Conversation
and assert_not_raises
* Check if no error is raised when fit is called with a different number of features * Check if estimator is reset when fit is called after a partial_fit * Check if error is raised when no of features changes between different partial_fit calls * Check if partial_fit does not overwrite the previous model * Check if classes argument is parsed and validated correctly * Check if clone(estimator).partial_fit == estimator.partial_fit * Check if partial_fit returns self
|
Running it on the same dataset yields "only"
Testing on C-ordered data gives "only" three failures, |
amueller
force-pushed
the
fortran_c_order_test
branch
from
c359d02 to
15169c2
Compare
|
Naturally doesn't find #4507 because that would have been too easy. |
|
(doesn't find #4507 because it requires |
|
By the way, this makes me think we are testing different array layouts for trees in https://github.com/scikit-learn/scikit-learn/blob/master/sklearn/tree/tests/test_tree.py#L741 It may be nice to generalize this test as a common test :) |
|
Adding the strided test would definitely be nice. This PR is not in a great shape as it basically exposes issues with the |
3 participants
Tries to add a common test for #4507 , uses #3907.
Either I'm doing something wrong (I hope so) or this is not looking good.Found my issue.