Metadata-Version: 2.1
Name: pamalia
Version: 0.0.1
Summary: UNKNOWN
Home-page: https://github.com/louisabraham/pamalia
Author: Louis Abraham
Author-email: louis.abraham@yahoo.fr
License: MIT
Platform: UNKNOWN
Requires-Python: >=3.6
Requires-Dist: sklearn

pamalia
=======

TODO
----

-  parallel evaluation with joblib (``n_jobs``)

-  implement genetic algorithm to find pipelines

-  copy parameters from tpot

-  add rules to prevent stupid things (PolynomialFeatures with many
   columns)

-  distribute genetic algorithms with dask

-  test joblib distributed backend with dask (nothing to do, just test)

-  fine-grained distribution with dask computation graph:

   -  trivial for prediction

   -  for fit, each step returns a cross-validated estimate and a fitted
      model. The fitted model is not used before the final step.

-  it is possible to implement cross-validation with a factor 2
   improvement when the cross-val and the training folds match

-  handle timeouts: https://github.com/dask/distributed/issues/391,
   https://github.com/dask/dask/issues/1183


