g

pages/3_🪙_Прогноз_Bitcoin USD. BiLSTM.py

@@ -0,0 +1,169 @@
+import streamlit as st
+import yfinance as yf
+import pandas as pd
+import numpy as np
+import matplotlib.pyplot as plt
+import torch
+import torch.nn as nn
+import torch.optim as optim
+from datetime import date, datetime, timedelta
+from model.BiLSTM_model import BiLSTM
+from sklearn.metrics import mean_squared_error
+from sklearn.preprocessing import MinMaxScaler
+from datetime import timedelta
+
+st.set_page_config(layout='wide', initial_sidebar_state='expanded')
+st.set_option('deprecation.showPyplotGlobalUse', False)
+st.title('ML Wall Street')
+st.image('images/img.png')
+
+START = "2021-01-01"
+TODAY = date.today().strftime("%Y-%m-%d")
+period = st.slider('Количество дней прогноза:', 1, 14, 14)
+
+df = yf.download('BTC-USD', START, TODAY)
+df.reset_index(inplace=True)
+
+latest_date = df['Date'].iloc[-1].strftime('%Y-%m-%d')
+st.markdown(f"<h3 style='text-align: center;'>Цены актуальны на последнюю дату закрытия торгов {latest_date}</h3>", unsafe_allow_html=True)
+
+def rmse(predictions, targets):
+    return np.sqrt(((predictions - targets) ** 2).mean())
+
+def mape(predictions, targets):
+    return np.mean(np.abs((targets - predictions) / targets)) * 100
+
+def weighted_mape(predictions, targets, weights):
+    errors = np.abs(targets - predictions)
+    weighted_errors = errors * weights
+    weighted_mape = np.sum(weighted_errors) / np.sum(np.abs(targets) * weights) * 100
+    return weighted_mape
+
+scaler = MinMaxScaler(feature_range=(0, 1))
+scaled_data = scaler.fit_transform(df['Adj Close'].values.reshape(-1,1))
+
+def create_dataset(data, time_step=1):
+    X, Y = [], []
+    for i in range(len(data)-time_step-1):
+        a = data[i:(i+time_step), 0]
+        X.append(a)
+        Y.append(data[i + time_step, 0])
+    return np.array(X), np.array(Y)
+
+X, y = create_dataset(scaled_data, time_step=period)
+
+# Разделяем данные на обучающую и тестовую выборки
+test_size = period  # 14 дней для теста
+train_size = len(X) - test_size
+X_train, X_test = X[:train_size], X[train_size:]
+y_train, y_test = y[:train_size], y[train_size:]
+
+X_train = torch.Tensor(X_train).unsqueeze(-1)  # Добавляем размерность
+X_test = torch.Tensor(X_test).unsqueeze(-1)    # Для совместимости с LSTM
+y_train = torch.Tensor(y_train)
+y_test = torch.Tensor(y_test)
+
+input_size = 1  # Входной размер равен 1, так как мы рассматриваем одну фичу - цену
+hidden_size = 128
+num_layers = 3
+output_size = 1
+model = BiLSTM(input_size, hidden_size, num_layers, output_size)
+criterion = nn.MSELoss()
+optimizer = optim.Adam(model.parameters(), lr=0.001)
+
+model = BiLSTM(1, 128, 3, 1)
+model.load_state_dict(torch.load('model/model_weights.pth'))
+model.eval()
+
+model.eval()
+with torch.no_grad():
+    test_predictions = model(X_test)
+    test_loss = criterion(test_predictions, y_test.view(-1, 1))
+
+    test_predictions = scaler.inverse_transform(test_predictions.cpu().numpy())
+    y_test = scaler.inverse_transform(y_test.view(-1, 1).cpu().numpy())
+    
+    test_rmse = rmse(test_predictions, y_test)
+    test_mape = mape(test_predictions, y_test)
+    weights = np.array([1.0, 0.9, 0.8, 0.7, 0.6, 0.5, 0.4])
+    test_weighted_mape = weighted_mape(test_predictions, y_test, weights)
+
+    # st.write(f'Test Loss: {test_loss.item():.4f}')
+    st.write(f'Test RMSE: {test_rmse:.4f}')
+    st.write(f'Test MAPE: {test_mape:.4f}%')
+    st.write(f'Test Weighted MAPE: {test_weighted_mape:.4f}%')
+
+# Предсказания на тестовых данных
+with torch.no_grad():
+    test_predictions = model(X_test)
+    adjusted_future_predictions = scaler.inverse_transform(test_predictions.cpu().numpy())
+
+# st.write(adjusted_future_predictions)
+# df_predictions = pd.DataFrame({'Predictions': adjusted_future_predictions.flatten()})
+# df_predictions['Predictions'] = df.iloc[:, 0].values
+# sum_test = pd.concat([df['Date'][train_size:], df_predictions], axis=1)
+# st.write(sum_test)
+# # adjusted_future_predictions = pd.DataFrame({'Date': df['Date'][train_size:].reset_index(), 'Predictions': adjusted_future_predictions})
+# # st.write(adjusted_future_predictions)
+# # pd.concat(df['Date'][train_size:], adjusted_future_predictions)
+filtered_df = df
+
+# Прогнозы с учетом трендов
+trend_first_day = adjusted_future_predictions[0] - filtered_df['Adj Close'].iloc[-1]
+trend_14_days = adjusted_future_predictions[-1] - adjusted_future_predictions[0]
+adjusted_future_predictions = [filtered_df['Adj Close'].iloc[-1] + trend_first_day + (trend_14_days / 14) * i for i in range(14)]
+
+# Вычисление даты, которая находится на 3 месяца назад
+start_date = datetime.strptime(latest_date, '%Y-%m-%d') - timedelta(days=90)
+
+# Преобразование строк столбца 'Date' в объекты datetime
+df['Date'] = pd.to_datetime(df['Date'])
+
+# Фильтра��ия данных за последние 3 месяца
+filtered_df = df[(df['Date'] >= start_date) & (df['Date'] <= latest_date)]
+
+# Генерация дат для будущих предсказаний
+future_dates = [filtered_df['Date'].values[-1] + np.timedelta64(i+1, 'D') for i in range(14)]
+
+# Расчет тренда на первый день
+trend_first_day = adjusted_future_predictions[0] - filtered_df['Adj Close'].iloc[-1]
+
+# Расчет тренда за 14 дней (как разницу между первым и последним днем)
+trend_14_days = adjusted_future_predictions[-1] - adjusted_future_predictions[0]
+
+# Прогнозы с учетом трендов
+adjusted_future_predictions = [filtered_df['Adj Close'].iloc[-1] + trend_first_day + (trend_14_days / 14) * i for i in range(14)]
+
+# Создание графика
+plt.figure(figsize=(12,6))
+plt.plot(filtered_df['Date'], filtered_df['Adj Close'], label='Actual', linewidth=2)
+plt.plot(future_dates, adjusted_future_predictions, label='Future Predictions (with Trends)', linestyle='--', linewidth=2)
+
+plt.title('Bitcoin Price Prediction (Last 3 Months)')
+plt.xlabel('Date')
+plt.ylabel('Price')
+plt.xticks(rotation=45)
+plt.legend()
+# Отображение графика в Streamlit
+st.pyplot(plt)
+
+# Генерация дат для будущих предсказаний
+future_dates = [df['Date'].values[-1] + np.timedelta64(i+1, 'D') for i in range(14)]
+
+# Расчет тренда на первый день
+trend_first_day = adjusted_future_predictions[0] - filtered_df['Adj Close'].iloc[-1]
+
+# Расчет тренда на последний день
+trend_last_day = adjusted_future_predictions[-1] - adjusted_future_predictions[0]
+
+# Вывод прогнозов
+print("Прогноз на 14 дней вперед:")
+for i, (date, pred) in enumerate(zip(future_dates, adjusted_future_predictions)):
+    print(f"День {i+1} ({date}): Прогноз: {float(pred):.2f}")
+
+# Вывод тренда на первый и последний день
+trend_first_day_text = "Рост" if trend_first_day > 0 else "Падение" if trend_first_day < 0 else "Нет тренда"
+trend_last_day_text = "Рост" if trend_last_day > 0 else "Падение" if trend_last_day < 0 else "Нет тренда"
+
+st.write(f"Тренд на первый день: {trend_first_day_text}")
+st.write(f"Тренд на последний день: {trend_last_day_text}")