Open alexbaden opened 8 years ago
I can already se a problem with the above approach, I think. Converting the "0" cuboids is probably creating a lot of 0 cuboids...
Might be worthwhile to write a script that does a single pass search, finds any stored 0 cuboids, and deletes them from the DB. Happy to investigate if everyone else agrees.
I was running the npz -> blosc conversion script and kept getting failures when trying to loop over the entire dataset.
After a bit of debugging, it appears that there are some cubes in the database that are all zeros...
I added a print statement for any cuboid that has
isNotZeros()
evaluate to false. It prints the x,y,z cube coordinates and the zindex.I then checked the database for this cube, and found it existed in the DB (even though the statement
isNotZeros()
returned false).And, sure enough, if we multiply out the cuboid coordinates by the cube dimension (e.g.
384,0,0
and check the cutout, I get an error...nd/sd/kristina15/NR2B_9thA/xy/2/384,440/0,100/1/
But, a cutout inside a cuboid that didn't return zeros works fine...
nd/sd/kristina15/NR2B_9thA/xy/2/100,200/0,100/1/
(note: this project on synaptomes has been fully converted already regardless of cube state, see below)
Here's a SQL dump of this table: https://www.dropbox.com/s/s5b8ookx6bq8sk2/NR2B_9thA_res2.sql?dl=0
The dataset is kristina15, channel is 16-bits.
I don't want to delete this zero cuboids until we've verified that there actually is no data in them. In the meantime, I changed the convert code to also convert 0 cuboids. That will at least let us look at the data and move forward with TRA related goals, and seems to be working. Was curious if there were any thoughts as to what is going on...
@randalburns @perlman