nshtrainer 1.3.1__tar.gz → 1.3.3__tar.gz

{nshtrainer-1.3.1 → nshtrainer-1.3.3}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.3
 Name: nshtrainer
-Version: 1.3.1
+Version: 1.3.3
 Summary: 
 Author: Nima Shoghi
 Author-email: nimashoghi@gmail.com

{nshtrainer-1.3.1 → nshtrainer-1.3.3}/pyproject.toml

@@ -1,6 +1,6 @@
 [project]
 name = "nshtrainer"
-version = "1.3.1"
+version = "1.3.3"
 description = ""
 authors = [{ name = "Nima Shoghi", email = "nimashoghi@gmail.com" }]
 requires-python = ">=3.10,<4.0"

{nshtrainer-1.3.1 → nshtrainer-1.3.3}/src/nshtrainer/model/base.py

@@ -85,12 +85,25 @@ def default_split_batched_predictions(
     """
     import torch.utils._pytree as tree
 
-    for sample_idx, batch_idx in enumerate(batch_indices):
+    for i, global_idx in enumerate(batch_indices):
+
+        def _verify_and_index(x: torch.Tensor):
+            # Make sure dim 0 length is equal to the batch size,
+            # otherwise we can't index it and should prompt
+            # the user to implement a splitter
+            if x.shape[0] != len(batch_indices):
+                raise ValueError(
+                    f"Batch size {x.shape[0]} does not match the number of batch indices {len(batch_indices)}. "
+                    "Please implement a custom `split_batched_predictions` method in your LightningModuleBase class."
+                )
+
+            return x[i]
+
         # Create a dictionary for each sample
         yield IndividualSample(
-            index=batch_idx,
-            batch=tree.tree_map(lambda x: x[sample_idx], batch),
-            prediction=tree.tree_map(lambda x: x[sample_idx], prediction),
+            index=global_idx,
+            batch=tree.tree_map(_verify_and_index, batch),
+            prediction=tree.tree_map(_verify_and_index, prediction),
         )
 
 

{nshtrainer-1.3.1 → nshtrainer-1.3.3}/src/nshtrainer/trainer/_config.py

@@ -374,6 +374,10 @@ class TrainerConfig(C.Config):
     """Tags for the run."""
     notes: list[str] = []
     """Human readable notes for the run."""
+    meta: dict[str, Any] = {}
+    """Metadata information for the run. This is a dictionary that can be used to store any additional information
+    about the run. It is not used by nshtrainer, but can be useful for logging or tracking purposes.
+    """
 
     @property
     def full_name(self):