DeepLearning - keras.initializers 예제

from tensorflow import keras

from tensorflow.keras import initializers

import matplotlib.pyplot as plt

initializer1 = keras.initializers.RandomNormal(mean=0, stddev=1.)

initializer2 = keras.initializers.RandomUniform(minval=0,maxval=1.)

initializer3 = keras.initializers.TruncatedNormal(mean=0, stddev=1.)

initializer4 = keras.initializers.Zeros()

initializer5 = keras.initializers.Ones()

initializer6 = keras.initializers.GlorotNormal() #Xavier

initializer7 = keras.initializers.GlorotUniform() #Xavier

initializer8 = keras.initializers.HeNormal() #He

initializer9 = keras.initializers.HeUniform() #He

def tit_practice(activation, epochs, optimizer):

    for k in range(1, 10):

        print('initializer:', globals()[f'initializer{k}'])

        fig = plt.figure(figsize=(30, 15))

        for i in range(1, 10):

            globals()[f'ax{i}'] = fig.add_subplot(3, 3, i)

        for i in range(1, 10):

            ip = Input(shape=(2,))

           # n = BatchNormalization()(ip)

            #n = Dropout(0.5)(n)

            n = Dense(3, activation=activation, kernel_initializer=globals()[f'initializer{k}'])

           # n = BatchNormalization()(n)

            n = Dense(1, activation='linear', kernel_initializer=globals()[f'initializer{k}'])

            model = Model(inputs=ip, outputs=n)

            model.compile(loss='mse', optimizer=optimizer, metrics='accuracy')

            hist = model.fit(X_train, y_train, epochs=epochs, verbose=0)

            globals()[f'ax{i}'].plot(hist.history['loss'], 'y', label='train loss', c='blue')

            globals()[f'ax{i}'].plot(hist.history['accuracy'], 'y', label='train accuracy', c='red')

            globals()[f'ax{i}'].plot(range(epochs), [0.15 for _ in range(epochs)], linestyle='--')

            initial_name = ['RandomNormal', 'RandomUniform', 'TruncatedNormal', 'Zeros', 'Ones', 'GlorotNormal', 'GlorotUniform', 'HeNormal', 'HeUniform']

            globals()[f'ax{i}'].set_title(f'{initial_name[k-1]} & {activation} & {optimizer}')

            globals()[f'ax{i}'].set_xlabel('epoch')

            globals()[f'ax{i}'].set_ylabel('loss')

            globals()[f'ax{i}'].legend(loc='upper left')

            globals()[f'ax{i}'].set_ylim(0, 1)

            # print(model.evaluate(X, y))

        plt.show()