Refitting model_final and nuisance averaging
#360
Conversation
vsyrgkanis
force-pushed
the
vasilis/refit
branch
5 times, most recently
from
acd4d11 to
39634db
Compare
fixed small issue with _d_t storage enabled refitting in drlearner enabled refit in ortho_iv enabled monte_carlo_iterations in ortho_iv added cache_values param to ortho_iv fixed docstrings to remove the model parameters and added dosctrings to the abstract methods. Removed doctest examples from ortholearner and rlearner as these classes can no longer be used as standalone and are abstract classes
vsyrgkanis
force-pushed
the
vasilis/refit
branch
from
39634db to
cf7122a
Compare
vsyrgkanis
force-pushed
the
vasilis/refit
branch
3 times, most recently
from
b315e2e to
cc2ae24
Compare
linting
fixed failing tests
fixed notebooks
fixed ortholearner docstring
reverted to allowing n_crossfit_splits and raising deprecation warning. fixed ortholearner doctest
fixed bugs for failing tests
fixed failing test bugs
added extra kwargs to _strata
old sklearn crashes with new scipy. we can revert once we enforce new sklearn
disallowing bootstrap inference when refitting
added many refit tests. fixed some leftover bugs based on new tests
typo in tests
changed monte_carlo_iterations to mc_iters and added mc_agg parameter to change the aggregation method {mean, median}
fixed nuisance aggregation when nuisances have different dimensions
removed _models_nuisance initilaization at init.
added rlearner residuals_ property
fixed flaky cate interpreter test
vsyrgkanis
force-pushed
the
vasilis/refit
branch
from
cc2ae24 to
ba0dbf0
Compare
… the final stage is estimated.
econml/dml.py
Outdated
|
|
||
| @property | ||
| def bias_part_of_coef(self): | ||
| return self.rlearner_model_final._fit_cate_intercept |
There was a problem hiding this comment.
Same here, why we need this attribute if it's same as fit_cate_intercept ? Another corner case I think we are missing now is from the parse_final_model_params, what if the user sets fit_cate_intercept=False but input the final model with fit_intercept=True, we should update our fit_cate_intercept depends on whether there is an intercept in the user defined final model?
There was a problem hiding this comment.
fit_cate_intercept is different than the intercept of the final model. The fit_cate_intercept is aobut whether we augment X with an constant of 1's. The fit intercept of the final model is about fitting an extr aoffset in the regresstion not a cate offset.
There was a problem hiding this comment.
Also observe that bias_part_of_coef, returns the fit_cate_intercept based on the fit time. See the implementation in of bias_part_of_coef in dml. It doesn't read the fit_cate_intercept, but it reads the parameter that was stored when fit was called.
So there wont be an inconsitency here. At least that was a corner case that I tried to address.
Co-authored-by: Keith Battocchi <kebatt@microsoft.com>
Support refitting only final model in DML after changing estimator parameters
Add support for monte carlo nuisance estimation, with multiple k-fold draws.
added rlearner residuals_ property that returns fitted residuals for training data (fixes plotting the residuals T & Y from the model #350 )
fixed flaky cate interpreter test
added refit example in the dml notebook