How to fix error "object has no attribute '_output_tensor_cache'"?

Question

I am getting this error when fitting a Keras model, defined using its functional API (inheriting from tf.keras.Model):

AttributeError: 'Model' object has no attribute '_output_tensor_cache'

How could I fix it? Here below a minimal piece of code that reproduces it, and then the full stack trace of the error. The error happens when calling Model.fit(), even before it gets into the implementation of Model.__call__().

I am using tensorflow-gpu (1.7.0).

import tensorflow as tf
import numpy as np


# Using Keras functional API
class Model(tf.keras.Model):
    def __init__(self):
        super(Model, self).__init__()
        self.inp = tf.keras.layers.Input(shape=(8,))
        self.fc1 = tf.keras.layers.Dense(32)
        self.fc2 = tf.keras.layers.Dense(10)

    def __call__(self, inputs, trainig=False):
        y = self.inp(inputs)
        y = self.fc1(y)
        y = self.fc2(y)
        return y


if __name__ == '__main__':
    # Just a random dataset, to try out the code
    X = np.random.rand(512, 8)
    y = np.random.randint(0, 9, size=(512,))

    model = Model()

    model.compile(loss=tf.keras.losses.categorical_crossentropy,
                  optimizer=tf.keras.optimizers.Adadelta(),
                  metric=['accuracy'])

    model.fit(X, y, batch_size=64, epochs=1, verbose=2, validation_split=.2)

Error stack trace:

Traceback (most recent call last):
  File "/home/fanta/workspace/wine-quality/minimal.py", line 29, in <module>
    model.fit(X, y, batch_size=64, epochs=1, verbose=2, validation_split=.2)
  File "/home/fanta/.local/python3.5/lib/python3.5/site-packages/tensorflow/python/keras/_impl/keras/engine/training.py", line 1150, in fit
    batch_size=batch_size)
  File "/home/fanta/.local/python3.5/lib/python3.5/site-packages/tensorflow/python/keras/_impl/keras/engine/training.py", line 704, in _standardize_user_data
    self._set_inputs(x)
  File "/home/fanta/.local/python3.5/lib/python3.5/site-packages/tensorflow/python/keras/_impl/keras/engine/training.py", line 880, in _set_inputs
    self._symbolic_set_inputs(inputs, training=training)
  File "/home/fanta/.local/python3.5/lib/python3.5/site-packages/tensorflow/python/keras/_impl/keras/engine/training.py", line 999, in _symbolic_set_inputs
    outputs = self.call(self.inputs[0], training=training)
  File "/home/fanta/.local/python3.5/lib/python3.5/site-packages/tensorflow/python/keras/_impl/keras/engine/network.py", line 631, in call
    if cache_key in self._output_tensor_cache:
AttributeError: 'Model' object has no attribute '_output_tensor_cache'

Process finished with exit code 1

Daniel Möller · Accepted Answer · 2018-04-17 15:06:02Z

1

Anything wrong in creating a regular model?

def createModel():

    inputs = tf.keras.layers.Input(shape=(8,))
    outputs = tf.keras.layers.Dense(32)(inputs)
    outputs = tf.keras.layers.Dense(10)(outputs)

    return tf.keras.Model(inputs,ouptuts)

answered Apr 17, 2018 at 15:06

Daniel Möller

86.8k24 gold badges202 silver badges222 bronze badges

Sign up to request clarification or add additional context in comments.

Comments

MDMoore313 · Accepted Answer · 2018-05-18 14:04:21Z

1

I got this error because used a Conv2D layer first and I didn't declare the input_shape parameter on my first layer. From Keras:

When using this layer as the first layer in a model, provide the keyword argument input_shape (tuple of integers, does not include the sample axis), e.g. input_shape=(128, 128, 3) for 128x128 RGB pictures in data_format="channels_last".

This might not help you, but could point you in the right direction or help someone else.

answered May 18, 2018 at 14:04

MDMoore313

3,3311 gold badge27 silver badges39 bronze badges

Comments

Stan Takt · Accepted Answer · 2018-10-12 05:39:50Z

0

Downgrading packages helped me:

pip install keras==2.0
pip install tensorflow==1.0

answered Oct 12, 2018 at 5:39

Stan Takt

1

Collectives™ on Stack Overflow

How to fix error "object has no attribute '_output_tensor_cache'"?

3 Answers 3

Comments

Comments

Comments

Your Answer

Hot Network Questions

Collectives™ on Stack Overflow

3 Answers 3

Comments

Comments

Comments

Your Answer

Sign up or log in

Post as a guest

Related