simulation_single option with pert_data.prepare_split gives ValueError

[murthy1770]

pert_data.prepare_split(split = 'simulation_single', seed=1) # get data split with seed
pert_data.get_dataloader(batch_size = 32, test_batch_size = 128) # prepare data loader

This gives Value Error. I am using a custom dataset with single gene perturbations only (CROP-Seq). What is the difference between simulation and simulation_single? If you have a dataset with single gene perturbations, what should one use?

---

ValueError Traceback (most recent call last)
Cell In[6], line 1
----> 1 pert_data.prepare_split(split = 'simulation_single', seed=1) # get data split with seed
2 pert_data.get_dataloader(batch_size = 32, test_batch_size = 128) # prepare data loader

File ~/.conda/envs/biomodels/lib/python3.12/site-packages/gears/pertdata.py:355, in PertData.prepare_split(self, split, seed, train_gene_set_size, combo_seen2_train_frac, combo_single_split_test_set_fraction, test_perts, only_test_set_perts, test_pert_genes, split_dict_path)
351 if split in ['simulation', 'simulation_single']:
352 # simulation split
353 DS = DataSplitter(self.adata, split_type=split)
--> 355 adata, subgroup = DS.split_data(train_gene_set_size = train_gene_set_size,
356 combo_seen2_train_frac = combo_seen2_train_frac,
357 seed=seed,
358 test_perts = test_perts,
359 only_test_set_perts = only_test_set_perts
360 )
361 subgroup_path = split_path[:-4] + '_subgroup.pkl'
362 pickle.dump(subgroup, open(subgroup_path, "wb"))

ValueError: too many values to unpack (expected 2)

[domcke]

I have the same issue.

[asunboi]

Likewise, I also have the same issue.

[asunboi]

Still unsure what the exact difference between single and simulation_single is, but it seems like single focuses only on training and testing single perturbations, whereas simulation_single trains on only single perturbations but can test with combo perturbations.

I haven't gotten the single split to work yet for some reason, but the error above in

ValueError: too many values to unpack (expected 2)

can be dealt with either changing line 193 of pertdata.py to be
if self.split_type == 'simulation' or self.split_type == 'simulation_single': instead of if self.split_type == 'simulation':

which returns the adata, subgroup that pertdata expects, since without the change it just returns self.adata.