How were the splits for the coq files decided per project?

[brando90]

I was curious & wanted to extend the data set size of coq-gym -- ideally automatically to include a lot. Were the files to do the splits decided manually? Or randomly? or maybe Coq was used to create a dag of dependencies and then a topological sort & then some arbitrary split based on the dag was made?

Thanks in advance! :)

[yangky11]

The projects are randomly split into train/val/test. All files within the same project belong to the same split.

[brando90]

ok but you didn't split per file I see! Is that to avoid the complication of the depedency dag thing? @yangky11

[yangky11]

We didn't want a test theorem to be accessible in the environment of a training theorem.