RegressionModel use_gpu parameters not working

[Angel030331]

Please use the template below to post a question to https://discourse.scverse.org/c/ecosytem/cell2location/.

Problem

the use_gpu parameters cannot be used in NB regression model training
...

• I follow the instructions from the cell2location tutorial (using on scvi-tools).
• I have adjusted required hyperparameters to my dataset and tissue N_cells_per_location and detection_alpha.
• I have provided 10X reaction/inlet as batch_key for reference NB regression.
• I have checked scverse Discourse and old Cell2location Community Forum, and did not find a solution.

Description of the data input and hyperparameters

...

DATA_PATH = '/Users/onkiwong/Desktop/Year_4/sem1/BIOF3001/Group_project/datasets/seqFISH+'
cellcount = os.path.join(DATA_PATH, 'raw_somatosensory_sc_exp.txt')
sp_data = os.path.join(DATA_PATH, 'Out_rect_locations.csv')
celltype = os.path.join(DATA_PATH, 'somatosensory_sc_labels.txt')

...

df_celltype = pd.read_csv(celltype, header=None, sep='\t')
df_celltype.columns = ['celltype']
df_celltype.index = adata_ref.obs.index
adata_ref.obs['Subset'] = df_celltype['celltype']
adata_ref.obs['Method'] = '3GEX'
adata_ref.obs['Sample'] = adata_ref.obs_names
adata_ref.obs['Sample'] = adata_ref.obs['Sample'].apply(lambda x: x[0:4])

from cell2location.utils.filtering import filter_genes
selected = filter_genes(adata_ref, cell_count_cutoff=5, cell_percentage_cutoff2=0.03, nonz_mean_cutoff=1.12)

5, 0.03, 1.12

In our case, a few genes are cut

adata_ref = adata_ref[:, selected].copy()

RegressionModel.setup_anndata(adata=adata_ref, batch_key='Sample', labels_key='Subset')

os.environ["THEANO_FLAGS"] = 'device=cuda,floatX=' + 'float32' + ',force_device=True' + ',dnn.enabled=False'
from cell2location.models import RegressionModel
mod = RegressionModel(adata_ref)

Use all data for training (validation not implemented yet, train_size=1)

mod.train(max_epochs=400, batch_size=None, train_size=1, lr=0.002, use_gpu=True)

plot ELBO loss history during training, removing first 20 epochs from the plot

mod.plot_history(20)

Single cell reference data: number of cells, number of cell types, number of genes

...

Single cell reference data: technology type (e.g. mix of 10X 3' and 5')

...

Spatial data: number of locations numbers, technology type (e.g. Visium, ISS, Nanostring WTA)

...

[danielgchen]

It looks the "use_gpu" parameter is deprecated in recent package updates, you can use the GPU via the "accelerator" parameter, e.g. mod.train(accelerator='gpu', batch_size=None)! I found that I needed to set the "batch_size" parameter to "None" in the new update order to fully copy the data to GPU or else it is not able to find the data and throws an error like IndexError: Dimension specified as 0 but tensor has no dimensions but might just be on my system.

[Angel030331]

It looks the "use_gpu" parameter is deprecated in recent package updates, you can use the GPU via the "accelerator" parameter, e.g. mod.train(accelerator='gpu', batch_size=None)! I found that I needed to set the "batch_size" parameter to "None" in the new update order to fully copy the data to GPU or else it is not able to find the data and throws an error like IndexError: Dimension specified as 0 but tensor has no dimensions but might just be on my system.

However, when I try this approach, a ValueError arises instead. Please refer to the attached diagram.

[Sawakotx]

Same issue. I just do not use the "use_gpu" parameter and it can work. Also, before import cell2location, I set this code for GPU(according to the common error of cell2location's documents [https://cell2location.readthedocs.io/en/latest/commonerrors.html].

#this line should go before importing cell2location
os.environ["THEANO_FLAGS"] = 'device=cuda,floatX=float32,force_device=True'
import torch
#used_device = torch.device('cpu')#'cuda:3' if torch.cuda.is_available() else
used_device =torch.device('cuda' if torch.cuda.is_available() else 'cpu')

Looking forward to more advice from cell2location team!

[hahia]

IndexError: Dimension specified as 0 but tensor has no dimensions

cell2location 0.1.4