MultiZMW stitching, model structure inputs, log(0) fixes

[michaelbrownid]

Hi,

The code now allows multiple ZMWs in a HDF5 file to run inference and stitching. I will switch my branch from master to dev-v0.1.0 after this to sync with you.

• MultiZMW stitching: multiple zmws in HDF5. readids in hdf5 delinate zmw/window boundaries for correct stitching. Many zmws can be sent to GPU for inference with results stitched at end.

  • model structure inputs: hardcoded values are now inputs. same create_model between training and inference. TODO: should hardcode values in python source files and import different models by name for continuity

  • fixed log(0) in QV values. Softmax can get numeric 1.0 resulting in undefined QV.

[GPUtester]

Can one of the admins verify this patch?

[tijyojwad]

add to allowlist

[tijyojwad]

@michaelbrownid this is an awesome change. would it be okay if we leave a couple of comments on this itself for suggested changes, or would you prefer we fork from this branch and make the changes we're thinking of and push separately?

[michaelbrownid]

Logistically, I'm good whatever is easiest to allow us both to be in sync.

[tijyojwad]

@michaelbrownid I pushed some updates to your branch itself - logic is the same, just updated some variable names and refactored a bit.

[tijyojwad]

add to allowlist

[tijyojwad]

rerun tests

[tijyojwad]

ok to test

[tijyojwad]

rerun tests

[mike-wendt]

rerun tests

[tijyojwad]

rerun tests

[raydouglass]

rerun tests

[raydouglass]

add to allowlist