[MNT] Probe statsmodels 0.15 with sktime-next

[amotl]

About

Probe statsmodels 0.15 with a patch to sktime by @aryamanDutta. Thank you!
Probe statsmodels 0.15 with sktime@main.

Details

Another automatic update failed CI.

TypeError: ETSResults.simulate() got an unexpected keyword argument 'random_state'
TypeError: ccf() got an unexpected keyword argument 'unbiased'

References

• [BUG] investigate and create a compatibility fix for statsmodels 1.15 failures #102
• [MNT] [Dependabot](deps): Update statsmodels requirement from <0.15.0,>=0.12.1 to >=0.12.1,<0.16.0 #110
• [BUG] _StatsModelsAdapter passes random_state to statsmodels get_prediction, breaking predict_interval on statsmodels 0.15 sktime#10968
• [BUG] Fix random_state handling in statsmodels adapter sktime#10972

[amotl]

We've reported @aryamanDutta's patch works well, according to the outcome on this project's CI. Thank you again.

• [BUG] Fix random_state handling in statsmodels adapter sktime#10972 (comment)

[fkiraly]

Ok, I understand this was for testing, right?

So we just need to wait for the new sktime release, and the problem will fix itself, correct?

[amotl]

Ok, I understand this was for testing, right?

Exactly. It was about validating the upstream patch by @aryamanDutta.

So we just need to wait for the new sktime release, and the problem will fix itself, correct?

Right. First, this patch needs to converge, and afterwards we need a new release.

• [BUG] Fix random_state handling in statsmodels adapter sktime#10972

[amotl]

Let me close this again. GH-110 will be sufficient to resolve this.

[amotl]

@fkiraly says the flaw was already fixed, so I am now validating against sktime@main per 25541ff.

[fkiraly]

hm, looks like it is failing still. Strange.
I am sure there were ets related failures, but I thought they were fixed.

Could you open an issue with the exact call?
And the patch that fixes it?

[amotl]

I am sure there were ets related failures, but I thought they were fixed.

I think it is about a different topic here.

Could you open an issue with the exact call? And the patch that fixes it?

I believe it is all done by @siddharth7113 and @aryamanDutta already. When using the patch like before 25541ff, the integration tests succeed. 💯

• Issue: [BUG] _StatsModelsAdapter passes random_state to statsmodels get_prediction, breaking predict_interval on statsmodels 0.15 sktime#10968
• Patch: [BUG] Fix random_state handling in statsmodels adapter sktime#10972