e61955b5553d98b4346817e0e9161f79fb23a40c,sonnet/python/modules/nets/mlp_test.py,MLPTest,testTranspose,#MLPTest#Any#Any#,185

Before Change

      self.assertEqual(mlp_transpose.layers[i].output_size,
                       mlp.layers[-1 - i].input_shape[1])

    data = np.random.rand(self.batch_size, self.input_size)
    init = tf.global_variables_initializer()
    with self.test_session() as sess:
      sess.run(init)
      sess.run(mlp_transposed_output, feed_dict={input_to_mlp: data})

    variables = mlp_transpose.get_variables()

    if use_bias:
      self.assertEqual(len(variables), len(self.output_sizes) * 2)

After Change

    self.assertEqual(mlp_transpose.use_bias, mlp.use_bias)
    self.assertEqual(mlp_transpose.activate_final, mlp.activate_final)

    if not tf.executing_eagerly():
      if activate_final:
        self.assertEqual(mlp_transposed_output.op.type, "Relu")
      elif use_bias:
        self.assertEqual(mlp_transposed_output.op.type, "Add")
      else:
        self.assertEqual(mlp_transposed_output.op.type, "MatMul")

    for i in range(0, len(mlp.layers)):
      self.assertEqual(mlp_transpose.layers[i].output_size,
                       mlp.layers[-1 - i].input_shape[1])

Instances