HT_Keras_issue_2017_10_25

gist1.py

import os
import time
import numpy as np

from keras import objectives
from keras import regularizers
from keras import backend as K
from keras.layers import Input, Dense, Activation
from keras.models import Model
from keras.optimizers import Adam
from keras.layers.normalization import BatchNormalization


def main():
    os.environ["CUDA_VISIBLE_DEVICES"] = '0'
    K.clear_session()

    train_x = np.random.rand(1024,100)
    train_y = np.random.rand(1024,16)
    valid_x = np.random.rand(256,100)
    valid_y = np.random.rand(256,16)

    init_method = 'glorot_uniform'
    model_in = Input(shape=(train_x.shape[1], ))
    # H = Dense(96, kernel_initializer=init_method, kernel_regularizer=regularizers.l2(0.02))(model_in)
    H = Dense(96, kernel_initializer=init_method)(model_in)
    H = Activation('relu')(H)
    model_out = Dense(16, kernel_initializer=init_method)(H)

    model = Model(model_in, model_out)
    model.summary()
    model.compile(loss='mean_squared_error', optimizer=Adam())

    epochs = 3

    for epoch in range(1, epochs + 1):
        h = model.fit(train_x, train_y, epochs=1,
                      batch_size=128, verbose=0,
                      validation_data=(valid_x, valid_y))

        print("Epoch: {:04d}    ".format(epoch),
              "train_loss= {:.4f}".format(h.history['loss'][0]),
              "                        ",
              "valid_loss= {:.4f}".format(h.history['val_loss'][0]))

        score_train = model.evaluate(x=train_x, y=train_y, batch_size=128, verbose=0)
        score_valid = model.evaluate(x=valid_x, y=valid_y, batch_size=128, verbose=0)

        y_predict = model.predict(train_x, batch_size=128, verbose=0)
        score_numpy_train = np.mean(np.square(train_y - y_predict), axis=-1)

        y_predict = model.predict(valid_x, batch_size=128, verbose=0)
        score_numpy_valid = np.mean(np.square(valid_y - y_predict), axis=-1)

        print("After training:",
              "train_loss= {:.4f}".format(score_train),
              "train_loss_numpy= {:.4f}".format(np.mean(score_numpy_train)),
              "valid_loss= {:.4f}".format(score_valid),
              "valid_loss_numpy= {:.4f}".format(np.mean(score_numpy_valid)))

    return

main()

gist2.py

import os
import time
import numpy as np

from keras import objectives
from keras import regularizers
from keras import backend as K
from keras.layers import Input, Dense, Activation
from keras.models import Model
from keras.optimizers import Adam
from keras.layers.normalization import BatchNormalization


def main():
    os.environ["CUDA_VISIBLE_DEVICES"] = '0'
    K.clear_session()

    train_x = np.random.rand(1024,100)
    train_y = np.random.rand(1024,16)
    valid_x = np.random.rand(256,100)
    valid_y = np.random.rand(256,16)

    init_method = 'glorot_uniform'
    model_in = Input(shape=(train_x.shape[1], ))
    H = Dense(96, kernel_initializer=init_method, kernel_regularizer=regularizers.l2(0.02))(model_in)
    # H = Dense(96, kernel_initializer=init_method)(model_in)
    H = Activation('relu')(H)
    model_out = Dense(16, kernel_initializer=init_method)(H)

    model = Model(model_in, model_out)
    model.summary()
    model.compile(loss='mean_squared_error', optimizer=Adam())

    epochs = 3

    for epoch in range(1, epochs + 1):
        h = model.fit(train_x, train_y, epochs=1,
                      batch_size=128, verbose=0,
                      validation_data=(valid_x, valid_y))

        print("Epoch: {:04d}    ".format(epoch),
              "train_loss= {:.4f}".format(h.history['loss'][0]),
              "                        ",
              "valid_loss= {:.4f}".format(h.history['val_loss'][0]))

        score_train = model.evaluate(x=train_x, y=train_y, batch_size=128, verbose=0)
        score_valid = model.evaluate(x=valid_x, y=valid_y, batch_size=128, verbose=0)

        y_predict = model.predict(train_x, batch_size=128, verbose=0)
        score_numpy_train = np.mean(np.square(train_y - y_predict), axis=-1)

        y_predict = model.predict(valid_x, batch_size=128, verbose=0)
        score_numpy_valid = np.mean(np.square(valid_y - y_predict), axis=-1)

        print("After training:",
              "train_loss= {:.4f}".format(score_train),
              "train_loss_numpy= {:.4f}".format(np.mean(score_numpy_train)),
              "valid_loss= {:.4f}".format(score_valid),
              "valid_loss_numpy= {:.4f}".format(np.mean(score_numpy_valid)))

    return

main()