Skip to content

Add reconstruction decoder (ReconDecoder) + recon_emb pipeline#16

Open
xiaotong743 wants to merge 2 commits into
mainfrom
recon
Open

Add reconstruction decoder (ReconDecoder) + recon_emb pipeline#16
xiaotong743 wants to merge 2 commits into
mainfrom
recon

Conversation

@xiaotong743

Copy link
Copy Markdown
Collaborator

Added a decoder module in src. other codes just for the exp

@review-notebook-app

Copy link
Copy Markdown

Check out this pull request on  ReviewNB

See visual diffs & provide feedback on Jupyter Notebooks.


Powered by ReviewNB

Comment thread experiments/recon_emb.py
"""
import os

os.environ.setdefault("TF_GPU_ALLOCATOR", "cuda_malloc_async")

Copy link
Copy Markdown
Collaborator

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

why was this needed? can you maybe add a comment to explain?

Comment thread experiments/recon_emb.py
from scaleflow.model._recon import Autoencoder, Decoder, ReconDecoder

# anndata moved write_elem across versions; import defensively.
try:

Copy link
Copy Markdown
Collaborator

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

just lower bound anndata to newer versions, that should be better

Comment thread experiments/recon_emb.py
# ── pure-reconstruction metrics on train / val / test ──
metrics = {}
for s in SPLITS:
pred = batched_apply(module, state.params, inp[idx[s]])

Copy link
Copy Markdown
Collaborator

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

why did you need this? out of gpu memory error? If so can you write in comments. It would be good to know. How big is the model? how big is the batches and how big is the whole thing?

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