Function Differences with torch.floor_divide

torch.floor_divide

torch.floor_divide(
    input,
    other,
    out=None
)

For more information, see torch.floor_divide.

mindspore.ops.FloorDiv

class mindspore.ops.FloorDiv(*args, **kwargs)(
    input_x,
    input_y
)

For more information, see mindspore.ops.FloorDiv.

Differences

PyTorch: The output will be rounded toward 0 rather than the floor.

MindSpore: The output will be rounded exactly toward floor.

Code Example

import mindspore as ms
import mindspore.ops as ops
import torch
import numpy as np

# In MindSpore, the output will be rounded toward the floor, so, after division, the output -0.33 will be rounded to -1.
input_x = ms.Tensor(np.array([2, 4, -1]), ms.int32)
input_y = ms.Tensor(np.array([3, 3, 3]), ms.int32)
floor_div = ops.FloorDiv()
output = floor_div(input_x, input_y)
print(output)
# Out：
# [0 1 -1]

# In torch, the output will be rounded toward 0, so, after division, the output -0.33 will be rounded to 0.
input_x = torch.tensor(np.array([2, 4, -1]))
input_y = torch.tensor(np.array([3, 3, 3]))
output = torch.floor_divide(input_x, input_y)
print(output)
# Out：
# tensor([0, 1, 0])