gisaia / ARLAS-proc

Workaround about data ingestion with computing frameworks
Apache License 2.0
4 stars 0 forks source link

HMM simple windowing without overlapping #72

Closed laurent-thiebaud-gisaia closed 5 years ago

laurent-thiebaud-gisaia commented 5 years ago

In order to improve performances. HMM are at their best with about 5000 points.

laurent-thiebaud-gisaia commented 5 years ago

Test "WithArlasMovingState transformation" should " compute the moving state of a dataframe's timeseries using windowing" result is not the same as without windowing. To be checked with Willi.

laurent-thiebaud-gisaia commented 5 years ago

Chcked with Willi: with too few points, windowing is relevant. I made the window bigger in the unit test to make it work.

laurent-thiebaud-gisaia commented 5 years ago

@sfalquier it can be reviewed now (no WIP anymore)