Improve Camelyon16 benchmark (#296)

* feat: Add phikon feature extractor

* feat: Improve Camelyon16 benchmark script

* feat: Add plots

flamby/datasets/fed_camelyon16/benchmark.py

@@ -14,6 +14,7 @@
     Baseline,
     BaselineLoss,
     FedCamelyon16,
+    Optimizer,
     collate_fn,
     metric,
 )
@@ -73,7 +74,7 @@ def main(num_workers_torch, log=False, log_period=10, debug=False, cpu_only=Fals
         if use_gpu:
             m = m.cuda()
         loss = BaselineLoss()
-        optimizer = optim.Adam(m.parameters(), lr=LR)
+        optimizer = Optimizer(m.parameters(), lr=LR)
         if log:
             # We create one summarywriter for each seed in order to overlay the plots
             writer = SummaryWriter(log_dir=f"./runs/seed{seed}")

flamby/datasets/fed_camelyon16/dataset.py

@@ -151,16 +151,16 @@ def __init__(
         if len(self.features_paths) < len(self.labels.index):
             if not (self.debug):
                 logging.warning(
-                    f"You have {len(self.features_paths)} features found in \
-                    {str(self.tiles_dir)} instead of {len(self.labels.index)} \
-                (full Camelyon16 dataset), please go back to the installation \
-                        instructions."
+                    f"You have {len(self.features_paths)} features found in"
+                    f" {str(self.tiles_dir)} instead of {len(self.labels.index)} (full"
+                    " Camelyon16 dataset), please go back to the installation"
+                    " instructions."
                 )
             else:
                 print(
-                    f"Warning you are operating on a reduced dataset in \
-                    DEBUG mode with in total \
-                {len(self.features_paths)}/{len(self.labels.index)} features."
+                    "Warning you are operating on a reduced dataset in  DEBUG mode with"
+                    " in total {len(self.features_paths)}/{len(self.labels.index)}"
+                    " features."
                 )
 
     def __len__(self):
@@ -251,7 +251,7 @@ def __init__(
         ]
 
 
-def collate_fn(dataset_elements_list, max_tiles=1000):
+def collate_fn(dataset_elements_list, max_tiles=10000):
     """Helper function to correctly batch samples from
     a Camelyon16Dataset accomodating for the uneven number of tiles per slide.
 

flamby/datasets/fed_ixi/dataset.py

@@ -130,7 +130,7 @@ def _validate_center(self) -> None:
         centers = list(self.CENTER_LABELS.keys()) + list(self.CENTER_LABELS.values())
         assert self.centers[0] in centers, (
             f"Center {self.centers[0]} "
-            f"is not compatible with this dataset. "
+            "is not compatible with this dataset. "
             f"Existing centers can be named as follow: {centers} "
         )