Srinivas Gorur-Shandilya

convolutional nets surprisingly very good and very fast to train, but they can get confused with dense spike trains (i think)

because these are the one that are problematic

estimateDelays should be much smarter

1

for example, start estimating a 20 second chunk, and keep adding 20 second chunks till delay estimate stabilizies.

create a NN introspective method

1

which embeds the NN training data using t-SNE, and allows the user to identify and refine inconsistinces in the training data

right now, using one IBI value is pretty shit and doesn't work well.

useful for a beginner tutorial

* add a ISI-based outlier detection method * one that compares similarity to its cluster, vs. others

if too many spikes detected, don't attempt to use NNpredict -- we know it's going to be rubbish

and it should reset itself to the first item every time markALlInView is called

as it's possible to scroll past the end of the data stream