Custom dataloader registry support #2932
Conversation
…try' into ori-2907-custom-dataloader-registry
…module / registry big change
for more information, see https://pre-commit.ci
…un, we will later adjust this file
Codecov ReportAttention: Patch coverage is
Additional details and impacted files@@ Coverage Diff @@
## main #2932 +/- ##
==========================================
- Coverage 90.14% 82.50% -7.65%
==========================================
Files 181 181
Lines 15644 15548 -96
==========================================
- Hits 14103 12828 -1275
- Misses 1541 2720 +1179
|
and fix the test for custom dataloaders
| adata.obs["batch"] = adata.obs[batch_keys].agg("".join, axis=1).astype("category") | ||
|
|
||
| scvi.model.SCVI.prepare_query_anndata(adata, save_path) | ||
| scvi.model.SCVI.load_query_data(registry=datamodule.registry, reference_model=save_path) |
There was a problem hiding this comment.
We should have more tests that actually fail - using different genes without prepare_query_anndata and different batch categories. Assert that it fails.
|
|
||
| scvi.model.SCVI.prepare_query_anndata(adata, model_census2) | ||
|
|
||
| scvi.model.SCVI.setup_anndata(adata, batch_key="batch") # needed? |
There was a problem hiding this comment.
checking that an AnnData model can be trained using datamodule. Do we really want it?
|
|
||
| user_attributes_model_census3 = model_census3._get_user_attributes() | ||
| pprint(user_attributes_model_census3) | ||
| _ = model_census3.get_elbo() |
| scvi.model.SCVI.prepare_query_anndata(adata, model_census3) | ||
| scvi.model.SCVI.load_query_data(adata, model_census3) | ||
|
|
||
| datamodule_inference = CensusSCVIDataModule( |
There was a problem hiding this comment.
check here that using different genes and different batches fails. You can take much fewer cells here, like 1000.
| # Create a dataloder of a CZI module | ||
| datapipe = datamodule_inference.datapipe | ||
| dataloader = experiment_dataloader(datapipe, num_workers=0, persistent_workers=False) | ||
| mapped_dataloader = ( |
|
|
||
| model = SCVI(adata, n_latent=n_latent) | ||
| model.train(max_epochs=1) | ||
| dataloader = model._make_data_loader(adata) |
There was a problem hiding this comment.
Does model._make_data_loader exist for all models? We should then add the test to the other models as well?
There was a problem hiding this comment.
Is the dataloader sufficient to also setup the model and does setup_datamodule work for it?
…try' into ori-2907-custom-dataloader-registry
for more information, see https://pre-commit.ci
…com/scverse/scvi-tools into ori-2907-custom-dataloader-registry
|
Check out this pull request on See visual diffs & provide feedback on Jupyter Notebooks. Powered by ReviewNB |
for more information, see https://pre-commit.ci
…oader-registry # Conflicts: # docs/tutorials/notebooks
…com/scverse/scvi-tools into ori-2907-custom-dataloader-registry
for more information, see https://pre-commit.ci
…com/scverse/scvi-tools into ori-2907-custom-dataloader-registry
for more information, see https://pre-commit.ci
No description provided.