search

cesium-ml / cesium

Machine Learning Time-Series Platform
Other
671 stars 101 forks source link

Remove Celery, use dask.multiprocessing in featurization #181

Closed bnaul closed 8 years ago

bnaul commented 8 years ago

Wait to merge til the corresponding PR in cesium_web is ready.

Fixes #179.

bnaul commented 8 years ago

I made the change suggested above about converting everything to TimeSeries.

featurize_time_series now just handles the logic of determining if one or multiple time series have been passed in; then we attempt to create a TimeSeries object for each, which catches any dimension mismatches. So I think the logic is more neatly compartmentalized. In particular: the compatible_shapes function I wrote before had to handle both the single- and multiple-time series cases, whereas the new one (now in time_series.py) already knows that just a single time series is being considered.

stefanv commented 8 years ago

What is the test coverage looking like on the dask graph section?

stefanv commented 8 years ago

This looks good to me. Over to you @acrellin