VITAE

• Fully integrated with scanpy.
• Add Jacobian regularizers for more stable results.
• Add MMD loss for batch corrections.

Bug fixing when anndata is given and processed = True (Issue #3 )

• scale_factor is set to be all ones, which is treated as a dummy variable.
• data_type is set to be Gaussian automatically.

Fix bug: preprocess anndata when label_names are given and processed=True.

• Fix bug: Issue #1 import scanpy.
• Fix bug: inference with all isolated nodes.
• Fix bug: dimred augment for visualization when training.
• Update model saving and loading.

• Add a progress bar for inference.
• Add warning for plotting when no_loop=False.
• Add uncertainty quantification plots.
• Add an optimizer for warmup alone.
• Improve visualization.

• Accelerate inference with close-form formulas of integration.
• Change default parameters.
• Remove unused objects D_JS.

• Reproducibility support.
  • Add float type adaptation based on global options tf.keras.backend.set_floatx and tf.keras.backend.floatx.
  • Change dtype to float64 for inputs of PCA.
  • Change dtype to float64 for computation of KNN graphs.
  • Set a random seed for constructing KNN graphs.
• Replace Louvain with Leiden clustering.
• Normalize embeddings for plotting.

• Add differentially gene expression tests.
• Add reproducibility materials for the manuscript.

Add train-test-splitting for pre-training and training.

• Create the evaluation set by randomly sampling 10% of cells.
• Optional to use stratified shuffle splitting.

https://pypi.org/project/pyvitae/

https://pypi.org/project/pyvitae/