These are chat archives for freeman-lab/zebra

20th
Jan 2015
Jason Wittenbach
@jwittenbach
Jan 20 2015 00:43
Ah, I can see how that would be tricky.
Is there a sample tiff-stack or some sort of volume image somewhere that I could use for playing around?
Jeremy Freeman
@freeman-lab
Jan 20 2015 01:39
pretty sure /nobackup/freeman/zebra/data/2014-04-17/OMR2/analysis/stats/stats_mean.mat has a fairly typical volume
2048 x 1046 x 40 or something
Jason Wittenbach
@jwittenbach
Jan 20 2015 01:40
Excellent!
Davis Bennett
@d-v-b
Jan 20 2015 02:50
ok tuning model for the orientation data
that seems like the guy
I'm on the api looking at circular tuning
is there an example notebook where this is done?
@freeman-lab might have one?
Jeremy Freeman
@freeman-lab
Jan 20 2015 03:32
not yet, need to add a notebook for that
but the simplest route is to get a Series where each record is the trial-averaged response for each condition
then load the model as an array of orientations, via TuningModel.load
and then just call model.fit
Davis Bennett
@d-v-b
Jan 20 2015 03:59
and that's just going to see the best von mises fit to the trial-averaged responses, given the angles
do I lose anything by looking at trial averages?
and have you done this analysis with neural data?
Jeremy Freeman
@freeman-lab
Jan 20 2015 04:04
yes, probably not, yes
orientation maps from the 2014 pape were done this way
more or less
that's the one i'm working on an example for
Davis Bennett
@d-v-b
Jan 20 2015 04:05
aha
also, looking at some neural data... damn even with nuclear localized indicator, it's blur town
Jeremy Freeman
@freeman-lab
Jan 20 2015 04:05
yeah...
Davis Bennett
@d-v-b
Jan 20 2015 04:05
I think a big factor is the front laser
since it is very wide by the time it hits the hindbrain
Jeremy Freeman
@freeman-lab
Jan 20 2015 04:06
i always thought things were sharper before we added the front later
*laser
Davis Bennett
@d-v-b
Jan 20 2015 04:07
I think it adds a huge amount of background
i mean, I have no idea what could explain the blur zone i see in these nuclear data
which is often back in the hindbrain
Jeremy Freeman
@freeman-lab
Jan 20 2015 04:08
right...
Davis Bennett
@d-v-b
Jan 20 2015 04:08
but if the front laser has diverged and is bathing a ton of nuclei in weak light, you'll get a widefield green blur
Jeremy Freeman
@freeman-lab
Jan 20 2015 04:15
so it's just alignment?
Davis Bennett
@d-v-b
Jan 20 2015 04:15
it's not alignment per se,
there's no way to align a super wide beam to a thin focal plane
if the beam is thin at the forebrain, it's going to be rather wide 200 um away
Davis Bennett
@d-v-b
Jan 20 2015 20:50
@freeman-lab I'm getting some errors after running my registration notebook
15/01/20 15:49:28 ERROR SparkDeploySchedulerBackend: Asked to remove non-existent executor 170
and it increments this for executors 171, 172, etc
this is after bringing down the ipython notebook server and bringing it back up
and I get these errors when calling sc.addPyFile()
so it's not a data-intensive thing
but this error doesn't prevent execution of sc.addPyFile()
e.g. I can call sc.addPyFile('foo') and then call foo.bar() in the next cell
while the errors are still piling up in the previous cell
Davis Bennett
@d-v-b
Jan 20 2015 20:59
practically it seems like I should re-launch spark
Jason Wittenbach
@jwittenbach
Jan 20 2015 21:03
If the file that you're passing to addPyFile() is local (i.e. not on the distributed FS), then that's just doing local stuff. The executors are the programs that actually run the code on the worker nodes, so yeah, restarting Spark is probably the way to go.