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:

  1. train and encode with scDLKit

  2. store the embedding in adata.obsm

  3. run sc.pp.neighbors and sc.tl.umap

  4. visualize with Scanpy plotting