documentation

mehta-lab · Aug 28, 2024 · 62212f6 · 62212f6
1 parent 3a926ba
commit 62212f6
Show file tree

Hide file tree

Showing 2 changed files with 4 additions and 3 deletions.
diff --git a/applications/contrastive_phenotyping/contrastive_cli/fit.yml b/applications/contrastive_phenotyping/contrastive_cli/fit.yml
@@ -38,7 +38,8 @@ trainer:
   enable_checkpointing: true
   inference_mode: true
   use_distributed_sampler: true
-  # important for contrastive learning to prevent local information leakage
+  # synchronize batchnorm parameters across multiple GPUs.
+  # important for contrastive learning to normalize the tensors across the whole batch.
   sync_batchnorm: true
 model:
   encoder:

diff --git a/applications/contrastive_phenotyping/contrastive_cli/plot_embeddings.py b/applications/contrastive_phenotyping/contrastive_cli/plot_embeddings.py
@@ -50,8 +50,8 @@
 # %%
 # Extract a track from the dataset and visualize its features.
 
-fov_name = "/B/4/4"
-track_id = 71
+fov_name = "/0/1/000000"  # "/B/4/4" FOV names can change between datasets.
+track_id = 21
 all_tracks_FOV = embedding_dataset.sel(fov_name=fov_name)
 a_track_in_FOV = all_tracks_FOV.sel(track_id=track_id)
 # Why is sample dimension ~22000 long after the dataset is sliced by FOV and by track_id?