Crunch 3 submission format

For crunch 3, can my infer just be reading a saved data frame of ranked genes ?

As long as you still include your train function which we can (re)run on unseen data and it does not crash.

I guess you were inspired by what Tarandros did in the quickstarter notebook?

The Crunch 3 logFC Quickstarter notebook says:

During the CrunchDAO submission, there is no need to run the train function. Simply, use the infer function loading the locally saved gene_ranking.csv file.

That does not invalidate the fact that the training code is still necessary.
The infer function can indeed just read what the train outputted.