Skip to content

Conversation

@seiyatsukamoto
Copy link

Bandpass model that has variable resampling, kernel length, and inference rate.

Seiya Tsukamoto and others added 3 commits October 31, 2025 20:40
seiya.tsukamoto

bandpass fixed

Delete projects/train/train/kernel_sampler.py

removed_plot

Delete projects/train/train/data/supervised/multimodal_multiband_plot.py

Add ml4gw generation params

Add multimodal functionality

Switch to amplfi-style prior for training

Fix class paths and update ml4gw dep

Re-organize data flow to get multimodal export and inference working

Add option for constraints on prior

Return only N sampled parameters

Restore accidentally-deleted snapshotter.py

Clean up after rebase

Clean up leftover pieces

fixed data init file

removed allowed offset

fixed val loader
removed multiband callbacks
@wbenoit26
Copy link
Contributor

@seiyatsukamoto What was the resampling/overlap configuration you trained your model with?

@seiyatsukamoto
Copy link
Author

The train.yaml:
kernel_length: 2.75
left_pad: 0.15
right_pad: 0.05
resample_rates: [2048, 1024, 512, 2048]
high_passes: [32, 32, 32, 32]
low_passes: [1024, 128, 64, 1024]
kernel_lengths: [0.5, 1, 2, 1]
inference_sampling_rates: [8, 4, 2, 8]
initial_offsets: [1, 2]

@wbenoit26
Copy link
Contributor

Got it. And can you link the W&B page for the training you did?

@seiyatsukamoto
Copy link
Author

https://api.wandb.ai/links/seiya-tsukamoto-ligo/p5yxxj4h Here's the link to the report. I don't think im on the wandb team so I can direct link the run.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants