Stuck on an issue?

Lightrun Answers was designed to reduce the constant googling that comes with debugging 3rd party libraries. It collects links to all the places you might be looking at while hunting down a tough bug.

And, if you’re still stuck at the end, we’re happy to hop on a call to see how we can help out.

Issue with running TF 2.1.0 with `validation_data` tuple present

See original GitHub issue

Weights and Biases version: wandb, version 0.8.28
Python version: Python 3.7.2
Operating System: Darwin

Description

I am not able to train TF models when passing validation_data on TF 2.1.0.

What I Did

import tensorflow_core as tf
from wandb.keras import WandbCallback
import wandb

wandb.init(
    project="test",
)

model = tf.keras.models.Sequential()
...
 model.fit(
                x=train_features,
                y=train_values,
                validation_data=(test_features, test_values),
                epochs=5,
                callbacks=[WandbCallback()],
                verbose=2
)

Results in:

  File "[redacted project file]", line 57, in run_training
    self.sequential = self._train_network(train_features, test_features, train_values, test_values)
  File "[redacted project file]", line 128, in _train_network
    verbose=2
  File "[redacted project loc]/venv/lib/python3.7/site-packages/tensorflow_core/python/keras/engine/training.py", line 819, in fit
    use_multiprocessing=use_multiprocessing)
  File "[redacted project loc]/venv/lib/python3.7/site-packages/wandb/keras/__init__.py", line 111, in new_v2
    set_wandb_attrs(cbk, val_data)
  File "[redacted project loc]/venv/lib/python3.7/site-packages/wandb/keras/__init__.py", line 79, in set_wandb_attrs
    elif isinstance(val_data, tuple) and isinstance(val_data[0], tf.Tensor):
AttributeError: module 'tensorflow' has no attribute 'Tensor'

Note 1: I am using tensorflow==2.1.0.

Note 2: The issue does not occur if I remove validation_data