prevent lattice squishing by sampling from a cuboid #73
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
curtischong
force-pushed
the
prevent-lattice-squishing
branch
from
1a4f0f6 to
9e97704
Compare
curtischong
changed the title
curtischong
force-pushed
the
prevent-lattice-squishing
branch
2 times, most recently
from
621b252 to
ba0789f
Compare
curtischong
force-pushed
the
prevent-lattice-squishing
branch
from
85fa933 to
7b78029
Compare
updated equations to match the mattergen mean and variance function to visualize the new lattice relative to the original lattice more wip correctly sample the mean lattice code to find the max/min angles print quantiles works very well add back visualizations when checking revert back to original variance sampling documentation for v fix forward pass put identity matrix on the same device add back square root on variance to match ddpm. I think the mattergen's variance is wrong only adjust the mean, keep the variance the same update jupyter notebook
curtischong
force-pushed
the
prevent-lattice-squishing
branch
from
7b78029 to
2b0e002
Compare
curtischong
force-pushed
the
prevent-lattice-squishing
branch
from
6a5ad73 to
54bcd51
Compare
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
the main problem is that when t is 200, the noise can really squish the lattice.
ideally, the noise we add won't make the lattice that squished.
however, even a few extra degrees can make the lattice really squished.
I think training the model to predict the volume of the final lattice will help it learn to not be stuck in quished lattices for 100 steps?
After only changing the mean in the forward pass, here are my findings: