search

irllabs / ml-lib

A machine learning library for Max and Pure Data
Other
274 stars 40 forks source link

macOS: ml.dtw (Dynamic Time Warping) crashes pd after a while during "mapping" #157

Closed stc closed 5 years ago

stc commented 5 years ago

sampling, teaching is ok. have to force quit after problem occurs. dtw is working smoothly w/ ofxGRT pd-0.49-1 on macOS mojave.

jamiebullock commented 5 years ago

Hi @stc, which version of ml-lib are you using?

Also, when you say you have to force quit, presumably Pd hangs... i.e. you get a spinning beachball?

stc commented 5 years ago

ml-0.18.0-alpha-OSX64, downloaded from github releases page. more precisely: after training, probability prediction seems to work for a while, then after 10-20 secs, it becomes less reliable, finally it stucks and sometimes the whole UI freezes, it's hard to reproduce freezing, can't figure out what event causing it (no beach ball)

jamiebullock commented 5 years ago

I can reproduce this and am looking into it.

jamiebullock commented 5 years ago

@stc when you map the time series are you delimiting each "gesture" with record 1 record 0 ?

I think the behaviour you described occurs when record is left "on". This leads to the DTW attempting to classifier an increasingly larger and larger input matrix as more data is added with map

The correct process is:

record 1 map... map... map... record 0 record 1 map... map... map... record 0 etc

stc commented 5 years ago

Indeed it seems that the order of record and other messages were sometimes not in the right order. Thanks for looking into it