Torch frontend lerp #10690

aparna2071 · 2023-02-20T21:52:25Z

fspyridakos

Thanks for your contribution! I left a few comments please go over them and ask for another review once you're done.

fspyridakos · 2023-02-21T09:41:55Z

ivy/functional/frontends/torch/pointwise_ops.py

+
+@to_ivy_arrays_and_back
+def lerp(start, end, weight):
+    return ivy.lerp(start, end, weight)


ivy.lerp doesn't exist in the ivy API, you can suggest its addition by following the steps here: https://lets-unify.ai/ivy/deep_dive/ivy_frontends.html#missing-ivy-functions.
That being said the function can simply composed by a few arithmetic operations as described here: https://pytorch.org/docs/stable/generated/torch.lerp.html.
Also please add the out argument

fspyridakos · 2023-02-21T09:42:14Z

ivy_tests/test_ivy/test_frontends/test_torch/test_pointwise_ops.py

+import torch
+from torch.testing import assert_allclose
+


Please remove these imports

fspyridakos · 2023-02-21T09:43:29Z

ivy/functional/frontends/torch/pointwise_ops.py

+
+
+@to_ivy_arrays_and_back
+def lerp(start, end, weight):


The first argument should be input as stated in the torch docs: https://pytorch.org/docs/stable/generated/torch.lerp.html

fspyridakos · 2023-02-21T09:45:52Z

ivy_tests/test_ivy/test_frontends/test_torch/test_pointwise_ops.py

+    expected_output = torch.lerp(start, end, weight)
+    output = frontend.torch.lerp(start, end, weight)
+
+    assert_allclose(expected_output, output, rtol=1e-03, atol=1e-05)


No need to do this here, test_frontend_fuction handles this

fspyridakos · 2023-02-21T09:48:16Z

ivy_tests/test_ivy/test_frontends/test_torch/test_pointwise_ops.py

+        input=(start, end, weight),
+        expected_output=expected_output,


don't pack all input arguments into one, make them seperate keyword args, i.e. input=start, end=end, weight=weight.
Remove expected_output
No need to include the out argument when you add it

aparna2071 · 2023-02-21T14:41:50Z

Thank you so much for your detailed review, will help me a lot.
I will make the necessary changes as soon as possible.

aparna2071 · 2023-02-21T16:46:41Z

Hi @fspyridakos, as far as the addition of ivy.lerp function is concerned, I've raised a new issue #10762.
Other than that, I've tried to make all necessary changes as suggested by you.
Kindly review & please feel free to suggest if any more changes are required.
Also, thanks a lot for detailed suggestions.

fspyridakos

Thanks for the changes! Also your ivy.lerp issue mentions the .interp function in jax and numpy, and ivy.interp already exists, if you think you can use that function here go ahead, but I think lerp and interp do different things, just chaning some elementwise functions is fine for now

fspyridakos · 2023-02-21T17:08:15Z

ivy/functional/frontends/torch/pointwise_ops.py

+    if out is None:
+        out = ivy.zeros_like(input)
+    out = ivy.multiply_add(ivy.subtract(end, input), weight, ivy.subtract(input, out), out=out)


most lines are redundant, you can just do return <operation> as with the above functions (out in the docs practically means the return) also look over your implementation, ivy.multiply_add doesn't exist

fspyridakos · 2023-02-21T17:08:47Z

ivy_tests/test_ivy/test_frontends/test_torch/test_pointwise_ops.py

+    fn_tree="torch.lerp",
+    dtype_and_input=helpers.dtype_and_values(
+        available_dtypes=helpers.get_dtypes("float"),
+        default=0.5,


I'm not sure what default is here, but you should probably use num_arrays=3

fspyridakos · 2023-02-21T17:09:05Z

ivy_tests/test_ivy/test_frontends/test_torch/test_pointwise_ops.py

+input_dtype, inputs = dtype_and_input
+start, end, weight = inputs
+helpers.test_frontend_function(
+input_dtypes=input_dtype,
+frontend=frontend,
+test_flags=test_flags,
+fn_tree=fn_tree,
+on_device=on_device,
+input_start=start,
+input_end=end,
+input_weight=weight,
+)


fix indentation here to match functions above

aparna2071 · 2023-02-21T18:23:04Z

Hi @fspyridakos, once again thanks a lot.
As far as ivy.lerp issue is concerned, I've deleted the links for interp & have tried to make all necessary changes as required for this PR.
Kindly review & please let me know if any more changes are required.

fspyridakos

Few more chnages and you're golden

fspyridakos · 2023-02-22T11:47:09Z

ivy/functional/frontends/torch/pointwise_ops.py

@@ -450,3 +450,9 @@ def hypot(input, other, *, out=None):
 @to_ivy_arrays_and_back
 def sigmoid(input, *, out=None):
    return ivy.sigmoid(input, out=out)
+
+
+@to_ivy_arrays_and_back


Please add float16 and bfloat16 as unsupprted dtypes for torch (like other functions above)

fspyridakos · 2023-02-22T11:48:21Z

ivy_tests/test_ivy/test_frontends/test_torch/test_pointwise_ops.py

+    fn_tree="torch.lerp",
+    dtype_and_input=helpers.dtype_and_values(
+        available_dtypes=helpers.get_dtypes("float"),
+        num_arrays=3,


Add the following:
large_abs_safety_factor=2.5, small_abs_safety_factor=2.5, safety_factor_scale="log",
Some large values will not interpolate properly due to float percision limitations, this prevents those from being generated

fspyridakos · 2023-02-22T11:50:28Z

ivy_tests/test_ivy/test_frontends/test_torch/test_pointwise_ops.py

+        input_start=start,
+        input_end=end,
+        input_weight=weight,


just have this be input=start, end=end and so on, the kwargs must have the same names as the function args

aparna2071 · 2023-02-22T13:36:11Z

Hi @fspyridakos, thanks for all your help.
I've made the changes, please let me know if anything else is required.

fspyridakos · 2023-02-22T15:07:22Z

LGTM, had to make 2 small changes but it's good now and the test passes. Thanks again for your contribution!

aparna2071 · 2023-02-22T15:08:38Z

Thanks a lot for merging @fspyridakos.

Co-authored-by: Fotios Spyridakos <[email protected]>

Torch frontend lerp

f4bb3ef

ivy-seed assigned fspyridakos Feb 20, 2023

ivy-leaves added the PyTorch Frontend Developing the PyTorch Frontend, checklist triggered by commenting add_frontend_checklist label Feb 20, 2023

fspyridakos suggested changes Feb 21, 2023

View reviewed changes

Made changes as suggested by reviewer to lerp

cfb46cc

aparna2071 requested a review from fspyridakos February 21, 2023 16:48

fspyridakos suggested changes Feb 21, 2023

View reviewed changes

Updated lerp

88ab7a0

aparna2071 requested a review from fspyridakos February 21, 2023 18:23

aparna2071 added 3 commits February 21, 2023 23:57

Added newline

7520e45

Newline error

c256387

Trying to remove newline error

179056d

fspyridakos suggested changes Feb 22, 2023

View reviewed changes

Made some changes

a475cbd

aparna2071 requested a review from fspyridakos February 22, 2023 13:36

fspyridakos added 2 commits February 22, 2023 15:06

Update pointwise_ops.py

bf800d5

Update test_pointwise_ops.py

97a8d45

fspyridakos merged commit b8151bf into ivy-llc:master Feb 22, 2023

Dsantra92 pushed a commit to Dsantra92/ivy that referenced this pull request Feb 22, 2023

Torch frontend lerp (ivy-llc#10690)

5d330a2

Co-authored-by: Fotios Spyridakos <[email protected]>

vedpatwardhan pushed a commit to vedpatwardhan/ivy that referenced this pull request Feb 23, 2023

Torch frontend lerp (ivy-llc#10690)

d1de37f

Co-authored-by: Fotios Spyridakos <[email protected]>

vedpatwardhan pushed a commit to vedpatwardhan/ivy that referenced this pull request Feb 23, 2023

Torch frontend lerp (ivy-llc#10690)

74bfd61

Co-authored-by: Fotios Spyridakos <[email protected]>

vedpatwardhan pushed a commit to vedpatwardhan/ivy that referenced this pull request Feb 26, 2023

Torch frontend lerp (ivy-llc#10690)

d0561f6

Co-authored-by: Fotios Spyridakos <[email protected]>

vedpatwardhan pushed a commit to vedpatwardhan/ivy that referenced this pull request Feb 26, 2023

Torch frontend lerp (ivy-llc#10690)

c5403e7

Co-authored-by: Fotios Spyridakos <[email protected]>

AnnaTz mentioned this pull request Feb 27, 2023

lerp #10683

Closed

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Torch frontend lerp #10690

Torch frontend lerp #10690

aparna2071 commented Feb 20, 2023

fspyridakos left a comment

fspyridakos Feb 21, 2023

fspyridakos Feb 21, 2023

fspyridakos Feb 21, 2023

fspyridakos Feb 21, 2023

fspyridakos Feb 21, 2023

aparna2071 commented Feb 21, 2023

aparna2071 commented Feb 21, 2023

fspyridakos left a comment

fspyridakos Feb 21, 2023

fspyridakos Feb 21, 2023

fspyridakos Feb 21, 2023

aparna2071 commented Feb 21, 2023

fspyridakos left a comment

fspyridakos Feb 22, 2023

fspyridakos Feb 22, 2023

fspyridakos Feb 22, 2023

aparna2071 commented Feb 22, 2023

fspyridakos commented Feb 22, 2023 •

edited

Loading

aparna2071 commented Feb 22, 2023

Torch frontend lerp #10690

Torch frontend lerp #10690

Conversation

aparna2071 commented Feb 20, 2023

fspyridakos left a comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

aparna2071 commented Feb 21, 2023

aparna2071 commented Feb 21, 2023

fspyridakos left a comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

aparna2071 commented Feb 21, 2023

fspyridakos left a comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

aparna2071 commented Feb 22, 2023

fspyridakos commented Feb 22, 2023 • edited Loading

aparna2071 commented Feb 22, 2023

fspyridakos commented Feb 22, 2023 •

edited

Loading