# Oct 11, 2024: aeon Module Meeting (Classification and Regression)
###### tags: `aeon-module-meeting`

#### Oct 11, 2024 @ 12:00PM UTC
#### Previous meeting: N/A

Present: TB MM CH

### Topics

- Bugs
currently four bugs labelled classification, one regression
*#1646 [BUG] BasicMotions results for HC2 not reproducible when numba turned off*
worth testing if this is still true with new mac tests
*#1696 [BUG] Arsenal predict not working properly?!*
not really a bug any more, but can leave it as it describes a threading issue
*#1726 [BUG] HC2 sporadic CI failure*
again, worth testing if this is still the case
*#2179 [BUG] freeze on import of aeon.classification.convolution_based*
came in during the meeting, need to look at it
*#2151 [BUG] MR-Hydra regressor non deterministic*
seems to have gone away https://github.com/aeon-toolkit/aeon/pull/2154

- PRs
there are five open draft classification PRs, two normal PRs for regression, one for coverage (#2175) and #2154. No actions needed

- Issues
17 open classification issues, seven for regression

Implement algorithms: TS-CHIEF (#170), PF2 (#428), disjoint-CNN (#1902), TD-MVDC (#2081),
Bugsx4 (See above),

*#663 [ENH] Interpretability for HIVE-COTE V2 Classifier *

would be nice, big job

*#1155 [ENH] Remove the unused inner types for collections

we agreed to remove nested_univ and df-lst. Tony will do it.

*#1599 [ENH] Tag for capability to making probabilistic predictions for classification*

Agreed in principle, complex in implementation, caused by composite algorithms that call sklearn predict_proba. Currently not really worth the effort to solve.

*#1686 [ENH] standardised naming and structures for shapelet classifiers*

definitely desirable, not spare capacity atm

*#1742 [ENH] Replace SFA with SFAFast in REDCOMETS*

hopefully the author zy18811 will find time to do this.

*#1777[ENH] Add class_weight parameter to classifiers that use sklearn based estimators with testing*

Not sure if this has been fully resolved

#1886 n_jobs/_n_jobs needs tidying up, tag changed because not just classification

*#1952 Add test that the estimator "algorithm_type" is valid*

should be easy, may not have explained well what I want

*#2089 [ENH] Split RocketClassifier and RocketRegressor into multiple classes*

Agreed to split Rockets if matthew wants to

### Future directions

- unequal length capabilities

Issue with numba lists and length
distance use: always convert to numba list, look at that and use this model to convert. Chris H.

- Unequal sampled (transformer)

- HIVE-COTE regression (research)

- possibly add axis parameter and unify with series

### Discussion

discussed current CI slowness, found problem and solved it.

### Actions

Bugs
- [ ] Test if #1646 and #1726 are still bugs
- [ ] Review #2151
- [ ] Read and respond to new bug
- [x] Split rockets
- [ ] Add test for valid estimator tags
- [ ] check if class weights finished, if not list algorithms still to do

### Next meeting date

TBC