Remove TF1-based examples (eva.py, exploitability_descent.py)

[pantha704]

Summary

Removes TF1-based examples from the open_spiel/python/examples/ directory as requested by @lanctot in #1434.

Changes

• Deleted eva.py - uses tensorflow.compat.v1
• Deleted exploitability_descent.py - uses tf.disable_v2_behavior()

Rationale

Per maintainer guidance in #1434:

"Any examples based on TF1 code still lingering in the examples directory (eva, exploitability_descent) should be removed."

A PyTorch version of EVA exists at open_spiel/python/pytorch/eva.py, so the TF1 example removal is safe.

Verification

• Verified no other files import these examples via grep
• Clean deletion of 194 lines

Addresses #1434 (partial)

[alexunderch]

why only these two....

[lanctot]

why only these two....

Agreed. Probably time to drop this whole directory:

https://github.com/google-deepmind/open_spiel/tree/master/open_spiel/python/algorithms/tf

Can we add it to this PR?

[alexunderch]

yes, let's get rid of that. I won't back to eva any time soon, almost everything else we have in pytorch/jax or I can re-add it, nothing urgent atp.

then, need to get rid of

open_spiel/open_spiel/scripts/python_extra_deps.sh

Line 57 in 09b2873

export OPEN_SPIEL_PYTHON_TENSORFLOW_DEPS="numpy==1.26.1 tensorflow==2.14.0 tensorflow-probability==0.22.1 tensorflow_datasets==4.9.7 keras==2.14.0"

and https://github.com/google-deepmind/open_spiel/blob/master/open_spiel/scripts/find_tensorflow.sh

I guess that's it!

you may feel sceptical about that, because some components related to tf may be re-added because, per se, of dataset processing with grain, but it would be fine to perform these things with jax stack, and not reintroduce new dependency line

[lanctot]

Oh yes, good point!

find_tensorflow.sh can go for sure.

I do believe tensorflow_datasets may still be used somewhere (maybe also tensorflow_probability) but I don't remember exactly. Maybe we can try removing all of those lines and see what happens, and just add back in the ones we still need.

[alexunderch]

Yes, let's try, I tried to remove those deps wherever I could. But still could miss smth. tensorflow_probability is being installed with distrax in the jax stack, so should be all good there.

[alexunderch]

@lanctot, I will aslo need to delete some other tensorflow things that don't break the tests but stay redundant because they were need to be compatible with tf.Sess and models after I am done with refactoring. Should I create a separate PR or we can concentrate on deletion here?

[pantha704]

Got it! Since @alexunderch was already working on the broader TF cleanup and refactoring, I'll leave the
algorithms/tf/ directory removal to them to avoid duplicating effort.
Happy to have this PR merged as-is for the examples cleanup, or closed if you'd prefer to consolidate everything in one PR. Let me know what works best!