Fix bug in new NNs with forecasting interval #1108
Conversation
|
🚀 Deployed on https://deploy-preview-1108--etna-docs.netlify.app |
Codecov Report
📣 This organization is not using Codecov’s GitHub App Integration. We recommend you install it so Codecov can continue to function properly for your repositories. Learn more @@ Coverage Diff @@
## inference-v2.1 #1108 +/- ##
=================================================
Coverage ? 86.86%
=================================================
Files ? 164
Lines ? 8954
Branches ? 0
=================================================
Hits ? 7778
Misses ? 1176
Partials ? 0 📣 We’re building smart automated test selection to slash your CI/CD build times. Learn more |
| ), | ||
| ], | ||
| ) | ||
| def test_forecast_in_sample_full_no_target_failed_assertion_error(self, model, transforms, example_tsds): |
There was a problem hiding this comment.
I've seen this pattern while refactoring nn for etna-v2
Why do we duplicate information about assertion raised?
There was a problem hiding this comment.
Can you explain please what exactly do you mean? Naming of the test?
There was a problem hiding this comment.
@to_be_fixed(raises=AssertionError) and test_forecast_in_sample_full_no_target_failed_assertion_error
- assertion_error duplication
- Instead of using test_forecast_in_sample_full_no_target with pytest.mark.xfail
There was a problem hiding this comment.
assertion_errorto distinguish different tests that failed in different ways.to_be_fixedis used to clarify that we want to fix that test in the future and I can write a particular error message to catch inside the test.
|
Please, take a look at the #1087, I explained what I've found about these tests and probable solutions. But I think all of them aren't really good and should be discussed. |
# Conflicts: # CHANGELOG.md
…nough, add check on size of context for DeepBaseModel and test for it
| @@ -93,6 +99,7 @@ def step(self, batch: MLPBatch, *args, **kwargs): # type: ignore | |||
| : | |||
| loss, true_target, prediction_target | |||
| """ | |||
| self._validate_batch(batch) | |||
There was a problem hiding this comment.
It seems we validate twice.
We can call forward pass in step
There was a problem hiding this comment.
As I understand, step isn't called during forecasting, it is called only during training.
There was a problem hiding this comment.
We use step in training_step and validation_step, it is for training.
Before submitting (must do checklist)
Proposed Changes
Change alignment during forecasting.
Closing issues
Closes #1087.