MDE in preprocess is not perfect

[liuxiawei]

the ov.pp.mde, the device is limit into cuda , and the params _kwargs can not modified and looks like confilt with constraint

[Starlitnightly]

you are right, in new version the argument will be change in ov.pp.mde:

def mde(adata,embedding_dim=2,n_neighbors=15, basis='X_mde',n_pcs=None, use_rep=None, knn=True, 
        transformer=None, metric='euclidean',verbose=False,
        key_added=None,random_state=0,repulsive_fraction=0.7,constraint=None):
    '''
    MDE
    '''
......
    if constraint is None:
        _kwargs = {
        "embedding_dim": embedding_dim,
        "constraint": pymde.Standardized(),
        "repulsive_fraction": repulsive_fraction,
        "verbose": verbose,
        "device": 'cuda',
        "n_neighbors": n_neighbors,
    }
    else:
        _kwargs = {
        "embedding_dim": embedding_dim,
        "constraint": constraint,
        "repulsive_fraction": repulsive_fraction,
        "verbose": verbose,
        "device": 'cuda',
        "n_neighbors": n_neighbors,
    }

[liuxiawei]

Thanks for you help! and it still limit device into cuda, It must use cuda?

[zuiluo1116]

Is there any chance for CPU only? For pyMDE support CPU situations.
Somthing like this (From https://pymde.org/):
import pymde import torch device = 'cuda' if torch.cuda.is_available() else 'cpu'

[zuiluo1116]

AND it worked like this.
Edited _preprocess.py at /usr/local/lib/python3.10/site-packages/omicverse/pp in docker.