0

class BinaryTruePositives(tf.keras.metrics.Metric):

def __init__(self, name='Results', **kwargs):
     super(BinaryTruePositives, self).__init__(name=name, **kwargs)
     self.true_positives = self.add_weight(name='tp', initializer='zeros')

def update_state(self, y_true, y_pred, sample_weight=None):

    y_true=tf.reshape(y_true, [16, 4])

    y_true=tf.cast(y_true, dtype=tf.float32)

    y_true = tf.squeeze(y_true)

    print("before pred")

    tf.print(y_pred)

    y_pred = tf.sign(y_pred)
    print("pred after sign")
    tf.print(y_pred)
    print("true")
    tf.print(y_true)

    equal_t = tf.equal(y_true, y_pred)

    reduce_t = tf.reduce_all(equal_t, axis=1)

    tf.print(reduce_t)

    z=tf.keras.backend.mean(reduce_t)

    tf.print(z)

    self.true_positives.assign_add(z)


def result(self):
    return self.true_positives

def reset_states(self):
    self.true_positives.assign(0.)

During the execution of model training, it throws an error of shape. Even though I 
reshape y_true. 

Batch_size=16
InvalidArgumentError:  [_Derived_]  Incompatible shapes: [64] vs. 
[16,4]
[[{{node

StatefulPartitionedCall/StatefulPartitionedCall/mean_squared_error/Sq ua redDifference}}]] [[StatefulPartitionedCall]] [[import/StatefulPartitionedCall_1/ReduceDataset]] [Op:__inference_wrapped_function_48795] [[{{node Function call stack: wrapped_function -> wrapped_function -> wrapped_function

classes are defined as [1, 1, 1, 1] for 0 class, [-1, -1, -1, -1] for 1 class, [1,-1, 1, -1] for 3 class

y_pred is [16,4] shape, but even though after reshaping y_true, it works for the first batch_size then displays an error of incompatible shapes

Amandeep Singh
  • 41
  • 5
  • try to find on which like `InvalidArgumentError` is thrown – Alberto Sinigaglia Aug 17 '22 at 16:25
  • it is just like InvalidArgumentError: [_Derived_] Incompatible shapes: [32] vs. [8,4] [[{{node StatefulPartitionedCall/StatefulPartitionedCall/mean_squared_error/SquaredDifference}}]] [[StatefulPartitionedCall]] [[import/StatefulPartitionedCall_1/ReduceDataset]] [Op:__inference_wrapped_function_48795] – Amandeep Singh Aug 17 '22 at 16:31
  • sorry, idk why i wrote "like", i meant "line" – Alberto Sinigaglia Aug 17 '22 at 16:55
  • Would it be right to think y_pred has shape (64,)? If it's that shape you may need to reshape y_true as (64,) rather than as (16, 4) as you're doing in the code you posted – David Harris Aug 17 '22 at 19:09
  • i have y_pred in shape [16,4], but on reshape y_true. it gives an output for first batch but then displays an error of shape – Amandeep Singh Aug 17 '22 at 20:16
  • If it runs on the first batch and then fails, it suggests there is something that is causing function tracing to fail. You may find it will work if you set the option tf.config.run_functions_eagerly(True) before running. But this is not a real solution - it will be slow, even if it does work. I'm unsure what is causing the problem - possibly you should not try to change the value of y_true within the function? You could try different variable names, but I am afraid I'm guessing a bit here. – David Harris Aug 17 '22 at 21:10

0 Answers0