Update internal functions to reflect deprecated keras functions

[EmilHvitfeldt]

This PR aims to close #613 and close #596.

With the changes to tensorflow, internal functions have been updated to reflect deprecated functions.

[topepo]

Also, please make an entry in the news file about this.

[EmilHvitfeldt]

This PR tries to handle a couple of different problems.

The first problem is the error emitted by Keras as mentioned in #613 and #596, where predict_classes and predict_proba were changed. There is now a conditional (version(keras) >= 2.6) on the package version of keras to switch between using predict() and predict_proba()

Secondly the use of keras::k_argmax() is conditioned to be used when the tensorflow version is v2.

Lastly, set_tf_seed() has been added, exported and used in tests. This function sets seed in R and tensorflow at the same time, conditionally switching between tensorflow::tf$random$set_seed and tensorflow::tf$random$set_random_seed for v1 and v2 respectively.

I have tested these changes locally using all combinations that could be installed together of

• Keras: 2.7.0, 2.6.1, 2.6.0, 2.4.0
• Tensorflow: 2.7.1, 2.6.3, 1.15.0

[github-actions]

This pull request has been automatically locked. If you believe you have found a related problem, please file a new issue (with a reprex: https://reprex.tidyverse.org) and link to this issue.