Add causal analysis solution #447
Conversation
kbattocchi
force-pushed
the
kebatt/causal_analysis
branch
7 times, most recently
from
7471971 to
9043b97
Compare
kbattocchi
force-pushed
the
kebatt/causal_analysis
branch
from
9043b97 to
32d321e
Compare
kbattocchi
force-pushed
the
kebatt/causal_analysis
branch
from
32d321e to
da0ff45
Compare
kbattocchi
force-pushed
the
kebatt/causal_analysis
branch
from
da0ff45 to
98b09af
Compare
|
|
||
| return insights, result | ||
|
|
||
| if self.feature_names is None: |
There was a problem hiding this comment.
Is this logic maybe in some sklearn utility
There was a problem hiding this comment.
If we are in a warm start why are we not using the same feature names as in the first time.
We could be getting some weirdness here if user passes once a numpy array and then a pandas data frame with warm start
kbattocchi
force-pushed
the
kebatt/causal_analysis
branch
from
98b09af to
9dc967b
Compare
| categorical: array-like of int, str, or bool | ||
| The features which are categorical in nature, expressed as either column indices, | ||
| column names, or boolean flags indicating which columns to pick | ||
| heterogeneity_inds: array-like of int, str, or bool, or None or list of array-like elements or None, default None |
There was a problem hiding this comment.
To avoid confusion I think we should make the default of heterogneity inds be 'all', which means use all variables, and None or empty list, should mean use no variables for heterogeneity.
There was a problem hiding this comment.
if it's an empty list, forest option will not work? it requires X? and also all the local view will not work neither
There was a problem hiding this comment.
I have made sure that empty h_inds now work with linear final models (even with local effects, although there's no point since there are no features to cause variation so all samples have the same results); I have not changed the spec of the method (e.g. enabling 'all') because that seems too risky at this stage for this release. As Maggie notes, empty h_inds will not work with forest dml and I don't see any easy path to making it work.
There was a problem hiding this comment.
Sounds good. Let's just elaborate a bit more in the docstring that to use no features for heterogneiety use the empty list.
| feature_names: list of str, default None | ||
| The names for all of the features in the data. Not necessary if the input will be a dataframe. | ||
| If None and the input is a plain numpy array, generated feature names will be ['X1', 'X2', ...]. | ||
| upper_bound_on_cat_expansion: int, default 5 |
There was a problem hiding this comment.
this arg seems not being applied to any place?
There was a problem hiding this comment.
Yeah, there's a TODO later noting that it's not yet implemented but should be.
I might even do the same uniformization even for local effects and maintain the feature_index inner index with the single feature name that we run the analysis for. |
kbattocchi
force-pushed
the
kebatt/causal_analysis
branch
from
c6f5f1e to
878b54d
Compare
kbattocchi
force-pushed
the
kebatt/causal_analysis
branch
from
878b54d to
6d4cb76
Compare
No description provided.