[BUG] val_strategy='bootstrap' does not create .json file for the bootstrapped split to be referenced later

jamesdolezal / slideflow

Deep learning library for digital pathology, with both Tensorflow and PyTorch support.

https://slideflow.dev

GNU General Public License v3.0

234 stars 39 forks source link

[BUG] val_strategy='bootstrap' does not create .json file for the bootstrapped split to be referenced later #301

Closed siddhir closed 1 year ago

siddhir commented 1 year ago

Description

.json file is not created with splits = /file/to/bootstrap_split.json

To Reproduce

boot_train,boot_val = train_dataset.split( val_strategy='bootstrap', val_fraction=0.2, labels='progressor', splits='/mnt/data/PROJECTS/HNSC_PREMAL/splits/bootstrap_split.json' )

Expected behavior

k-fold splits are saved appropriately but boostrapped splits are not

jamesdolezal commented 1 year ago

This is actually intended behavior, as bootstrap validation involves random partitioning of data into training/validation, and these random partitions are never reused.