Set consistent_label_distribution to True (#584)

* set consistent_label_distribution to True * Update README.md
alibaba · Apr 26, 2023 · b372410 · b372410
1 parent 3ff4664
commit b372410
Show file tree

Hide file tree

Showing 2 changed files with 2 additions and 2 deletions.
diff --git a/federatedscope/core/configs/README.md b/federatedscope/core/configs/README.md
@@ -51,7 +51,7 @@ The configurations related to the data/dataset are defined in `cfg_data.py`.
 |           `data.server_holds_all`            | (bool) False | Only use in global mode, whether the server (workers with idx 0) holds all data, useful in global training/evaluation case | -                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                 |
 |               `data.subsample`               | (float) 1.0 |  Only used in LEAF datasets, subsample clients from all clients | -                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                 |
 |                `data.splits`                 | (list) [0.8, 0.1, 0.1] | Train, valid, test splits | -                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                 |
-| `data.` </br>`consistent_label_distribution` | (bool) False | Make label distribution of train/val/test set over clients keep consistent during splitting | -                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                 |
+| `data.` </br>`consistent_label_distribution` | (bool) True | Make label distribution of train/val/test set over clients keep consistent during splitting | -                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                 |
 |               `data.cSBM_phi`                | (list) [0.5, 0.5, 0.5] | Phi for cSBM graph dataset | -                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                 |
 |                `data.loader`                 | (string) '' | Graph sample name, used in minibatch trainer | 'graphsaint-rw': use `GraphSAINTRandomWalkSampler` as DataLoader; 'neighbor': use `NeighborSampler` as DataLoader.                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                |
 |           `dataloader.num_workers`           | (int) 0 | num_workers in DataLoader | -                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                 |

diff --git a/federatedscope/core/configs/cfg_data.py b/federatedscope/core/configs/cfg_data.py
@@ -22,7 +22,7 @@ def extend_data_cfg(cfg):
     # idx 0) holds all data, useful in global training/evaluation case
     cfg.data.subsample = 1.0
     cfg.data.splits = [0.8, 0.1, 0.1]  # Train, valid, test splits
-    cfg.data.consistent_label_distribution = False  # If True, the label
+    cfg.data.consistent_label_distribution = True  # If True, the label
     # distributions of train/val/test set over clients will be kept
     # consistent during splitting
     cfg.data.cSBM_phi = [0.5, 0.5, 0.5]