These are chat archives for freeman-lab/discussion

Aug 2016
Jeremy Freeman
Aug 02 2016 15:29
@justinplittle these are the steps to track an upstream project after a pull request
git remote add --track master upstream git://
git fetch upstream
git merge upstream/master
Jason Wittenbach
Aug 02 2016 20:39
Jeremy Freeman
Aug 02 2016 21:52
oh whoa i didn't know about that paper @jwittenbach !
super relevant
Jason Wittenbach
Aug 02 2016 21:53
yeah, definitely
@syncrostone and I went through it…definitely think we could do better
they still only get ~50-60% for precision and recall
and they take a really simplistic approach
but yeah, totally relevant
they simply filter each time-point independently through 4 layers
and then take a max across time
before a final filter
Jeremy Freeman
Aug 02 2016 22:00
oh weird, what's the justification for that?
Jason Wittenbach
Aug 02 2016 22:05
they don’t really say
well, they do mention one thing:
during preprocessing, they downsample in time to a point where any calcium event end being entirely within one downsample window
so other than on/off, they have effecitvely removed all of the temporal information
so each “frame” is effectively just a different set of active neurons
so I guess that’s the motivation:
the filters find “on” neurons
and then the max across time just merges all the neurons that it found in each frame
Jeremy Freeman
Aug 02 2016 22:14
ohh i see
kinda makes sense
how do their metrics compare to the ones we're using?
@syncrostone this is the paper on benchmarking event detection algorithms
Jeremy Freeman
Aug 02 2016 23:49
@sethvincent someone just emailed me with a minidocs issue, this is the screenshot
when running the example
he has node@v0.12.7 maybe the issue is support for tagged template strings?
poking around i think there were only added in node 4+
Seth Vincent
Aug 02 2016 23:52
ah yeah it's the tagged template strings
Jeremy Freeman
Aug 02 2016 23:52
cool will tell him