Question

我正在训练获取数据的模型，并使用新数据测试这些模型。如果模型的性能不如我希望的那样，ı必须使用新数据（new_data + old_data）生成新模型。由于训练需要时间，因此我尝试训练我拥有的模型，但是由于使用新数据（new_data + old_data）对旧模型进行缩放（标准化），因此这些模型在新的训练过程中将无法很好地发挥作用，因此无法进行训练他们而不是生成新模型。这些解决方案是否存在消除标准化错误的解决方案，从而可以有效地训练旧模型？在ı训练预训练模型时，我使用以下代码：

    import time
    import warnings
    import numpy as np
    import random
    import pandas as pd
    import smtplib
    import schedule
    from datetime import datetime 
    from sklearn.preprocessing import MinMaxScaler
    from keras.layers.core import Dense, Activation, Dropout
    from keras.layers.recurrent import LSTM
    from keras.layers import Bidirectional
    from keras.models import Sequential
    from keras.callbacks import ModelCheckpoint
    from keras.callbacks import ReduceLROnPlateau
    from keras import optimizers
    from keras import metrics
    from keras.models import load_model
    from keras.layers import Dense, LSTM, CuDNNLSTM, BatchNormalization, Bidirectional, Dropout, Activation, LeakyReLU
    warnings.filterwarnings("ignore", category = FutureWarning)
    pd.set_option('display.max_columns', None)  
    pd.set_option('display.expand_frame_repr', False)
    pd.set_option('max_colwidth', -1)
    seq_len = 50
    train_test_ratio = 0.0
    df = pd.read_csv('datasets/dk1007-0810LTCabc.csv', dtype=float, header=None)
    model = load_model('models/LTC4/RNN_Final-56-0.02')
    df[df == 0] = 1
    df1  = df.iloc[:, :]
    ##y_df1 = df1.iloc[:,10:11]#,asks1[1:2],askamount[2:3],bids1[3:4],bidsamount[4:5],price[5:6],price_5m_change[6:7],price_10m_change[7:8],price_15m_change[8:9],price_20m_change[9:10],price_30m_change[10:11],price_60m_change[11:12],asks1_change,bids1_change,asksamount/bidsamount,price_change_1m
    y_df1 = df1.iloc[:,6:7]#,asks1[1:2],askamount[2:3],bids1[3:4],bidsamount[4:5],price[5:6],price_5m_change[6:7],price_10m_change[7:8],price_15m_change[8:9],price_20m_change[9:10],price_30m_change[10:11],price_60m_change[11:12],price_60m_change_target[12:13],price_30m_change_target[13:14],price_20m_change_target[14:15],price_15m_change_target[15:16],price_10m_change_target[16:17],price_5m_change_target[17:18],asks1_change,bids1_change,asksamount/bidsamount,price_change_1m

    df = df1.iloc[:,12:]

    ts = 0
    scaler = MinMaxScaler()
    df = scaler.fit_transform(df)

    scaler2 = MinMaxScaler()
    y_df = scaler2.fit_transform(y_df1)
    df1 = (df,y_df)
    seq_data  = []
    data_X = []
    data_Y = []

    for i in range(0,len(df)-seq_len):
        dft = df1[0][i:i+seq_len]
        seq_data.append([dft, df1[1][i+seq_len]])
    random.shuffle(seq_data)
    for i in range(0, len(df)-seq_len):
    ##    if seq_data[i][1][0] != 0.5417594375446555:
        data_X.append(seq_data[i][0])
        data_Y.append(seq_data[i][1])
    X_train = np.array(data_X)
    Y_train = np.array(data_Y)

    #split into train and testing data
    row = round(train_test_ratio * X_train.shape[0])
    x_test = X_train[:int(row), :]
    x_train = X_train[int(row):,:]
    y_test = Y_train[:int(row)]
    y_train= Y_train[int(row):]

    print("X_train:", x_train.shape)
    print("y_train:", y_train.shape)
    print("X_test:", x_test.shape)
    print("y_test:", y_test.shape)


    filepath = "RNN_Final-{epoch:02d}-{val_loss:.2f}"  # unique file name that will include the epoch and the validation loss for that epoch
    checkpoint = ModelCheckpoint("models/{}".format(filepath, monitor='val_loss', verbose=1, save_best_only=True, mode='min')) # saves only the best ones
    reduce_lr = ReduceLROnPlateau(monitor = "val_loss", factor = 0.75, patience = 15,
                                             verbose = 1, mode = "auto", min_delta = 1e-04, cooldown = 5,
                                              min_lr = 0)

    start = time.time()
    model.fit(x_train, y_train, batch_size=16, epochs=500, validation_split=0.1, callbacks=[checkpoint, reduce_lr])
    model.save('0310_5m')
    print('training time : ', time.time() - start)

在keras中训练预训练模型

0 个答案: