Visualization#
scDLKit includes quick baseline visualizations so researchers do not need to rebuild plotting code for every sanity check.
Available plots#
loss curves
latent UMAP
latent PCA
reconstruction scatter
classification confusion matrix
model-comparison bar chart
Typical representation workflow#
runner.plot_losses()
runner.plot_latent(method="umap", color="label")
Scanpy after scDLKit#
For the main PBMC notebook, the recommended path is:
train and encode with scDLKit
store the embedding in
adata.obsmrun
sc.pp.neighborsandsc.tl.umapvisualize with Scanpy plotting