search

damellis / ESP

The Example-based Sensor Predictions (ESP) system applies machine learning to real-time sensor data.
BSD 3-Clause "New" or "Revised" License
224 stars 52 forks source link

feat(ui): advanced feature view #363

Closed nebgnahz closed 8 years ago

nebgnahz commented 8 years ago
  1. On analysis, training, and prediction tabs, show the features from the last stage live if 'f' is pressed
  2. On the training data view, the live data is replaced with features (as in 1); at the same time, the training sample plot is shortened and the adjoining feature plots are shown. Right now, the features are still computed on the fly (calling populateSampleFeatures). Ideally the features could have been saved [1].

[1] To save the features, then we need a vector of vectors holding all the feature data; they are for each class and each sample. It seems to make sense to implement this in the training data manager. I am leaving that as a future (separate) change request.

References #136

damellis commented 8 years ago

One issue is that it looks like the training samples are fed through the same pipeline as the live data, which makes the live data features glitch when I navigate through the training samples. Should we have a clone of the pipeline that we only use for extracting features from the training samples?

nebgnahz commented 8 years ago

Yes we should! Maybe a new issue?

Not merging to master frequent enough is a problem...

damellis commented 8 years ago

It seems like this no longer crashes if there aren't any feature extraction modules, so you should probably go ahead and merge it. I'll open a new issue for the pipeline duplication.