Implement forecast decomposition for SMA-based models #1180
Conversation
alex-hse-repository
changed the title
|
🚀 Deployed on https://deploy-preview-1180--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 @@
## master #1180 +/- ##
===========================================
+ Coverage 69.13% 87.64% +18.51%
===========================================
Files 177 177
Lines 10380 10394 +14
===========================================
+ Hits 7176 9110 +1934
+ Misses 3204 1284 -1920
... and 66 files with indirect coverage changes 📣 We’re building smart automated test selection to slash your CI/CD build times. Learn more |
etna/models/seasonal_ma.py
Outdated
| ts.df.loc[:, pd.IndexSlice[:, "target"]] = y_pred | ||
|
|
||
| if return_components: | ||
| df.loc[-prediction_size:, pd.IndexSlice[:, "target"]] = y_pred |
There was a problem hiding this comment.
We still use df.loc[-prediction_size:, pd.IndexSlice[:, "target"]], this probably won't work on all pandas versions (we have error in testing under different pandas versions).
The :prediction_size implies that first index is integer, but it isn't.
There was a problem hiding this comment.
Sorry, missed this place, you are right. Fixed it
There was a problem hiding this comment.
May be we should make it more clear why in forecast we add y_pred into df. Because autoregression logic lies within def _forecast and isn't present on level of def forecast.
There was a problem hiding this comment.
Or we should make it more clear what kind of df we are going to use inside _predict_components. We require that it contains lags that was used to make a prediction (taking into account auto-regression), it isn't clear.
There was a problem hiding this comment.
Add docstring to _predict_components with description of df
|
|
||
| Notes | ||
| ----- | ||
| This model supports in-sample and out-of-sample prediction decomposition. |
There was a problem hiding this comment.
This model supports in-sample and out-of-sample prediction decomposition.
I'm not sure that it is an implementation detail and should be in Notes.
Prediction components are corresponding target lags with weights of 1/window
May it could be better to write :math:1 / window
Before submitting (must do checklist)
Proposed Changes
Closing issues
closes #1166