...
little variation
dead neurons
still - definite difference at 4000, appears like enough to discriminate
Classifiying using Codewords
A few experiments - the mean's of the nolasing vs lasing codewords are 16.8 apart (euclidean distance).
The question is how linearly separable are the lasing/no-lasing codewords.
Classifying based on euclidean distance to the mean gives 59% accuracy.
Selecting the one best feature of the 4096 components of the codeword, and classifying based on euclidean distance to the mean gives 63% accuracy. The below plot is the accuracy of each of the 4096 components, when used as a single feature, sorted by best feature. Strange why there are two elbows?
A linear classifier, trained with cross entropy loss, get 99.5%, but this is probably not the best classifier to use - I think better will be SVM based with a good margin.
A linear classifier seems to get 99.5% using these codewords.
Finding the Reference
We are not interested in a good classifier though, our metric is if we can find the right reference for a lasing shot, by using eclidean distance in the codeword space.
...