ChatGPTで特徴量エンジニアリング〜MAとRSIを特徴量として追加〜

import numpy as np
import pandas as pd
import talib
from sklearn.preprocessing import MinMaxScaler
from sklearn.metrics import mean_squared_error
from tensorflow.keras.models import Sequential
from tensorflow.keras.layers import Dense, GRU
from tensorflow.keras.optimizers import Adam
from tensorflow.keras.regularizers import l1_l2

# 移動平均とRSIを計算する関数
def add_MA(data, window_size):
ma = talib.SMA(data, window_size)
return np.column_stack((data, ma))

def add_RSI(data, window_size):
rsi = talib.RSI(data, window_size)
return np.column_stack((data, rsi))

# データの読み込み
data = pd.read_csv('your_data.csv') # 適切なファイル名に置き換えてください
data = data['Close'].values # 終値を取得

# 移動平均とRSIを追加
data_MA = add_MA(data, 5)
data_RSI = add_RSI(data_MA, 14)

# データの正規化
scaler = MinMaxScaler(feature_range=(0, 1))
data = scaler.fit_transform(data_RSI)

# データ分割用関数
def create_dataset(dataset, look_back=1):
x, y = [], []
for i in range(len(dataset) - look_back - 1):
x.append(dataset[i:(i + look_back), :])
y.append(dataset[i + look_back, 0])
return np.array(x), np.array(y)

# データ分割
look_back = 10
train_size = int(len(data) * 0.7)
test_size = len(data) - train_size
train, test = data[:train_size, :], data[train_size:, :]
x_train, y_train = create_dataset(train, look_back)
x_test, y_test = create_dataset(test, look_back)

# モデルの構築
model = Sequential()
model.add(GRU(32, return_sequences=True, input_shape=(look_back, 3), kernel_regularizer=l1_l2(l1=1e-5, l2=1e-4)))
model.add(GRU(32, kernel_regularizer=l1_l2(l1=1e-5, l2=1e-4)))
model.add(Dense(1))

# モデルのコンパイルと学習
model.compile(optimizer=Adam(lr=0.0005), loss='mean_squared_error')
history = model.fit(x_train, y_train, epochs=150, batch_size=64, validation_data=(x_test, y_test))

# 予測と評価
y_train_pred = model.predict(x_train)
y_test_pred = model.predict(x_test)

# 正規化の逆変換
y_train_pred = scaler.inverse_transform(np.column_stack((y_train_pred, np.zeros((y_train_pred.shape[0], 1)))))
y_test_pred = scaler.inverse_transform(np.column_stack((y_test_pred, np.zeros((y_test_pred.shape[0], 1)))))
y_train_true = scaler.inverse_transform(train[look_back:-1])
y_test_true = scaler.inverse_transform(test[look_back:-1])
# RMSEの計算
train_rmse = np.sqrt(mean_squared_error(y_train_true[:, 0], y_train_pred[:, 0]))
test_rmse = np.sqrt(mean_squared_error(y_test_true[:, 0], y_test_pred[:, 0]))
print('学習データのRMSE: ', train_rmse)
print('テストデータのRMSE: ', test_rmse)

from keras.layers import Bidirectional

model.add(Bidirectional(GRU(64, return_sequences=True)))
model.add(Dropout(0.2))
model.add(Bidirectional(GRU(32, return_sequences=False)))
model.add(Dropout(0.2))

import numpy as np
import pandas as pd
import talib
from sklearn.preprocessing import MinMaxScaler
from sklearn.metrics import mean_squared_error
from keras.models import Sequential
from keras.layers import GRU, Dense, Dropout
from keras.optimizers import Adam
from keras.regularizers import l1_l2
from keras.layers import Bidirectional

# 移動平均とRSIを計算する関数
def add_MA(data, window_size):
ma = talib.SMA(data, window_size)
return ma[window_size - 1:]

def add_RSI(data, window_size):
rsi = talib.RSI(data, window_size)
return rsi[window_size - 1:]

# データの読み込み
data = pd.read_pickle('/content/drive/My Drive/backtest/bar/df_H4.pkl') # 適切なファイル名に置き換えてください
data = data['Close'].values # 終値を取得

# 移動平均とRSIを追加
data_MA = add_MA(data, 5)
data_RSI = add_RSI(data_MA, 14).reshape(-1, 1)

# データの形状が一致するように調整
min_length = min(len(data_MA), len(data_RSI))
data_MA = data_MA[:min_length]
data_RSI = data_RSI[:min_length]

data = np.column_stack((data_MA, data_RSI))

# 欠損値を削除
data = data[~np.isnan(data).any(axis=1)]

# データの正規化
scaler = MinMaxScaler(feature_range=(0, 1))
data = scaler.fit_transform(data)

# データ分割用関数
def create_dataset(dataset, look_back=1):
x, y = [], []
for i in range(len(dataset) - look_back - 1):
x.append(dataset[i:(i + look_back), :])
y.append(dataset[i + look_back, 0])
return np.array(x), np.array(y)

# データ分割
look_back = 10
train_size = int(len(data) * 0.7)
test_size = len(data) - train_size
train, test = data[:train_size, :], data[train_size:, :]
x_train, y_train = create_dataset(train, look_back)
x_test, y_test = create_dataset(test, look_back)

# モデルの構築
model = Sequential()
model.add(Bidirectional(GRU(32, return_sequences=True, kernel_regularizer=l1_l2(l1=1e-5, l2=1e-4)), input_shape=(look_back, x_train.shape[2])))
model.add(Bidirectional(GRU(32, kernel_regularizer=l1_l2(l1=1e-5, l2=1e-4))))
model.add(Dropout(0.2))
model.add(Dense(1))

# モデルのコンパイルと学習
model.compile(optimizer=Adam(lr=0.0005), loss='mean_squared_error')
history = model.fit(x_train, y_train, epochs=150, batch_size=64, validation_data=(x_test, y_test))

# 予測と評価
y_train_pred = model.predict(x_train)
y_test_pred = model.predict(x_test)

# 正規化の逆変換
y_train_pred = scaler.inverse_transform(np.column_stack((y_train_pred, np.zeros((y_train_pred.shape[0], 1)))))
y_test_pred = scaler.inverse_transform(np.column_stack((y_test_pred, np.zeros((y_test_pred.shape[0], 1)))))
y_train_true = scaler.inverse_transform(train[look_back:-1])
y_test_true = scaler.inverse_transform(test[look_back:-1])

# RMSEの計算
train_rmse = np.sqrt(mean_squared_error(y_train_true[:, 0], y_train_pred[:, 0]))
test_rmse = np.sqrt(mean_squared_error(y_test_true[:, 0], y_test_pred[:, 0]))

print('学習データのRMSE: ', train_rmse)
print('テストデータのRMSE: ', test_rmse)