Code quality of submissions

Competition Broad Institute Crunch #3
1

Dear crunchDAO team,

In my submitted notebooks, I took the model used in the infer function from the resources dir via a hardcoded filepath. However, my train function (which I do not use in the submission) writes the models to the model_directory_path (like in the train function provided by the random submission on crunchDAO).

My question: Is it ok if it is like this in my code? Because it means the notebook could not be run like this, the model would have to be accessed differently in the infer function. Do we have to make a conditional statement in the infer function, that checks if there is a model either in the model_directory_path dir or the resources dir, so that the submission can we made with the uploaded model, and the notebook is runnable on its own in full? Or does this not matter all to much?

And the second related question: We had one notebook, and there just exchanged the code for the different submissions. But I realised that I forgot to change the docstring of the train and infer function. Now, all submissions have the same docstring, which was specific to one submission. Is this a problem that has to be fixed, or does it also not really matter?

Thanks for any info!

Best,
Thomas

2

  1. Hardcoded relative paths are fine. However, hardcoded absolute paths are not. I always recommend using the model_directory_path as you will be prepared for any possible changes in the future of CrunchDAO.

  2. We do not do any analysis on the docstrings. So if you copy/paste the infer’s docstring into the train’s docstring, nothing will happen.

    Since it looks like you asked this question for the Crunch #3, remember that the most important part is the REPORT.md, which will not be editable after you submitted.

3

Thanks!

Yes, all the info about the method in in our REPORT.md, and that was updated for each new submission.

And we have relative paths to get the model: ./resources/model

Best,
Thomas