Repository navigation
Setting for Models to use for Latent Space Explorer #3
Description
Activity
To add a little more detail. To switch a model, the user needs to manually upload the model to the server where the
lse_operatatoris being run, edit the settings file to reference the model and change the setting for current models.We would love for the lse_operator to be linked to MLFlow so that it can do much of this work. In our case, the UI for switching models would be in the Latent Space Explorer. @taxe10
For each scan, every frame is going to go through a process that involves two different models. An initial neural network that produces a latent space vector and dimensionality reduction.
We're configuring these models ahead of time here
However we would like to let the user choose which model to run before a scan. I'm thinking we'll store the current setting in redis and have a resful service to read and update it. This probably won't happen before the upcoming beamtime, but it's something to think about.
For what it's worth, this local configuration of available models is what we hope replace with the ALS Computing Hub.
@SeijDeLeon @xiaoyachong