Integrated Gradients Notebook - np.ndarray kwargs but only tf.Tensor supported

[RobertSamoilescu]

This PR fixes issue #527.
The newer versions of transformers do not support np.ndarray for optional arguments (i.e., attention_mask).
The error is fixed by casting np.darray to tf.Tensor before passing to explain or forward methods.

# the values of the kwargs have to be `tf.Tensor`. 
# see transformers issue #14404: https://github.com/huggingface/transformers/issues/14404
kwargs = {k: tf.constant(v) for k,v in z_test_sample.items() if k == 'attention_mask'}

In addition, I included import matplotlib.cm as it is required for matplotlib >= 3.4.2.

Also used a smaller validation dataset as when using the full testing dataset I ran into GPU memory issues.

    # using the entire testing dataset might result in memory issues when running on GPU
    model_out.fit(train_embbedings, y_train, 
                  validation_data=(test_embbedings[:100], y_test[:100]),
                  epochs=epochs, 
                  batch_size=batch_size,
                  callbacks=[cp_callback],
                  verbose=1)

[review-notebook-app]

Check out this pull request on 

See visual diffs & provide feedback on Jupyter Notebooks.

---

Powered by ReviewNB

[codecov]

Codecov Report

Merging #528 (cc9522c) into master (6100135) will not change coverage.
The diff coverage is n/a.

@@           Coverage Diff           @@
##           master     #528   +/-   ##
=======================================
  Coverage   82.34%   82.34%           
=======================================
  Files          76       76           
  Lines       10334    10334           
=======================================
  Hits         8510     8510           
  Misses       1824     1824