Program for classifying network state in a given window before light pulses.
Currently taking 10 seconds before a light pulse and extracting 22 features from this window.
- Refactor network_loader - bug fix on the window and downsampling
- Features are currently linearly dependent.
- Implement feature selection
- Sort out the code from early days vs ipython notebook copies etc
1 . EX110215T12
- Stationarity testing Potential features: vector strength of cross freq coulping Wavelet coefficients stationarity of prev to light - diff to when dive into blocks? coastline PCs of AR coefs, over time many small stationary bins PCs in general - will not work?! The low freq components? - including the weird change ups m hines did, per sec? Or sum of wavelets Eigenvalues
Asses by having counter for x and just plotting? per one. For AR "fingerprint" can use subplots.