keras_matrix_weight.py

import numpy as np
import keras
from sklearn.model_selection import train_test_split

"""
ベクトルx,yは与えられているとする．
y = A * x となる行列Aを求める．
- 行列とベクトルの積, <https://stackoverflow.com/questions/46570963/keras-lambda-layer-for-matrix-vector-multiplication>
"""


def build_model(input_dim=3, output_dim=4):
    x = keras.models.Input(shape=(input_dim,))
    A = np.zeros((input_dim, output_dim))
    y = keras.layers.Dense(output_dim, trainable=True, use_bias=False, name="matrix")(x)
    model = keras.models.Model(x, y)
    return model

def load_data():
    A = np.random.random((4,3))
    print(A)
    x, y = [], []
    for i in range(10000):
        x_i = np.random.random(3)
        y_i = np.dot(A, x_i)
        x.append(x_i)
        y.append(y_i)
    x = np.asarray(x)
    y = np.asarray(y)
    x_train, x_test, y_train, y_test = train_test_split(x, y)
    return (x_train, y_train), (x_test, y_test)

def run():
    (x_train, y_train), (x_test, y_test) = load_data()
    print(x_train.shape, x_test.shape, y_train.shape)
    model = build_model()
    print(model.summary())
    model.compile(
        optimizer='rmsprop',
        loss='mae',
        metrics=['accuracy']
    )   
    model.fit(
        x_train,
        y_train,
        validation_data=(x_test, y_test),
        epochs=10,
    )   
    for layer in model.layers:
        if layer.name == "matrix":
            weights = layer.get_weights()
            print(weights)

if __name__ == "__main__":
    run()