kerasで頭に描いたネットワーク構造を実現するためのTips ~ functional API 編 ~ ref: https://qiita.com/Mco7777/items/1339d01bc6ef028e7b44

file4.txt

y_i = \sum_{k=1}^{5} (w_{ki}*x_k) + b_i \quad (1 \leq i \leq 5)

functionalAPI.py

from keras.models import Model
from keras.layers import Input, Dense, Activation

model_input = Input(shape=(5,))
mid = Dense(10)(model_input)
output = Activation('softmax')(mid)
model = Model(inputs=model_input, outputs=output)

functionalAPI2.py

my_dense = Dense(10)
my_activation = Activation('softmax')

model_input = Input(shape=(5,))
mid = my_dense(model_input)
output = my_activation(mid)
model = Model(inputs=model_input, outputs=output)

functionalAPI3.py

my_dense = Dense(5)

model_input = Input(shape=(5,))
mid = my_dense(model_input)
mid2 = my_dense(mid)
output = Activation('softmax')(mid2)
model = Model(inputs=model_input, outputs=output)

functionalAPI4.py

import numpy as np
from keras import backend as K

# 適当な学習
model.compile(loss='mean_squared_error', optimizer='adam')
X = np.random.rand(10,5)
y = np.random.rand(10,5)
model.fit(X, y, epochs=5)

# 出力
print([K.eval(w) for w in model.weights])

# [array([[ 0.16185357, -0.15696804, -0.36604205, -0.02380839, -0.35026637],
#         [-0.13462609,  0.09170605,  0.13241905, -0.628663  , -0.67678481],
#         [ 0.15944755, -0.67250979,  0.35241356,  0.24243712,  0.10116834],
#         [ 0.61587954, -0.04093929,  0.36012208, -0.30871043, -0.31387496],
#         [ 0.66632921, -0.31295586,  0.59317648,  0.44108745,  0.09751857]], dtype=float32),
#  array([ 0.00472282,  0.00472244,  0.00472797,  0.00468933, -0.00472205], dtype=float32)]

functionalAPI5.py

from keras.models import Model
from keras.layers import Input, Dense, Activation, Multiply

model_input1 = Input(shape=(5,))
model_input2 = Input(shape=(5,))
mid = Dense(10)(model_input1)
mid2 = Dense(10)(model_input2)
multiplied = Multiply()([mid, mid2])
additional_dense = Dense(5)(multiplied)
additional_dense2 = Dense(5)(additional_dense)
output1 = Activation('softmax')(additional_dense)
output2 = Activation('softmax')(additional_dense2)
model = Model(inputs=[model_input1, model_input2], outputs=[output1, output2])