ctheodoris

/

Geneformer

like 296

Thank you for making this model available!

I have had previous success running InSilicoPerturber on a subsetted Genecorpus-30M dataset composed of ~2700 cells, using the same parameters (I am simply using the Genecorpus-30M dataset to test how the InSilicoPerturbation functions works). However, after changes were pushed on Aug 2, 2023 to fix the attention mask issue, I now receive the following error:

isp = InSilicoPerturber(perturb_type="delete",
                        perturb_rank_shift=None,
                        genes_to_perturb=["ENSG00000135100"],
                        combos=0,
                        anchor_gene=None,
                        model_type="Pretrained",
                        num_classes=0,
                        emb_mode="cell",
                        cell_emb_style="mean_pool",
                        filter_data=None,
                        cell_states_to_model=None,
                        max_ncells=None,
                        emb_layer=-1,
                        forward_batch_size=50,
                        nproc=16,
                        token_dictionary_file = "/home/ubuntu/Geneformer/geneformer/token_dictionary.pkl")

isp.perturb_data("/home/ubuntu/Geneformer",
                 "/data/subset_genecorpus/",
                 "/data/subset_genecorpus/delete_cell/",
                 "delete_cell_HNF1A")

Filter (num_proc=16): 100%|███████| 2741/2741 [00:12<00:00, 214.74 examples/s]
Map (num_proc=16): 100%|███████████████| 37/37 [00:12<00:00,  2.92 examples/s]
Map (num_proc=16): 100%|██████████████| 37/37 [00:00<00:00, 159.23 examples/s]
Map (num_proc=16): 100%|██████████████| 37/37 [00:00<00:00, 166.24 examples/s]
Map (num_proc=16): 100%|██████████████| 37/37 [00:00<00:00, 165.63 examples/s]
Traceback (most recent call last):                                            
  File "<stdin>", line 1, in <module>                                         
  File "/home/ubuntu/Geneformer/geneformer/in_silico_perturber.py", line 974, 
in perturb_data                                                               
    self.in_silico_perturb(model,                                             
  File "/home/ubuntu/Geneformer/geneformer/in_silico_perturber.py", line 1052,
 in in_silico_perturb                                                         
    cos_sims_data = quant_cos_sims(model,                                     
  File "/home/ubuntu/Geneformer/geneformer/in_silico_perturber.py", line 444, 
in quant_cos_sims                                                             
    cos_sims += [cos(minibatch_emb, minibatch_comparison).to("cpu")]          
  File "/opt/tensorflow/lib/python3.10/site-packages/torch/nn/modules/module.p
y", line 1501, in _call_impl                                                  
    return forward_call(*args, **kwargs)                                      
  File "/opt/tensorflow/lib/python3.10/site-packages/torch/nn/modules/distance
.py", line 87, in forward                                                     
    return F.cosine_similarity(x1, x2, self.dim, self.eps)                    
RuntimeError: The size of tensor a (2047) must match the size of tensor b (204
6) at non-singleton dimension 1                                               

I've referenced Discussion #85 to help with this issue; however changing the batch
size to 200 still raises the same error. I also have the latest version of Geneformer pulled.

Could I get some help with why this error is now raising? Thank you!