You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
We need to generate embeddings, like on v0.1, on all the training set.
Since we are currently doing random cropping on the training set on loading, we also need to implement full coverage.
This is especially important since we now have different embeddings for each instrument. Ideally all embeddings are instrument agnostic, but we don't know yet, so it's important to maximize the overlap to check for that.
When can we expect the script used to generate the v1 embeddings to be created/updated? The app team is hoping they can begin to integrate this ASAP to see if they can get v1 embeddings into the app in time for SatSummit.
We generated embeddings for multiple regions and have an embedding builder. So running the embeddings on the training set might not be so useful anymore.
We need to generate embeddings, like on v0.1, on all the training set.
Since we are currently doing random cropping on the training set on loading, we also need to implement full coverage.
This is especially important since we now have different embeddings for each instrument. Ideally all embeddings are instrument agnostic, but we don't know yet, so it's important to maximize the overlap to check for that.
Related #207.
The text was updated successfully, but these errors were encountered: