search

histocartography / patho-quant-explainer

Code for replicating results presented in the paper: "Quantifying Explainers of Graph Neural Networks in Computational Pathology"
MIT License
43 stars 7 forks source link

about training code #2

Closed tikboaHIT closed 3 years ago

tikboaHIT commented 3 years ago

Thank your excellent work!!!

have you plan to release the training process code?

guillaumejaume commented 3 years ago

Hi,

We didn't plan to add the training script on this repository but feel free to adapt this script:

https://github.com/histocartography/hact-net/blob/main/core/train.py that can be run with:

python train.py --cg_path <SOME-SAVE-PATH>/hact-net-data/cell_graphs/ --config_fpath ../data/config/cg_bracs_cggnn_7_classes_pna.yml -b 8 --in_ram --epochs 60 -l 0.0005

used in our other paper "Hierarchical Graph Representations in Digital Pathology". Of course, it depends the type of GNN layer you wanna use (GIN, PNA), the number of classes (3, 7). Let me know if you encounter issues while training the cell graph GNN.

tikboaHIT commented 3 years ago

Thank you! I will try.