Add torch.matmul function. #1780

gchanan · 2017-06-12T16:31:02Z

Includes test_torch, test_autograd and docs changes.

torch/functional.py

+            return torch.mm(tensor1, tensor2)
+        else:
+            return torch.mm(tensor1, tensor2, out=out)
+    elif dim_tensor1 >= 2 and dim_tensor2 >= 2:


test/test_autograd.py

+            # if non-Variable torch function returns a scalar, compare to scalar
+            if not torch.is_tensor(unpacked_result):
+                assert(packed_result.dim() == 1)
+                assert(packed_result.nelement() == 1)


torch/functional.py

+    try:
+        dim_tensor2 = tensor2.dim()
+    except AttributeError:  # not a tensor
+        return NotImplemented


torch/functional.py

+        if out is None:
+            return torch.mm(tensor1.unsqueeze(0), tensor2).squeeze(0)
+        else:
+            return torch.mm(tensor1.unsqueeze(0), tensor2, out=out).squeeze_(0)


torch/functional.py

+            return torch.mm(tensor1.unsqueeze(0), tensor2).squeeze(0)
+        else:
+            return torch.mm(tensor1.unsqueeze(0), tensor2, out=out).squeeze_(0)
+    elif dim_tensor1 == 2 and dim_tensor2 == 2:


torch/functional.py

+        else:
+            return torch.mm(tensor1.unsqueeze(0), tensor2, out=out).squeeze_(0)
+    elif dim_tensor1 == 2 and dim_tensor2 == 2:
+        if out is None:


torch/functional.py

+            return torch.mm(tensor1, tensor2)
+        else:
+            return torch.mm(tensor1, tensor2, out=out)
+    elif dim_tensor1 >= 2 and dim_tensor2 >= 2:


torch/functional.py

+            # But 1) is inconsistent with other functions (e.g. torch.bmm) that will maintain
+            # output non-contiguity if the size is correct (perhaps we should change this globally?)
+            # And 3) is a surprising output to accept if we aren't accepting 1).
+            # So let's just force accepting contiguous tensors.


at least 3-dimensional.

gchanan · 2017-06-13T20:03:18Z

Latest push should have addressed the review comments.

Add torch.matmul function.

213a7bb

Includes test_torch, test_autograd and docs changes.

gchanan mentioned this pull request Jun 12, 2017

Remove flattening for torch.dot #1781

Merged

Add __all__ to functional so imports are accidentally imported.

7d6948c

fmassa reviewed Jun 12, 2017

View reviewed changes

Include unbind in __all__.

21087db

apaszke reviewed Jun 12, 2017

View reviewed changes

gchanan added 3 commits June 13, 2017 12:20

Add matmul case for when one argument is 1-dimensional and the other

efaecef

at least 3-dimensional.

Add squeeze_ to Variable.

3808948

Use squeeze_ instead of squeeze for matmul.

70f159e

soumith approved these changes Jun 14, 2017

View reviewed changes

soumith merged commit 4e35652 into pytorch:master Jun 14, 2017

Add torch.matmul function. #1780

Add torch.matmul function. #1780

Uh oh!

Conversation

gchanan commented Jun 12, 2017

Uh oh!

This comment was marked as off-topic.

Uh oh!

This comment was marked as off-topic.

Uh oh!

This comment was marked as off-topic.

Uh oh!

This comment was marked as off-topic.

Uh oh!

This comment was marked as off-topic.

Uh oh!

This comment was marked as off-topic.

Uh oh!

This comment was marked as off-topic.

Uh oh!

This comment was marked as off-topic.

Uh oh!

This comment was marked as off-topic.

Uh oh!

This comment was marked as off-topic.

Uh oh!

This comment was marked as off-topic.

Uh oh!

This comment was marked as off-topic.

Uh oh!

This comment was marked as off-topic.

Uh oh!

This comment was marked as off-topic.

Uh oh!

This comment was marked as off-topic.

Uh oh!

This comment was marked as off-topic.

Uh oh!

This comment was marked as off-topic.

Uh oh!

This comment was marked as off-topic.

Uh oh!

This comment was marked as off-topic.

Uh oh!

gchanan commented Jun 13, 2017

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

4 participants