search

codeneuro / spikefinder-python

python module for testing spike detection algorithms
MIT License
1 stars 3 forks source link

NaN handling #3

Open philippberens opened 8 years ago

philippberens commented 8 years ago

We should decide how to handle NaNs in submission (or how many to allow) and implement it

freeman-lab commented 8 years ago

The current implementation now removes any rows that have a nan in either the measured or predicted rates prior to downsampling. In addition, when running the command line tool to score an entire dataset, we take the nanmedian of scores across neurons. In my testing I occasionally found NaNs in the output of the optimization function for the loglik computation for just one neuron, though I wasn't entirely sure why, probably worth tracking down.