spatial_dropout_example.py

from keras.layers import Input, Dense, LSTM, Conv1D, Activation
from keras.layers import Dropout, AlphaDropout, BatchNormalization
from keras.layers import GlobalAveragePooling1D, Reshape, multiply
from keras.models import Model
import keras.backend as K

import numpy as np
import matplotlib.pyplot as plt

#############################################################################
def make_model(batch_shape):
    ipt = Input(batch_shape=batch_shape)
    x   = Conv1D(20, 4, activation='tanh')(ipt)
    x   = Dropout(0.2, noise_shape=(batch_shape[0], 1, K.int_shape(x)[-1]))(x)
    x   = GlobalAveragePooling1D()(x)
    out = Dense(1, activation='sigmoid')(x)

    model = Model(ipt, out)
    model.compile('adam', 'binary_crossentropy')
    return model

def make_data(batch_shape):
    return (np.random.randn(*batch_shape),
            np.random.randint(0, 2, (batch_shape[0], 1)))
    
def get_layer_outputs(model, layer_idx, input_data, learning_phase=0):
    layer = model.layers[layer_idx]
    layers_fn = K.function([model.input, K.learning_phase()], [layer.output])
    return layers_fn([input_data, learning_phase])[0]
#############################################################################
    
batch_shape = (32, 100, 16)
model = make_model(batch_shape)
x, y = make_data(batch_shape)

model.train_on_batch(x, y)

#############################################################################
outs = get_layer_outputs(model, 2, x, 1)
plt.imshow(outs[0].T, cmap='bwr', vmin=-1, vmax=1)

plt.title("Conv1D-Dropout w/ noise_shape=(batch_size, 1, channels)", weight='bold')
plt.xlabel("Timesteps", weight='bold')
plt.ylabel("Channels",  weight='bold')
plt.gcf().set_size_inches(10, 4)