Last updated
Last updated
Install pylift:
To start, you simply need a pandas.DataFrame
with a treatment column of 0s and 1s (0 for control, 1 for test) and a numeric outcome column. Implementation can be as simple as follows:
up.fit()
can also be passed a flag productionize=True
, which when True
will create a productionizable model trained over the entire data set, stored in self.model_final
(though it is contentious whether it's safe to use a model that has not been evaluated in production -- if you have enough data, it may be prudent not to). This can then be pickled with self.model_final.to_pickle(PATH)
, as usually done with sklearn
-style models.
A fully-fledged example of pylift in action can be found in pylift/examples/simulated_data/sample.ipynb
.