Differences with torch.bucketize

torch.bucketize

torch.bucketize(input, boundaries, *, out_int32=False, right=False, out=None)

For more information, see torch.bucketize.

mindspore.ops.bucketize

class mindspore.ops.bucketize(input, boundaries, *, right=False)

For more information, see mindspore.ops.bucketize.

Usage

MindSpore API functions is consistent with that of PyTorch, with differences in the data types supported by the parameters.

PyTorch: input supports the scalar and Tensor types. boundaries supports the Tensor type, and the data type of the returned index can be specified via out_int32.

MindSpore: input supports Tensor type. boundaries supports list type, no out_int32 parameter.

Categories	Subcategories	PyTorch	MindSpore	Difference
Parameters	Parameter 1	input	input	Consistent functiona, different supported data types
	Parameter 2	boundaries	boundaries	Consistent function, different supported data types
	Parameter 3	out_int32	-	PyTorch out_int32 specifies the type of index to return, while MindSpore does not have this parameter.
	Parameter 4	right	right	Consistent
	Parameter 5	out	-	PyTorch out can obtain outputs, while MindSpore does not have this parameter.

Code Example

import torch

boundaries = torch.tensor([1, 3, 5, 7, 9])
v = torch.tensor([[3, 6, 9], [3, 6, 9]])
out1 = torch.bucketize(v, boundaries)
out2 = torch.bucketize(v, boundaries, right=True)
print(out1)
# Out:
# tensor([[1, 3, 4],
#        [1, 3, 4]])

print(out2)
# Out:
# tensor([[2, 3, 5],
#        [2, 3, 5]])

from mindspore import Tensor, ops
boundaries = [1, 3, 5, 7, 9]
v = Tensor([[3, 6, 9], [3, 6, 9]])
out1 = ops.bucketize(v, boundaries)
out2 = ops.bucketize(v, boundaries, right=True)
print(out1)
# Out:
# [[1 3 4]
#  [1 3 4]]

print(out2)
# Out:
# [[2 3 5]
#  [2 3 5]]