Training is too slow

[abdalazizrashid]

So, I'm trying to reproduce the paper results, but the training is too slow; it's taking more than 4 for a single problem on hours on Google Colab machines. I forked the repo and did some refactoring and fix some of the problems, but still there so much more to be done, but the main obstacle is the training time. Also, the reported solving time in the paper is at most few thousand seconds can you please elaborate on that? Please take a look at this notebook where I attempt to reproduce the results.

[lacava]

may be related to https://github.com/SJ001/AI-Feynman/issues/40