FEATURE: Binary classifier -- oscillatory vs non-oscillatory time series
Summary
(Describe the feature you're requesting)
Current behaviour/setbacks
(Describe what is currently happening and why this is sub-optimal)
Desired behaviour/advantages
(Describe the desired behaviour and what this will improve)
Implementation sketch
- Pre-trained on BY4741 time series from experiment 20016?
- Featurisation:
catch22
(does it fit our paradigm to have a post-process depend on another post-process? If not, this feature could go intoskeletons
?) - Model: likely RF, or SVM. Need to optimise hyperparameters first.
- Output: if SVM, probability of oscillatory using
predict_proba
-- this gives more info than just whether oscillatory or not