Train_uni_batchnorm.py

import pandas
import matplotlib.pyplot as plt
import numpy
import matplotlib.pyplot as plt
import pandas
import math
import keras
from keras.models import Sequential
from keras.layers import Dense
from keras.layers import LSTM, GRU
from keras.layers import Dropout
from sklearn.preprocessing import MinMaxScaler
from sklearn.metrics import mean_squared_error
import numpy as np
import random

#dataset = pandas.read_csv('train.csv', usecols=[1], engine='python')
#dataset = dataset[:1600]

file = []
file.append(pandas.read_csv('train.oslo-vestby-report.2011-02-11_1618CET.log', names = ["band"], sep = '\t'))
file.append(pandas.read_csv('train_sim_traces/tram.jernbanetorget-ljabru-report.2010-12-09_1222CET.log', names = ["band"], sep = '\t'))
file.append(pandas.read_csv('train_sim_traces/ferry.nesoddtangen-oslo-report.2010-09-20_1542CEST.log', names = ["band"], sep = '\t'))
file.append(pandas.read_csv('train_sim_traces/car.aarnes-elverum-report.2011-02-10_1611CET.log', names = ["band"], sep = '\t'))
file.append(pandas.read_csv('train_sim_traces/bus.ljansbakken-oslo-report.2011-01-31_1025CET.log', names = ["band"], sep = '\t'))
file.append(pandas.read_csv('train_sim_traces/metro.kalbakken-jernbanetorget-report.2010-09-14_2303CEST.log', names = ["band"], sep = '\t'))

print(len(file))
print('Training data loaded!')



for epoch in range(10):

    random.shuffle(file)

    for cur in file:

        cur_data = np.asarray(cur)
        cur_data = np.vstack(cur_data)

        # plt.figure(figsize=(25,9))
        # plt.plot(dataset)
        # plt.show()

        scaler = MinMaxScaler(feature_range=(0, 1))
        cur_data = scaler.fit_transform(cur_data)

        train_size = int(len(cur_data) * 0.9)
        test_size = len(cur_data) - train_size
        train, test = cur_data[0:train_size, :], cur_data[train_size:len(cur_data), :]
        train_rev, test_rev = train[::-1], test[::-1]


        def create_dataset(dataset, look_back=1):
            dataX, dataY = [], []

            for i in range(len(dataset) - look_back - 1):
                a = dataset[i:(i + look_back), 0]
                dataX.append(a)
                dataY.append(dataset[i + look_back, 0])

            return numpy.array(dataX), numpy.array(dataY)


        look_back = 1
        trainX, trainY = create_dataset(train, look_back)
        testX, testY = create_dataset(test, look_back)

        trainX_rev, trainY_rev = create_dataset(train_rev, look_back)
        testX_rev, testY_rev = create_dataset(test_rev, look_back)

        # reshape input to be [samples, time steps, features]
        trainX = numpy.reshape(trainX, (trainX.shape[0], trainX.shape[1], 1))
        testX = numpy.reshape(testX, (testX.shape[0], testX.shape[1], 1))

        trainX_rev = numpy.reshape(trainX_rev, (trainX_rev.shape[0], trainX_rev.shape[1], 1))
        testX_rev = numpy.reshape(testX_rev, (testX_rev.shape[0], testX_rev.shape[1], 1))

        trainY_inverse = scaler.inverse_transform([trainY])
        testY_inverse = scaler.inverse_transform([testY])

        trainY_rev_inverse = scaler.inverse_transform([trainY_rev])
        testY_rev_inverse = scaler.inverse_transform([testY_rev])

        try:
            model_mse = keras.models.load_model('model_mse_uni3.h5')
            print('load existing model succeed')
        except:
            print('Training from new model')
            model_mse = Sequential()
            model_mse.add(GRU(256, input_shape=(trainX.shape[1], trainX.shape[2]), return_sequences=True))
            #model_mse.add(Dropout(0.4))
            #model_mse.add(GRU(256, return_sequences=True))
            model_mse.add(Dropout(0.4))
            # model_mse.add(Dropout(0.2))
            model_mse.add(GRU(64))
            # model_mse.add(Dropout(0.2))

            model_mse.add(Dense(1))
            model_mse.compile(loss='mean_squared_error', optimizer='adam')

        # model_mse.summary()

        try:
            model_mse.fit(trainX, trainY, epochs=20, batch_size=64, verbose=2)
            model_mse.save('model_mse_uni3.h5')
        except KeyboardInterrupt:
            model_mse.save('model_mse_uni3.h5')

        try:
            model_mse.fit(trainX_rev, trainY_rev, epochs=20, batch_size=64, verbose=2)
            model_mse.save('model_mse_uni3.h5')
        except KeyboardInterrupt:
            model_mse.save('model_mse_uni3.h5')

        model_mse.save('model_mse_uni3.h5')



model_mse.save('model_mse_uni3.h5')

print('Training process completed! Model saved!')