# Differences with torch.nn.functional.avg_pool1d [](https://gitee.com/mindspore/docs/blob/r2.3.0rc2/docs/mindspore/source_en/note/api_mapping/pytorch_diff/avg_pool1d.md) The following mapping relationships can be found in this file. | PyTorch APIs | MindSpore APIs | | :-------------------: | :-----------------------: | | torch.nn.functional.avg_pool1d | mindspore.ops.avg_pool1d | | torch.nn.functional.avg_pool2d | mindspore.ops.avg_pool2d | | torch.nn.functional.avg_pool3d | mindspore.ops.avg_pool3d | ## torch.nn.functional.avg_pool1d ```text torch.nn.functional.avg_pool1d(input, kernel_size, stride=None, padding=0, ceil_mode=False, count_include_pad=True) ``` For more information, see [torch.nn.functional.avg_pool1d](https://pytorch.org/docs/1.8.1/nn.functional.html#torch.nn.functional.avg_pool1d). ## mindspore.ops.avg_pool1d ```text mindspore.ops.avg_pool1d(input_x, kernel_size=1, stride=1, padding=0, ceil_mode=False, count_include_pad=True) ``` For more information, see [mindspore.ops.avg_pool1d](https://mindspore.cn/docs/en/r2.3.0rc2/api_python/ops/mindspore.ops.avg_pool1d.html). ## Differences PyTorch: Perform average pooling operations on time series data. MindSpore: MindSpore API function is basically the same as pytorch, with different default values for some inputs. | Categories | Subcategories| PyTorch | MindSpore |Differences | | ---- | ----- | ------- | --------- |------------------ | | Parameters | Parameter 1 | input | input_x | Different parameter names | | | Parameter 2 | kernel_size | kernel_size | The pytorch parameter has no default value and the MindSpore parameter has a default value of 1. | | | Parameter 3 | stride | stride | The default value of pytorch parameter is None, which is consistent with kernel_size by default, and the default value of MindSpore Parameter is 1. | | | Parameter 4 | padding | padding | | | | Parameter 5 | ceil_mode | ceil_mode | | | | Parameter 6 | count_include_pad | count_include_pad | | ### Code Example 1 ```python # PyTorch import torch import numpy as np input = torch.tensor([[[1, 2, 3, 4, 5, 6, 7]]], dtype=torch.float32) output = torch.nn.functional.avg_pool1d(input, kernel_size=3, stride=2) print(output) # tensor([[[ 2., 4., 6.]]]) # MindSpore import mindspore from mindspore import Tensor, ops input_x = Tensor([[[1, 2, 3, 4, 5, 6, 7]]], mindspore.float32) output = ops.avg_pool1d(input_x, kernel_size=3, stride=2) print(output) # [[[ 2. 4. 6.]]] ```