mmvec
mmvec copied to clipboard
biocore
Neural networks for microbe-metabolite interaction analysis
This is the most immediate struggle Right now, `--i-microbes` and `--i-metabolites` may be a bit too restrictive, since it indicates that it is limited to only microbes and metabolites. We...
This function is currently converting sparse bioms to a dense representation. https://github.com/biocore/rhapsody/blob/master/rhapsody/util.py#L111 This maybe problematic when scaling these algorithms to very large datasets.
It maybe advantageous to have a way to stop fitting after the model has reached convergence.
Pytorch is proving to be much much easier to debug. This will be key, especially when enabling Bayesian inference via variational inference.
It may make sense to decouple the creation of the OrdinationResults object from the model fitting. This is advantageous for the following reasons 1. This help with software modularity 2....
If the input tables are sorted / clustered beforehand -- the algorithm will drastically underperform. There needs to be some scrambling beforehand -- preferably at each iteration. See torch dataloader:...
Maybe worthwhile to add in high level commands to enable hyper-parameter tuning At the very least, provide a tutorial on how to do this. https://github.com/fmfn/BayesianOptimization
There are currently a bunch of files deposited, and it isn't clear how these files can be used. It'll be easier for the standalone CLI to deposit a OrdinationResults object...
Right now, our optimization scheme is pretty dumb - its just iterating through the microbial reads one at a time. This can be drastically speed up (likely 100-1000x faster) if...