edublancas / dstools Goto Github PK

See https://github.com/TomAugspurger/engarde

skpipeline tries to save feature importances if the model has them, do the same with coefficients

pipeline started, ended. train started, ended (maybe just training time)

When using Pipeline, the user sends a dictionary in the load method, the keys for this dictionary are arbitrary and this does not cause any trouble.

But when using SKExperiment and trying to train an experiment, the MetaEstimator assumes certain naming conventions for the keys. I should make this explicit. Probably raising and exception in skpipeline.load when data does not have the appropiate keys

I should probably take out the Pipeline, Experiment and Record classes to a new repo and leave this one as just a bunch of tools for Data Science.

Validate that every necessary function was provided to the pipeline instance. Either adding them to the constructor (but it´s going to make it to verbose) or check them at the top of __call__. This will help the user detect errors early.

if a pipeline is created with workers > 1 and tinydb backend, raise an exception since tiny db does not support concurrent 'connections'.