_textbox, encrypted_vect_textbox, error_box_3],
|
| 453 |
)
|
| 454 |
|
| 455 |
gr.Markdown("# Step 4: Run the FHE evaluation")
|
|
|
|
| 465 |
label="Encrypted vector:", max_lines=4, interactive=False
|
| 466 |
)
|
| 467 |
|
| 468 |
+
# decrypt_target_botton.click(
|
| 469 |
+
# decrypt_prediction,
|
| 470 |
+
# inputs=[encrypted_vect_textbox, user_id_textbox],
|
| 471 |
+
# outputs=[decrypt_target_textbox],
|
| 472 |
+
# )
|
| 473 |
+
|
| 474 |
+
clear_button.click(
|
| 475 |
+
clear_all_btn,
|
| 476 |
+
outputs=[
|
| 477 |
+
box_default,
|
| 478 |
+
error_box_1,
|
| 479 |
+
error_box_2,
|
| 480 |
+
error_box_3,
|
| 481 |
+
user_id_textbox,
|
| 482 |
+
eval_key_textbox,
|
| 483 |
+
quant_vect_textbox,
|
| 484 |
+
user_vector_textbox,
|
| 485 |
+
eval_key_len_textbox,
|
| 486 |
+
encrypted_vect_textbox,
|
| 487 |
+
*check_boxes,
|
| 488 |
+
],
|
| 489 |
)
|
| 490 |
|
| 491 |
demo.launch()
|