FIX - Improve the error message when the TextEncoder is fitted without installing the additional dependencies #1769
Conversation
|
Hi @fxzhou22, thanks for working on this PR. It's already in a good shape, but I made a suggestion on how to keep all the tests in the main The format checks are failing because of issues on our side, I'll deal with that. |
Hi @rcap107 , thank you for the feedback! It should be my fault, but I don't seem to find your suggestion, would you mind pointing me to where the comment is please? However, I looked again into the
In this way, As a result, I remove also Thank you also for the format checks part. |
replace importorskip inside encoder() to avoid the problem and changes test_missing_import_error with direct creation
That's strange 🤔 I tried to ping you in the comment, can you see it?
I think it's better if we use the monkeypatch, rather than modifying the tests that are already there |
Sorry, I really don't find it... I'm quite new here and this is in fact my first contribution, I apologize for it. I'll look into monkeypatch and undo my changes to already existing tests. |
There was a problem hiding this comment.
This test can be run from the same file using a small monkeypatch like so:
def test_missing_import_error(encoder, monkeypatch):
"""Test that a clear error is raised when sentence_transformers is missing.
We mock the missing dependency by hiding it from sys.modules, then verify
that TextEncoder.fit() raises an ImportError with a helpful message.
"""
monkeypatch.setitem(sys.modules, "sentence_transformers", None)
st = clone(encoder)
x = pd.Series(["oh no"])
err_msg = (
"Missing optional dependency 'sentence_transformers'.*"
"TextEncoder requires sentence-transformers.*"
"install\\.html#deep-learning-dependencies"
)
with pytest.raises(ImportError, match=err_msg):
st.fit(x)By monkeypatching sys.modules we can make it look as if sentence_transformers is missing, and avoid issues with the importorskip at the top.
There was a problem hiding this comment.
Yes, now I get it, thanks a lot!
There was a problem hiding this comment.
This file can be removed with the inclusion of the monkeypatch in the main test file
It was my fault, I did not submit the review, my bad 😅 You should be able to see it now |
Yes, thanks a lot and no worries! I've tried with your suggested code, it seems to me that
In my opinion, the case in the bottom left corner is the situation/test case that we expect, when the user dosen't have I'll thus commit my changes with |
On the CI, the version of the test that uses monkeypatch runs in the configuration that includes sentence_transformers, and removes it to emulate the situation in which the dependency is missing. So in practice it would work just fine to test the import error.
I do like your solution better, since it's still testing the import error in the other configurations. |
|
Looks good to me, thanks a lot @fxzhou22 |
Thanks for the explication and glad that I've helped. It was a good experience and I've learned a lot. Thank you also for the clear documentation that guided me well wtih my first contribution! |
Hello, this PR would:
TextEncoderis used without installing the optionalsentence_transformersdependency.Changes
skrub/_text_encoder.py: Added explicit error message and reference to installation guidetest_text_encoder_missing.py: New test file created withtest_missing_import_errorNote: It seems to me that in original case,
test_text_encoder.pyhaspytest.importorskip("sentence_transformers")which skips the entire file when the package is not installed. It is the reason why I created a new test file for it.test_text_encoder.py: Added documentation and error message updated fortest_missing_import_errorCHANGES.rst: Added changelog entry