Improve best batch and random forest classes #29
Conversation
AldoGl
force-pushed
the
improve_BB_RF_classes
branch
from
9734f0f to
ce6c690
Compare
Codecov Report
Additional details and impacted files@@ Coverage Diff @@
## main #29 +/- ##
==========================================
+ Coverage 97.72% 97.74% +0.01%
==========================================
Files 23 23
Lines 1230 1240 +10
==========================================
+ Hits 1202 1212 +10
Misses 28 28
|
| self.a = a | ||
| self.b = b | ||
| self.perturbation_range = perturbation_range |
There was a problem hiding this comment.
It is OK for me. To make the new feature perfect, I would add some value checks on a, b, and perturbation_range. In particular:
a > 0.0b > 0.0perturbation_range > 1
There was a problem hiding this comment.
Thanks, it makes sense. Probably perturbation_range should be greater or equal to 1
There was a problem hiding this comment.
Thanks, it makes sense. Probablu
perturbation_rangeshould be greater or equal to 1
If perturbation_range = 1, integers(1, self.perturbation_range) gives error:
np.random.default_rng().integers(1, 1) # raises ValueError: low >= high
| """ | ||
| super().__init__(batch_size, random_state, max_deduplication_passes) | ||
|
|
||
| self._n_estimators = n_estimators | ||
| self._criterion = criterion | ||
| self._n_classes = n_classes |
There was a problem hiding this comment.
Given the implementation of prepare_data_for_classifier, n_classes should be at least X, where at the bare minimum 1 should not break the code, but ideally we could require more bins, e.g. at least 2?
AldoGl
force-pushed
the
improve_BB_RF_classes
branch
from
ce6c690 to
cba47fd
Compare
|
Thank you @marcofavoritobi for your feedback. I implemented the checks you mentioned |
Proposed changes
I generalise the best batch constructor by letting it take the
aandbparameter of the beta-binomial distribution as well as theperturbation_rangeapplied to the parameter values.I generalise the random forest constructor by letting it take the number of classes (
n_classes) used by the classifier in the fitting phase.