Spaces:
Runtime error
Runtime error
Update pages/3_BiLTSM_Прогноз bitcoin.py
Browse files
pages/3_BiLTSM_Прогноз bitcoin.py
CHANGED
|
@@ -1,46 +1,36 @@
|
|
| 1 |
import streamlit as st
|
| 2 |
-
import
|
| 3 |
-
import torch.nn as nn
|
| 4 |
-
import numpy as np
|
| 5 |
import pandas as pd
|
|
|
|
| 6 |
import matplotlib.pyplot as plt
|
| 7 |
-
|
| 8 |
-
|
| 9 |
-
import yfinance as yf
|
| 10 |
-
from model.lstm_model import BiLSTM
|
| 11 |
import torch.optim as optim
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 12 |
|
|
|
|
|
|
|
|
|
|
|
|
|
| 13 |
|
| 14 |
-
|
| 15 |
-
|
| 16 |
-
|
| 17 |
-
# TODAY = "2024-01-01" # Замените на текущую дату или нужную вам дату окончания
|
| 18 |
-
# btc_usdt = yf.download('BTC-USD', START, TODAY)
|
| 19 |
-
# btc_usdt.reset_index(inplace=True)
|
| 20 |
-
# return btc_usdt
|
| 21 |
|
| 22 |
-
|
| 23 |
-
|
|
|
|
| 24 |
|
| 25 |
-
|
| 26 |
-
|
| 27 |
-
btc_usdt = pd.read_csv(data_path)
|
| 28 |
|
| 29 |
scaler = MinMaxScaler(feature_range=(0, 1))
|
| 30 |
-
scaled_data = scaler.fit_transform(
|
| 31 |
-
|
| 32 |
-
def rmse(predictions, targets):
|
| 33 |
-
return np.sqrt(((predictions - targets) ** 2).mean())
|
| 34 |
-
|
| 35 |
-
def mape(predictions, targets):
|
| 36 |
-
return np.mean(np.abs((targets - predictions) / targets)) * 100
|
| 37 |
|
| 38 |
-
def weighted_mape(predictions, targets, weights):
|
| 39 |
-
errors = np.abs(targets - predictions)
|
| 40 |
-
weighted_errors = errors * weights
|
| 41 |
-
weighted_mape = np.sum(weighted_errors) / np.sum(np.abs(targets) * weights) * 100
|
| 42 |
-
return weighted_mape
|
| 43 |
-
|
| 44 |
def create_dataset(data, time_step=1):
|
| 45 |
X, Y = [], []
|
| 46 |
for i in range(len(data)-time_step-1):
|
|
@@ -49,94 +39,105 @@ def create_dataset(data, time_step=1):
|
|
| 49 |
Y.append(data[i + time_step, 0])
|
| 50 |
return np.array(X), np.array(Y)
|
| 51 |
|
| 52 |
-
|
| 53 |
-
|
| 54 |
-
X, y = create_dataset(scaled_data, time_step)
|
| 55 |
|
| 56 |
# Разделяем данные на обучающую и тестовую выборки
|
| 57 |
-
test_size =
|
| 58 |
train_size = len(X) - test_size
|
| 59 |
X_train, X_test = X[:train_size], X[train_size:]
|
| 60 |
y_train, y_test = y[:train_size], y[train_size:]
|
| 61 |
|
| 62 |
-
X_train = torch.Tensor(X_train).unsqueeze(-1)
|
| 63 |
-
X_test = torch.Tensor(X_test).unsqueeze(-1)
|
| 64 |
y_train = torch.Tensor(y_train)
|
| 65 |
y_test = torch.Tensor(y_test)
|
| 66 |
|
| 67 |
-
# Загрузка весов модели
|
| 68 |
-
weights_path = "model/model_weights.pth"
|
| 69 |
input_size = 1
|
| 70 |
-
hidden_size = 128
|
| 71 |
-
num_layers = 3
|
| 72 |
output_size = 1
|
| 73 |
-
|
| 74 |
-
# Создание BiLSTM модели
|
| 75 |
model = BiLSTM(input_size, hidden_size, num_layers, output_size)
|
| 76 |
criterion = nn.MSELoss()
|
| 77 |
-
optimizer =
|
| 78 |
|
| 79 |
-
|
| 80 |
-
model.load_state_dict(torch.load(weights_path))
|
| 81 |
model.eval()
|
| 82 |
-
|
| 83 |
-
# Получение предсказаний с помощью загруженной модели
|
| 84 |
with torch.no_grad():
|
| 85 |
test_predictions = model(X_test)
|
| 86 |
test_loss = criterion(test_predictions, y_test.view(-1, 1))
|
| 87 |
|
| 88 |
-
test_predictions = scaler.inverse_transform(test_predictions.numpy())
|
| 89 |
-
y_test = scaler.inverse_transform(y_test.
|
| 90 |
-
|
| 91 |
-
|
|
|
|
| 92 |
weights = np.array([1.0, 0.9, 0.8, 0.7, 0.6, 0.5, 0.4])
|
| 93 |
-
test_weighted_mape =
|
| 94 |
|
| 95 |
-
|
| 96 |
-
st.
|
| 97 |
-
st.write(
|
| 98 |
|
| 99 |
-
# Генерация предсказаний для будущих дат
|
| 100 |
-
with torch.no_grad():
|
| 101 |
-
future_predictions = model(X_test) # Предсказания для тестовой выборки
|
| 102 |
-
future_predictions = scaler.inverse_transform(future_predictions.numpy())
|
| 103 |
|
| 104 |
-
#
|
| 105 |
-
future_dates = [
|
| 106 |
-
|
| 107 |
-
# Преобразование предсказаний для будущих дат в формат DataFrame
|
| 108 |
-
future_df = pd.DataFrame({
|
| 109 |
-
'Date': future_dates,
|
| 110 |
-
'Predicted_Price': future_predictions.flatten()
|
| 111 |
-
})
|
| 112 |
-
|
| 113 |
-
# Вывод графика предсказаний на будущие даты
|
| 114 |
-
st.line_chart(future_df.set_index('Date')['Predicted_Price'], use_container_width=True)
|
| 115 |
-
st.title("Future Bitcoin Price Predictions")
|
| 116 |
-
|
| 117 |
-
# Вывод метрик по кнопке
|
| 118 |
-
if st.button("Show Metrics"):
|
| 119 |
-
st.write("### Test Metrics")
|
| 120 |
-
st.write(f"**Test Loss:** {test_loss.item():.4f}")
|
| 121 |
-
st.write(f"**Test RMSE:** {test_rmse:.4f}")
|
| 122 |
-
st.write(f"**Test MAPE:** {test_mape:.4f}%")
|
| 123 |
-
st.write(f"**Test Weighted MAPE:** {test_weighted_mape:.4f}%")
|
| 124 |
-
# Вывод графика предсказаний на будущие даты
|
| 125 |
-
st.line_chart(future_df.set_index('Date')['Predicted_Price'], use_container_width=True)
|
| 126 |
-
st.title("Future Bitcoin Price Predictions")
|
| 127 |
-
|
| 128 |
-
# Вывод метрик
|
| 129 |
-
st.write("### Future Predictions (Next 14 Days)")
|
| 130 |
-
st.write(f"**Average Predicted Price:** {future_df['Predicted_Price'].mean():.2f}")
|
| 131 |
-
st.write(f"**Maximum Predicted Price:** {future_df['Predicted_Price'].max():.2f}")
|
| 132 |
-
st.write(f"**Minimum Predicted Price:** {future_df['Predicted_Price'].min():.2f}")
|
| 133 |
|
| 134 |
-
#
|
| 135 |
-
trend_first_day =
|
| 136 |
-
trend_last_day =
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 137 |
|
| 138 |
-
|
| 139 |
-
|
| 140 |
|
| 141 |
-
|
| 142 |
-
st.write(
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
import streamlit as st
|
| 2 |
+
import yfinance as yf
|
|
|
|
|
|
|
| 3 |
import pandas as pd
|
| 4 |
+
import numpy as np
|
| 5 |
import matplotlib.pyplot as plt
|
| 6 |
+
import torch
|
| 7 |
+
import torch.nn as nn
|
|
|
|
|
|
|
| 8 |
import torch.optim as optim
|
| 9 |
+
from datetime import date, datetime, timedelta
|
| 10 |
+
from model.lstm_model import BiLSTM
|
| 11 |
+
from sklearn.metrics import mean_squared_error
|
| 12 |
+
from sklearn.preprocessing import MinMaxScaler
|
| 13 |
+
import plotly.graph_objects as go
|
| 14 |
|
| 15 |
+
st.set_page_config(layout='wide', initial_sidebar_state='expanded')
|
| 16 |
+
st.set_option('deprecation.showPyplotGlobalUse', False)
|
| 17 |
+
st.title('ML Wall Street')
|
| 18 |
+
st.image('images/img.png')
|
| 19 |
|
| 20 |
+
START = "2021-01-01"
|
| 21 |
+
TODAY = date.today().strftime("%Y-%m-%d")
|
| 22 |
+
period = st.slider('Количество дней прогноза:', 1, 14, 14)
|
|
|
|
|
|
|
|
|
|
|
|
|
| 23 |
|
| 24 |
+
df = yf.download('BTC-USD', START, TODAY)
|
| 25 |
+
df.reset_index(inplace=True)
|
| 26 |
+
df['Date'] = pd.to_datetime(df['Date']) # Преобразование в формат datetime
|
| 27 |
|
| 28 |
+
latest_date = df['Date'].iloc[-1].strftime('%Y-%m-%d')
|
| 29 |
+
st.markdown(f"<h3 style='text-align: center;'>Цены актуальны на последнюю дату закрытия торгов {latest_date}</h3>", unsafe_allow_html=True)
|
|
|
|
| 30 |
|
| 31 |
scaler = MinMaxScaler(feature_range=(0, 1))
|
| 32 |
+
scaled_data = scaler.fit_transform(df['Adj Close'].values.reshape(-1, 1))
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 33 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 34 |
def create_dataset(data, time_step=1):
|
| 35 |
X, Y = [], []
|
| 36 |
for i in range(len(data)-time_step-1):
|
|
|
|
| 39 |
Y.append(data[i + time_step, 0])
|
| 40 |
return np.array(X), np.array(Y)
|
| 41 |
|
| 42 |
+
X, y = create_dataset(scaled_data, time_step=period)
|
|
|
|
|
|
|
| 43 |
|
| 44 |
# Разделяем данные на обучающую и тестовую выборки
|
| 45 |
+
test_size = period
|
| 46 |
train_size = len(X) - test_size
|
| 47 |
X_train, X_test = X[:train_size], X[train_size:]
|
| 48 |
y_train, y_test = y[:train_size], y[train_size:]
|
| 49 |
|
| 50 |
+
X_train = torch.Tensor(X_train).unsqueeze(-1)
|
| 51 |
+
X_test = torch.Tensor(X_test).unsqueeze(-1)
|
| 52 |
y_train = torch.Tensor(y_train)
|
| 53 |
y_test = torch.Tensor(y_test)
|
| 54 |
|
|
|
|
|
|
|
| 55 |
input_size = 1
|
| 56 |
+
hidden_size = 128
|
| 57 |
+
num_layers = 3
|
| 58 |
output_size = 1
|
|
|
|
|
|
|
| 59 |
model = BiLSTM(input_size, hidden_size, num_layers, output_size)
|
| 60 |
criterion = nn.MSELoss()
|
| 61 |
+
optimizer = optim.Adam(model.parameters(), lr=0.001)
|
| 62 |
|
| 63 |
+
model.load_state_dict(torch.load('model/model_weights.pth'))
|
|
|
|
| 64 |
model.eval()
|
|
|
|
|
|
|
| 65 |
with torch.no_grad():
|
| 66 |
test_predictions = model(X_test)
|
| 67 |
test_loss = criterion(test_predictions, y_test.view(-1, 1))
|
| 68 |
|
| 69 |
+
test_predictions = scaler.inverse_transform(test_predictions.cpu().numpy())
|
| 70 |
+
y_test = scaler.inverse_transform(y_test.view(-1, 1).cpu().numpy())
|
| 71 |
+
|
| 72 |
+
test_rmse = np.sqrt(mean_squared_error(test_predictions, y_test))
|
| 73 |
+
test_mape = np.mean(np.abs((y_test - test_predictions) / y_test)) * 100
|
| 74 |
weights = np.array([1.0, 0.9, 0.8, 0.7, 0.6, 0.5, 0.4])
|
| 75 |
+
test_weighted_mape = np.sum(np.abs(y_test - test_predictions) * weights) / np.sum(np.abs(y_test) * weights) * 100
|
| 76 |
|
| 77 |
+
st.write(f'Test RMSE: {test_rmse:.4f}')
|
| 78 |
+
st.write(f'Test MAPE: {test_mape:.4f}%')
|
| 79 |
+
st.write(f'Test Weighted MAPE: {test_weighted_mape:.4f}%')
|
| 80 |
|
|
|
|
|
|
|
|
|
|
|
|
|
| 81 |
|
| 82 |
+
# Получение дат для будущих предсказаний
|
| 83 |
+
future_dates = pd.date_range(df['Date'].iloc[-1] + timedelta(days=1), periods=14)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 84 |
|
| 85 |
+
# Прогнозы с учетом трендов
|
| 86 |
+
trend_first_day = test_predictions[0] - df['Adj Close'].iloc[-1]
|
| 87 |
+
trend_last_day = test_predictions[-1] - test_predictions[0]
|
| 88 |
+
adjusted_future_predictions = [df['Adj Close'].iloc[-1] + trend_first_day + (trend_last_day / 14) * i for i in range(14)]
|
| 89 |
+
|
| 90 |
+
# Создание графика
|
| 91 |
+
plt.figure(figsize=(12, 6))
|
| 92 |
+
plt.plot(df['Date'], df['Adj Close'], label='Actual', linewidth=2)
|
| 93 |
+
plt.plot(future_dates, adjusted_future_predictions, label='Предсказание на 14 дней', linestyle='--', linewidth=2)
|
| 94 |
+
|
| 95 |
+
plt.title('Предсказание стоимости Bitcoin на 14 Дней')
|
| 96 |
+
plt.xlabel('Date')
|
| 97 |
+
plt.ylabel('Price')
|
| 98 |
+
plt.xticks(rotation=45)
|
| 99 |
+
plt.legend()
|
| 100 |
|
| 101 |
+
# Отображение графика в Streamlit
|
| 102 |
+
st.pyplot(plt)
|
| 103 |
|
| 104 |
+
# Вывод прогнозов
|
| 105 |
+
st.write("Прогноз на 14 дней вперед:")
|
| 106 |
+
for i, (date, pred) in enumerate(zip(future_dates, adjusted_future_predictions)):
|
| 107 |
+
st.write(f"День {i + 1} ({date}): Прогноз: {float(pred):.2f}")
|
| 108 |
+
|
| 109 |
+
# Вывод тренда на первый и последний день
|
| 110 |
+
trend_first_day_text = "Рост" if trend_first_day > 0 else "Падение" if trend_first_day < 0 else "Нет тренда"
|
| 111 |
+
trend_last_day_text = "Рост" if trend_last_day > 0 else "Падение" if trend_last_day < 0 else "Нет тренда"
|
| 112 |
+
|
| 113 |
+
st.write(f"Тренд на первый день: {trend_first_day_text}")
|
| 114 |
+
st.write(f"Тренд на последний день: {trend_last_day_text}")
|
| 115 |
+
|
| 116 |
+
# Создание графика с использованием Plotly
|
| 117 |
+
fig = go.Figure()
|
| 118 |
+
# График для фактических данных
|
| 119 |
+
fig.add_trace(go.Scatter(x=df['Date'], y=df['Adj Close'], mode='lines', name='Фактические данные', line=dict(width=2)))
|
| 120 |
+
# График для предсказаний на 14 дней
|
| 121 |
+
fig.add_trace(go.Scatter(x=future_dates, y=adjusted_future_predictions, mode='lines', name='Прогноз (14 дней)', line=dict(width=2, dash='dash')))
|
| 122 |
+
|
| 123 |
+
# Настройки макета графика
|
| 124 |
+
fig.update_layout(
|
| 125 |
+
title='Прогноз стоимости Bitcoin на следующие 14 дней',
|
| 126 |
+
xaxis_title='Дата',
|
| 127 |
+
yaxis_title='Цена',
|
| 128 |
+
xaxis=dict(tickangle=-45),
|
| 129 |
+
legend=dict(x=0, y=1, traceorder='normal'),
|
| 130 |
+
margin=dict(l=0, r=0, t=30, b=0),
|
| 131 |
+
|
| 132 |
+
# range slider
|
| 133 |
+
xaxis_rangeslider=dict(
|
| 134 |
+
visible=True,
|
| 135 |
+
thickness=0.05, # Толщина слайдера
|
| 136 |
+
bordercolor='white', # Цвет границы слайдера
|
| 137 |
+
bgcolor='rgba(219, 219, 219, 0.25)', # Цвет фона слайдера
|
| 138 |
+
range=[df['Date'].iloc[0], future_dates[-1]] # Обновление диапазона дат
|
| 139 |
+
)
|
| 140 |
+
)
|
| 141 |
+
|
| 142 |
+
# Отображение графика
|
| 143 |
+
st.plotly_chart(fig, use_container_width=True)
|