Parametric UMAP by pytorch

Execute the init_env.sh first to initialize your environment

execute command

python main.py --config=./configs/{yourconfigfile}.yaml --dataset={yourdataset}

the image of successful execute

To-do list

• Do ablation study between GCDR and PUMAP.
• Rewrite the train in projection_3D.py
• Implement the cutting of KNN Graph
• Some distortion occurred with different batch_size.
• Add ISOMAP.
• Rewrite loss function with torch.nn.module
• Rewrite the PUMAP_module in UMAPModel.py
• Unify the dtype of tensor.

The mindmap for rewriting the train in projection_3D.py

1. Implement a trainer with checkpoint-saving function so that we can store the intermeidate results.
2. Leverage the trainer to improve our arch.