tpcav 0.2.1__tar.gz → 0.2.3__tar.gz

{tpcav-0.2.1/tpcav.egg-info → tpcav-0.2.3}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: tpcav
-Version: 0.2.1
+Version: 0.2.3
 Summary: Testing with PCA projected Concept Activation Vectors
 Author-email: Jianyu Yang <yztxwd@gmail.com>
 License-Expression: MIT AND (Apache-2.0 OR BSD-2-Clause)

{tpcav-0.2.1 → tpcav-0.2.3}/pyproject.toml

@@ -4,7 +4,7 @@ build-backend = "setuptools.build_meta"
 
 [project]
 name = "tpcav"
-version = "0.2.1"
+version = "0.2.3"
 description = "Testing with PCA projected Concept Activation Vectors"
 authors = [{name = "Jianyu Yang", email = "yztxwd@gmail.com"},]
 readme = "README.md"

{tpcav-0.2.1 → tpcav-0.2.3}/test/test_cav_trainer.py

@@ -182,7 +182,9 @@ class TPCAVTest(unittest.TestCase):
 
         self.assertTupleEqual(batch[0].shape, (builder.batch_size, 4, 1024))
 
-        tpcav_model = TPCAV(DummyModelSeq(), layer_name="layer1")
+        device = 'cuda:0' if torch.cuda.is_available() else 'cpu'
+
+        tpcav_model = TPCAV(DummyModelSeq(), layer_name="layer1").to(device)
         tpcav_model.fit_pca(
             concepts=builder.all_concepts(),
             num_samples_per_concept=10,

{tpcav-0.2.1 → tpcav-0.2.3}/tpcav/cavs.py

@@ -205,8 +205,8 @@ class CavTrainer:
                     c, num_samples=num_samples
                 )
                 fscore, weight = _train(
-                    concept_embeddings,
-                    self.control_embeddings,
+                    concept_embeddings.cpu(),
+                    self.control_embeddings.cpu(),
                     Path(output_dir) / c.name,
                     self.penalty,
                 )
@@ -223,8 +223,8 @@ class CavTrainer:
                 res = pool.apply_async(
                     _train,
                     args=(
-                        concept_embeddings,
-                        self.control_embeddings,
+                        concept_embeddings.cpu(),
+                        self.control_embeddings.cpu(),
                         Path(output_dir) / c.name,
                         self.penalty,
                     ),
@@ -413,6 +413,7 @@ def run_tpcav(
     batch_size=8,
     bws=None,
     input_transform_func=helper.fasta_chrom_to_one_hot_seq,
+    fit_pca=True,
     p=4
 ):
     """
@@ -475,11 +476,12 @@ def run_tpcav(
     # create TPCAV model on top of the given model
     tpcav_model = TPCAV(model, layer_name=layer_name)
     # fit PCA on sampled all concept activations of the last builder (should have the most motifs)
-    tpcav_model.fit_pca(
-        concepts=motif_concept_builders[-1].all_concepts() + bed_builder.concepts if  bed_builder is not None else motif_concept_builders[-1].all_concepts(),
-        num_samples_per_concept=num_samples_for_pca,
-        num_pc="full",
-    )
+    if fit_pca:
+        tpcav_model.fit_pca(
+            concepts=motif_concept_builders[-1].all_concepts() + bed_builder.concepts if  bed_builder is not None else motif_concept_builders[-1].all_concepts(),
+            num_samples_per_concept=num_samples_for_pca,
+            num_pc="full",
+        )
     #torch.save(tpcav_model, output_path / "tpcav_model.pt")
 
     # create trainer for computing CAVs

{tpcav-0.2.1 → tpcav-0.2.3}/tpcav/concepts.py

@@ -41,6 +41,8 @@ def _construct_motif_concept_dataloader_from_control(
     motif_mode: str,
     batch_size: int,
     num_workers: int,
+    start_buffer=0,
+    end_buffer=0
 ) -> DataLoader:
     """Mirror the motif-based dataloader logic used in the TCAV script."""
     datasets = []
@@ -51,8 +53,8 @@ def _construct_motif_concept_dataloader_from_control(
             motif=motif,
             motif_mode=motif_mode,
             num_motifs=num_motifs,
-            start_buffer=0,
-            end_buffer=0,
+            start_buffer=start_buffer,
+            end_buffer=end_buffer,
             print_warning=False,
             infinite=False,
         )
@@ -190,7 +192,7 @@ class ConceptBuilder:
                     added.append(concept)
         return added
 
-    def build_motif_concept(self, motifs, concept_name, control_regions=None, motif_mode="pwm"):
+    def build_motif_concept(self, motifs, concept_name, control_regions=None, motif_mode="pwm", start_buffer=0, end_buffer=0):
         if control_regions is None:
             if not self.control_concepts:
                 raise ValueError("Call build_control or pass control_regions first.")
@@ -207,6 +209,8 @@ class ConceptBuilder:
             motif_mode=motif_mode,
             batch_size=self.batch_size,
             num_workers=self.num_workers,
+            start_buffer=start_buffer,
+            end_buffer=end_buffer
         )
         concept = Concept(
             id=self._reserve_id(),

{tpcav-0.2.1 → tpcav-0.2.3}/tpcav/tpcav_model.py

@@ -140,7 +140,7 @@ class TPCAV(torch.nn.Module):
         residual, projected = self.project_activations(avs)
         if projected is not None:
             return torch.cat((projected, residual), dim=1)
-        return residual
+        return residual.detach()
 
     def forward_from_embeddings_at_layer(
         self,

{tpcav-0.2.1 → tpcav-0.2.3/tpcav.egg-info}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: tpcav
-Version: 0.2.1
+Version: 0.2.3
 Summary: Testing with PCA projected Concept Activation Vectors
 Author-email: Jianyu Yang <yztxwd@gmail.com>
 License-Expression: MIT AND (Apache-2.0 OR BSD-2-Clause)