Where communities thrive


  • Join over 1.5M+ people
  • Join over 100K+ communities
  • Free without limits
  • Create your own community
People
Repo info
Activity
    Jeremy Freeman
    @freeman-lab
    Whereas the other is a property of any blocked algorithm, to be invoked during the fit
    Jason Wittenbach
    @jwittenbach
    Hmm, I guess I can see that
    though you really can’t set / make sense of the first without knowing the latter
    Jeremy Freeman
    @freeman-lab
    k good to know that was confusing
    will consider tweaking the API to make block size an algorithm parameter
    any luck beyond those parameter issues?
    Jason Wittenbach
    @jwittenbach
    Just finished running it all the way through the pipeline — completely failed
    sources.png
    might need to play with the parameters a bit :-P
    that’s the mean image on a single plane as the base-image though
    so I’m not sure this dataset is a good one to use in the first place
    as I don’t really see any neurons
    Jeremy Freeman
    @freeman-lab
    Ha! Glad it ran at least
    But yeah that looks pretty blurry to begin with
    With the right parameters I got pretty nice results on some mouse data
    Jason Wittenbach
    @jwittenbach
    yeah, even if I squint, I don’t see any neurons in that mean image
    Jeremy Freeman
    @freeman-lab
    Exactly
    And for this algorithm at least there: really no notion of "good"
    So
    It's sort of doing the best it cab
    Yumuu
    @Yumuu
    guys, @jwittenbach @freeman-lab , did you try registration today. seems to be super slow again...
    Jeremy Freeman
    @freeman-lab
    which step is slow? is it during the prepare function?
    try calling thunder-janelia update and then starting again
    the change i recently made may help
    Yumuu
    @Yumuu
    yep, it's that one. it's done now. I'll update and see if it can be faster for the next one. thanks!
    Yumuu
    @Yumuu
    guys, every time I start spark, besides the node I requested, I also got an additional QLOGIN,is it normal?
    Jeremy Freeman
    @freeman-lab
    yup, that's normal
    Davis Bennett
    @d-v-b
    @nvladimus I made some changes to your compression functions above and put them in my branch of zebra
    here's s gist of my version of those functions: https://gist.github.com/d-v-b/5322a73a7bd6cd8a8735
    the main change is adding a kwarg that determines whether old files get deleted
    Davis Bennett
    @d-v-b
    @freeman-lab how hard would it be to compress / decompress in saveAsSeries() / loadSeries()
    i think I asked you this before
    I think you said it would be tricky but I don't remember why :)
    Nikita Vladimirov
    @nvladimus
    ok, thanks for update, @d-v-b
    Nikita Vladimirov
    @nvladimus
    @d-v-b , did the pipeline for registration change after last thunder update? My routine registration now takes forever.
    Davis Bennett
    @d-v-b
    I don't remember any big changes..
    i'm running your registration pipeline on 25 nodes for a 2048 volume dataset and it seems to perform fine
    ahhh maybe you need to set the partitioning
    ok
    when you run tsc.loadImages() do you specify the number of partitions?
    e.g.
    numFrames = volt.getStackFreq(rawDirs[curExp])[2]
    # load images
    dat = tsc.loadImages(rawDirs[curExp], dims, ext='stack', npartitions=numFrames)
    Nikita Vladimirov
    @nvladimus
    no, this looks new
    what is the # of partitions?
    Davis Bennett
    @d-v-b
    it's the number of chunks to divide the entire dataset into, I think
    I asked @freeman-lab about this a while ago, he said there's no perfect way to partition the data, so thunder uses the default partitioning
    but for image operations like registration I use the number of timepoints in the dataset
    Nikita Vladimirov
    @nvladimus
    does it speed up the operation?
    Davis Bennett
    @d-v-b
    yeah it should
    Jeremy Freeman
    @freeman-lab
    yes... the default used to be the number of frames