ys-zong / const Goto Github PK

Hi Author!
I met some errors when running conST, which reported the error "did not define the function contrastive_loss". I didn't find the code of contrastive_loss in the model.py or in the training.py. Could you help me solve it?

Thanks!

Hello, can you confirm that the torchvision version is indeed 0.9.2?

Hey,

I was trying to run your code on my local but leidenalg keeps giving this error:
BaseException: Could not construct partition: Weight vector not the same size as the number of edges.
I have tried using every version of leidenalg from 0.7.0 to 0.9.0.
Do you happen to know any work around.

Complete Error Message:
`BaseException Traceback (most recent call last)
/tmp/ipykernel_58748/2566975975.py in
35 sc.pp.neighbors(adata_conST, n_neighbors=params.eval_graph_n)
36
---> 37 eval_resolution = res_search_fixed_clus(adata_conST, n_clusters)
38 print(eval_resolution)
39 cluster_key = "conST_leiden"

/media/nikhilmehta/New Volume/01IITK/Sem 7/CS690/conST/src/utils_func.py in res_search_fixed_clus(adata, fixed_clus_count, increment)
36 '''
37 for res in sorted(list(np.arange(0.01, 2.5, increment)), reverse=True):
---> 38 sc.tl.leiden(adata, random_state=0, resolution=res)
39 count_unique_leiden = len(pd.DataFrame(adata.obs['leiden']).leiden.unique())
40 if count_unique_leiden == fixed_clus_count:

~/anaconda3/envs/cs690/lib/python3.7/site-packages/scanpy/tools/_leiden.py in leiden(adata, resolution, restrict_to, random_state, key_added, adjacency, directed, use_weights, n_iterations, partition_type, neighbors_key, obsp, copy, **partition_kwargs)
142 partition_kwargs['resolution_parameter'] = resolution
143 # clustering proper
--> 144 part = leidenalg.find_partition(g, partition_type, **partition_kwargs)
145 # store output into adata.obs
146 groups = np.array(part.membership)

~/anaconda3/envs/cs690/lib/python3.7/site-packages/leidenalg/functions.py in find_partition(graph, partition_type, initial_membership, weights, n_iterations, max_comm_size, seed, **kwargs)
81 kwargs['weights'] = weights
...
--> 856 singleton partition.
857
858 weights : list of double, or edge attribute

BaseException: Could not construct partition: Weight vector not the same size as the number of edges.`

Could you please add a licence to your code. It is unclear if the tool is open source.

Thank you very much for your time.

Greetings, Currently, I have been assigned the task of replicating your conST method; however, I am unable to locate any others such as conST_151674.pth. Could you kindly guide me on where to find them? Thank you sincerely for your assistance.

hello,
I was trying to retrain the conST model without using the trained weights conST_151673.pth, but i I encountered difficulties while performing the following step . So can you share the code about this step.

Thank you!

Hello,
I noticed that several layers had wrong input dimensions when I turned use_img to True, and I have corrected them in my repo forked from yours: frickyinn/conST.
But with MAE image features, the ARI result were little lower than merely using gene expression. I think that maybe I was using the wrong hyper-parameters when I tried to change the dimensions. So could you help me solve this?

Thank you!

Hello,
I was trying to run your model following the provided notebook conST_cluster.ipynb.

For the Histology feature extraction part, following your steps, I am facing an error from the run_mae_extract_feature.py file

Creating model: pretrain_mae_base_patch16_224
Traceback (most recent call last):
  File "run_mae_extract_feature.py", line 97, in <module>
    main(opts)
  File "run_mae_extract_feature.py", line 58, in main
    model = get_model(args)
  File "run_mae_extract_feature.py", line 42, in get_model
    model = create_model(
  File "/usr/local/lib/python3.8/dist-packages/timm/models/factory.py", line 78, in create_model
    raise RuntimeError('Unknown model (%s)' % model_name)
RuntimeError: Unknown model (pretrain_mae_base_patch16_224)

My timm version is timm-0.6.12. Exploring the rwightman/pytorch-image-models repo, I am led to believe I need to install this model on timm (similar issue on the repo huggingface/pytorch-image-models#158 ).

Could you provide me with the necessary files for the pretrain_mae_base_patch16_224 model?