[BUG] HiveCote output

Describe the bug Fitting hive-cote model gives unrecognizable number outputs and doesn’t finish executing, even on a small dataset.

To Reproduce Run fit with a small dataset (mine is 500x10), I interrupted after 30 hours on this small dataset.

>from sktime.classification.hybrid import HIVECOTEV1

>hive = HIVECOTEV1()
>hive.fit(X_train_small, y_train_small)

1.4102227687835693
1.4351227283477783
1.4401609897613525
1.455118179321289
1.4720587730407715
1.4735655784606934
1.5098981857299805
1.5518741607666016
1.5618183612823486
1.6083710193634033
1.6144781112670898
1.656531572341919
1.694200038909912
1.8179378509521484
1.1604790687561035
1.2566297054290771
1.267608404159546
1.2775826454162598
1.2984962463378906
1.6186723709106445
1.684495449066162
1.1329700946807861
1.162933588027954
1.1708612442016602
1.1918137073516846
1.2217142581939697
1.2486572265625
1.2994804382324219
1.3000919818878174
1.3204693794250488
1.3333721160888672
1.424189805984497
1.4730608463287354
1.5324997901916504
1.6767830848693848
2.2469911575317383
1.10504150390625
1.1977901458740234
1.2267138957977295
1.2467164993286133
1.250666618347168
1.3663420677185059
1.3813018798828125
1.1080362796783447
1.119640827178955
1.2005653381347656
1.2197377681732178
1.2207260131835938
1.2436511516571045
1.2863152027130127
1.1598975658416748
1.173907995223999
1.182831048965454
1.223266363143921
1.2944936752319336
1.3483796119689941
1.3493454456329346
1.225809097290039
1.2561888694763184
1.2595922946929932
1.2845253944396973
1.3055100440979004
1.306506633758545
1.3862872123718262
1.4042775630950928
1.4251701831817627
1.8021807670593262
1.153918981552124
1.1748120784759521
1.1948018074035645
1.2117547988891602
1.2277166843414307
1.2336978912353516
1.2676081657409668
1.267608880996704
1.271592140197754
1.146115779876709
1.2246768474578857
1.2275745868682861
1.2276766300201416
1.2586297988891602
1.2646138668060303
1.312474250793457
1.1439476013183594
1.2127492427825928
1.2346980571746826
1.2396836280822754
1.2486522197723389
1.248659372329712
1.2735872268676758
1.2815752029418945
1.0818612575531006
1.2257189750671387
1.2336993217468262
1.2556405067443848
1.2866203784942627
1.2915489673614502
1.3842966556549072
1.4561092853546143
4.397695541381836
4.439384460449219
4.521897077560425
1.4102232456207275
1.4172465801239014
1.4710273742675781
1.5079617500305176
1.5219314098358154
1.5228848457336426
1.5463685989379883
1.5488560199737549
1.5538420677185059
1.5658457279205322
1.5777888298034668
1.0661485195159912
1.0791501998901367
1.1140198707580566
1.1509478092193604
1.1888630390167236
1.1957767009735107
1.2287139892578125
1.2297098636627197
1.2337005138397217
1.2407174110412598
1.27756667137146
1.2776179313659668
1.2795777320861816
1.2805571556091309
1.1738529205322266
1.207770824432373
1.2187480926513672
1.2227346897125244
1.2317297458648682
1.2686076164245605
1.2725942134857178
1.283567190170288
1.288553237915039
1.2236785888671875
1.250654935836792
1.2656142711639404
1.2686102390289307
1.2745912075042725
1.2925684452056885
1.3703694343566895
1.141946792602539
1.1509218215942383
1.1798830032348633
1.217745065689087
1.2371776103973389
1.2376549243927002
1.2377171516418457
1.2726309299468994
1.295534610748291
1.298499345779419
1.205770492553711
1.2506554126739502
1.2636210918426514
1.2875206470489502
1.297529935836792
1.1938278675079346
1.1987836360931396
1.2456707954406738
1.246654748916626
1.260629653930664
1.2665889263153076
1.2766263484954834
1.2795774936676025
1.2795794010162354
1.1908149719238281
1.191812515258789
1.2007882595062256
1.2057664394378662
1.2562260627746582
1.2586488723754883
1.259627103805542
1.2735936641693115
1.2835674285888672
1.1528823375701904
1.16489577293396
1.2047667503356934
1.2137830257415771
1.2206969261169434
1.2227301597595215
1.276543378829956
1.3703701496124268
1.154890537261963
1.1579084396362305
1.169846773147583
1.2337028980255127
1.2346899509429932
1.23968505859375
1.2566413879394531
1.2586708068847656
1.1200060844421387
1.1329705715179443
1.1868271827697754
1.2107627391815186
1.2247257232666016
1.2307095527648926
1.2386882305145264
1.2456769943237305
1.7991812229156494
1.8540723323822021
2.4433372020721436
1.5897419452667236
1.642646312713623
1.7273809909820557
2.219067096710205
1.108036756515503
1.1898179054260254
1.1987824440002441
1.206784963607788
1.2157487869262695
1.2247209548950195
1.2386524677276611
1.2634143829345703
1.3773200511932373
1.3872885704040527
1.3922829627990723
1.164886713027954
1.254988193511963
1.2556400299072266
1.2715990543365479
1.3055362701416016
1.3185110092163086
1.3523328304290771
1.3623898029327393
1.0801029205322266
1.1274166107177734
1.191850185394287
1.2496368885040283
1.265561580657959
1.3234786987304688
1.3713512420654297
1.3812839984893799
1.3863086700439453
1.3932538032531738
1.4052231311798096
1.13299560546875
1.2047674655914307
1.2396745681762695
1.2576580047607422
1.3164677619934082
1.36539888381958
1.453104019165039
1.4697601795196533
1.680494785308838
1.1129837036132812
1.1220283508300781
1.1429426670074463
1.18086576461792
1.190821886062622
1.1928248405456543
1.201784610748291
1.2247192859649658
1.2257499694824219
1.2277185916900635
1.3833296298980713
1.458068609237671
1.5454039573669434
2.000641107559204
1.1140172481536865
1.231666088104248
1.3254904747009277
1.360323190689087
1.4071955680847168
1.5943021774291992
1.6433846950531006
1.928835391998291
---------------------------------------------------------------------------
KeyboardInterrupt                         Traceback (most recent call last)
<timed exec> in <module>

Expected behavior An understandable output and a sense of progress, e.g. passing a verbose=True argument.

Further details: I checked the code of the class too, but I don’t understand where the outputs come from, since its simply fitting 4 different algos (Shapelettransform, TimeSeriesForest, RISE, cBOSS) and saving the weights. I can’t see when the code terminates.

Versions sklearn: 0.23.1

from sktime import show_versions; show_versions()

System:
    python: 3.8.5 
executable: C:\Users\
   machine: Windows-10

Python dependencies:
          pip: 21.0.1
   setuptools: 50.3.0.post20201006
      sklearn: 0.23.1
       sktime: 0.5.3
  statsmodels: 0.12.2
        numpy: 1.19.5
        scipy: 1.4.1
       Cython: 0.29.21
       pandas: 1.0.5
   matplotlib: 3.2.2
       joblib: 0.17.0
        numba: 0.52.0
     pmdarima: None
      tsfresh: None

Thanks for the great repo!

[MatthewMiddlehurst]

Yeah that really should not have been outside contrib! I have made a number of fixes and additions to the HC1 class. I’m happy to keep it where it is with these changes.

I have included a verbose setting to give periodic progress updates as suggested, as well as a n_jobs parameter to make use of multiple threads where applicable (mainly the cross-validation). There was a recursive call to fit embarrassingly, so it should no longer take 30 hours (or forever) to fit.

Unit testing remains an issue, I have set it up so you can alter the parameters of each individual classifier which should help, but I think its going to be a costly one even with that. Any suggestions welcome here or on the PR.

[mloning]

Thanks @TonyBagnall for the quick reply!

@MJimitater if you’re interested in looking into this to find out what has gone wrong, that’d be much appreciated too!