Function Differences with torch.nn.Softmax

torch.nn.Softmax

class torch.nn.Softmax(dim=None)(input) -> Tensor

For more information, see [torch.nn.Softmax](https://pytorch.org/docs/1.8.1/generated/torch.nn.Softmax.html.

mindspore.nn.Softmax

class mindspore.nn.Softmax(axis=-1)(x) -> Tensor

For more information, see mindspore.nn.Softmax.

Differences

PyTorch: a generalization of the binary classification function on multiclassification, which aims to present the results of multiclassification in the form of probabilities.

MindSpore: MindSpore API implements the same function as PyTorch, but the parameter names are different.

Code Example

The two APIs achieve the same function and have the same usage.

# PyTorch
import torch
import numpy
from torch import tensor
import torch.nn as nn

x = torch.FloatTensor([1, 1])
softmax = nn.Softmax(dim=-1)(x)
print(softmax.numpy())
# [0.5 0.5]

# MindSpore
import mindspore
import numpy as np
from mindspore import Tensor

x = Tensor(np.array([1, 1]), mindspore.float16)
softmax = mindspore.nn.Softmax()
output = softmax(x)
print(output)
# [0.5 0.5]