Full Mouse atlas
git clone https://github.com/kimmo1019/scDEC.git
Downloaded Mouse atlas dataset which has been formatted as input of scDEC from the zenode repository. Note that it should be located in the datasets folder. Note that we provide the preprocessed data named X_Y.npz.
python main_clustering.py --data Mouse_atlas --K 40 --dx 10 --dy 20
The training will take about 5 hours on a Linux server with Xeon(R) E5-2680 CPUs and a Tesla P100 GPU. The results will be saved in the results/Mouse_atlas/20200810_152831_x_dim=10_y_dim=20_alpha=10.0_beta=10.0_ratio=0.2 folder.
python eval.py --data Mouse_atlas --timestamp 20200810_152831
Note that the timestamp should be consistent with the training timestamp.
The visualization of latent features will be saved as scDEC_embedding.png in the same results folder.
The clustering results will be saved as scDEC_cluster.txt in the same results folder. Each row denotes the predicted cluster id of each cell.