Bump version of Python, torch and pytorch-lightning

pycave/clustering/kmeans/metrics.py

@@ -31,7 +31,7 @@ def __init__(
 
     def update(self, data: torch.Tensor, assignments: torch.Tensor) -> None:
         indices = assignments.unsqueeze(1).expand(-1, self.num_features)
-        self.centroids.scatter_add_(0, indices, data)
+        self.centroids.scatter_add_(0, indices, data.float())
 
         counts = assignments.bincount(minlength=self.num_clusters).float()
         self.cluster_counts.add_(counts)
@@ -149,7 +149,7 @@ def update(self, data: torch.Tensor, shortest_distances: torch.Tensor) -> None:
         # Then, we sample from the data `num_choices` times and replace if needed
         choices = (squared_distances + eps).multinomial(self.num_choices, replacement=True)
         self.choices.masked_scatter_(
-            use_choice_from_data.unsqueeze(1), data[choices[use_choice_from_data]]
+            use_choice_from_data.unsqueeze(1), data[choices[use_choice_from_data]].float()
         )
 
         # In any case, the cumulative distances are updated

pycave/utils/lightning_module.py

@@ -24,7 +24,7 @@ def configure_optimizers(self) -> None:
     def training_step(self, batch: torch.Tensor, batch_idx: int) -> None:
         self.nonparametric_training_step(batch, batch_idx)
 
-    def training_epoch_end(self, outputs: List[torch.Tensor]) -> None:
+    def on_training_epoch_end(self, outputs: List[torch.Tensor]) -> None:
         self.nonparametric_training_epoch_end()
 
     @abstractmethod

pyproject.toml

@@ -15,10 +15,10 @@ version = "0.0.0"
 [tool.poetry.dependencies]
 lightkit = "^0.5.0"
 numpy = "^1.20.3"
-python = ">=3.8,<3.11"
-pytorch-lightning = "^1.6.0"
-torch = "^1.8.0"
-torchmetrics = ">=0.6,<0.12"
+python = ">=3.8,<=3.11"
+pytorch-lightning = "<2.2"
+torch = "<2.2.0"
+torchmetrics = ">=0.6,<1.4.0"
 
 [tool.poetry.group.pre-commit.dependencies]
 black = "^22.12.0"