Merge pull request #101 from Nzteb/reg_api

(WIP)Change embedder penalty API
uma-pi1 · May 25, 2020 · 952ec5a · 952ec5a
2 parents 1ec6a1d + 925318a
commit 952ec5a
Show file tree

Hide file tree

Showing 2 changed files with 23 additions and 18 deletions.
diff --git a/kge/model/embedder/lookup_embedder.py b/kge/model/embedder/lookup_embedder.py
@@ -121,10 +121,10 @@ def penalty(self, **kwargs) -> List[Tensor]:
                 ]
             else:
                 # weighted Lp regularization
-                unique_ids, counts = torch.unique(
-                    kwargs["batch"]["triples"][:, kwargs["slot"]], return_counts=True
+                unique_indexes, counts = torch.unique(
+                    kwargs["indexes"], return_counts=True
                 )
-                parameters = self._embeddings(unique_ids)
+                parameters = self._embeddings(unique_indexes)
                 if p % 2 == 1:
                     parameters = torch.abs(parameters)
                 result += [
@@ -135,10 +135,10 @@ def penalty(self, **kwargs) -> List[Tensor]:
                             / p
                             * (parameters ** p * counts.float().view(-1, 1))
                         ).sum()
-                        # In contrast to unweighted Lp regulariztion, rescaling by number of
-                        # triples is necessary here so that penalty term is correct in
-                        # expectation
-                        / len(kwargs["batch"]["triples"]),
+                        # In contrast to unweighted Lp regularization, rescaling by
+                        # number of triples/indexes is necessary here so that penalty
+                        # term is correct in expectation
+                        / len(kwargs["indexes"]),
                     )
                 ]
         else:  # unknown regularization

diff --git a/kge/model/kge_model.py b/kge/model/kge_model.py
@@ -426,19 +426,24 @@ def append_num_parameter(job, trace):
         job.post_epoch_trace_hooks.append(append_num_parameter)
 
     def penalty(self, **kwargs) -> List[Tensor]:
+        # Note: If the subject and object embedder are identical, embeddings may be
+        # penalized twice. This is intended (and necessary, e.g., if the penalty is
+        # weighted).
         if "batch" in kwargs and "triples" in kwargs["batch"]:
-            kwargs["batch"]["triples"] = kwargs["batch"]["triples"].to(
-                self.config.get("job.device")
+            triples = kwargs["batch"]["triples"].to(self.config.get("job.device"))
+            return (
+                super().penalty(**kwargs)
+                + self.get_s_embedder().penalty(indexes=triples[:, S], **kwargs)
+                + self.get_p_embedder().penalty(indexes=triples[:, P], **kwargs)
+                + self.get_o_embedder().penalty(indexes=triples[:, O], **kwargs)
+            )
+        else:
+            return (
+                super().penalty(**kwargs)
+                + self.get_s_embedder().penalty(**kwargs)
+                + self.get_p_embedder().penalty(**kwargs)
+                + self.get_o_embedder().penalty(**kwargs)
             )
-        return (
-            # Note: If the subject and object embedder are identical, embeddings may be
-            # penalized twice. This is intended (and necessary, e.g., if the penalty is
-            # weighted).
-            super().penalty(**kwargs)
-            + self.get_s_embedder().penalty(slot=S, **kwargs)
-            + self.get_p_embedder().penalty(slot=P, **kwargs)
-            + self.get_o_embedder().penalty(slot=O, **kwargs)
-        )
 
     def get_s_embedder(self) -> KgeEmbedder:
         return self._entity_embedder