Simple Feed Forward Neural Net Using tf.keras written for reproducing issue #27316 mentioned in tensorflow/issues

simple_fcnn.py

import numpy as np
import tensorflow as tf
import tensorflow.keras.layers as L
import tensorflow.keras.activations as Z
from tensorflow.keras import Model

from pprint import pprint

def feed_forward_nn(input_shape):
    with tf.variable_scope("keras_"):
        _input_ = L.Input(shape=input_shape)
        layer_1 = L.Dense(5, Z.relu)(_input_)
        output = L.Dense(1, Z.tanh)(layer_1)
    return Model(inputs=_input_, outputs=output)
    
def main():
    model1 = feed_forward_nn((3, ))
    model2 = feed_forward_nn((3, ))
    print("Model weights before:\n")
    pprint(model1.get_weights())
    pprint(model2.get_weights())
    print()
    model1.set_weights(model2.get_weights())
    are_wts_equal = all([(i == j).all() for i, j in zip(model1.get_weights(), model2.get_weights())])
    print("Model weights after:\n")
    pprint(model1.get_weights())
    pprint(model2.get_weights())
    print()
    print("\nAre Model Weights Equal (using <tf.keras.Model>.get_weights()): {}".format(are_wts_equal))
    print("\nIf YES then these should be equal!!!")
    some_input = np.random.random((1, 3))
    print(model1.predict(some_input))
    print(model2.predict(some_input))
    sess = tf.Session()
    sess.run(tf.global_variables_initializer())
    sess.run(tf.get_collection(tf.GraphKeys.GLOBAL_VARIABLES))
    print(model1.predict(some_input))
    print(model2.predict(some_input))
    print(sess.run(model1.trainable_weights))
    print(sess.run(model2.trainable_weights))
    return model1, model2

if __name__ == "__main__":
    model1, model2 = main()