Pytorch入门01
【摘要】 torch.mean函数原型：torch.mean(data,dim=None, keepdim=False)作用：计算指定维度的平均值参数：data：输入的数据dim：指定的维度，如果为None则表示对所有维度进行keepdim：是否保留计算平均值的维度，保留后该维度的长度为1示例：import torchdata = torch.randn(2,3,4)print("原始数据")pri...
torch.mean
函数原型：torch.mean(data,dim=None, keepdim=False)
作用：计算指定维度的平均值
参数：
-
data：输入的数据
-
dim：指定的维度，如果为None则表示对所有维度进行
-
keepdim：是否保留计算平均值的维度，保留后该维度的长度为1
示例：
import torch
data = torch.randn(2,3,4)
print("原始数据")
print(data.shape)
print(data)
default_mean = torch.mean(data, dim=1)
print("默认均值")
print(default_mean.shape)
print(default_mean)
keepdim_mean = torch.mean(data, dim=1, keepdim=True)
print("keepdim均值")
print(keepdim_mean.shape)
print(keepdim_mean)
"""
原始数据
torch.Size([2, 3, 4])
tensor([[[-0.1666, 0.9206, 1.3668, -1.9337],
[-1.9155, 1.0233, -1.0685, -0.6268],
[ 1.3937, -1.0046, 1.5545, -0.1198]],
[[ 1.2142, 0.1488, -1.0612, 0.9261],
[ 1.0475, 0.0313, 0.6086, 0.3535],
[-2.2811, -0.8609, 0.0288, 0.8412]]])
默认均值
torch.Size([2, 4])
tensor([[-0.2295, 0.3131, 0.6176, -0.8934],
[-0.0065, -0.2269, -0.1413, 0.7069]])
keepdim均值
torch.Size([2, 1, 4])
tensor([[[-0.2295, 0.3131, 0.6176, -0.8934]],
[[-0.0065, -0.2269, -0.1413, 0.7069]]])
"""
torch.squeeze & torch.unsqueeze
函数原型：torch.squeeze(input, dim)
作用：移除张量中大小为 1 的维度的函数
参数：
-
input：输入数据
-
dim：是否指定维度，如果不指定则移除张量中所有长度为1的维度
函数原型：torch.unsqueeze(input, dim)
作用：在指定位置插入大小为 1 的维度
参数：
-
input：输入数据
-
dim：指定维度，必填项
示例：
import torch
data = torch.randn(2,3)
unsqueeze_data = torch.unsqueeze(data,dim=0)
print("在0维上进行unsqueeze后数据")
print(unsqueeze_data.shape)
print(unsqueeze_data)
squeeze_data = torch.squeeze(unsqueeze_data,dim=1)
print("在1维上进行squeeze后数据")
print(squeeze_data.shape)
print(squeeze_data)
squeeze_data = torch.squeeze(squeeze_data,dim=0)
print("在0维上进行squeeze后数据")
print(squeeze_data.shape)
print(squeeze_data)
"""
在0维上进行unsqueeze后数据
torch.Size([1, 2, 3])
tensor([[[ 0.5905, -0.0414, -1.8667],
[ 1.5337, 0.0957, 1.1172]]])
在1维上进行squeeze后数据
torch.Size([1, 2, 3])
tensor([[[ 0.5905, -0.0414, -1.8667],
[ 1.5337, 0.0957, 1.1172]]])
在0维上进行squeeze后数据
torch.Size([2, 3])
tensor([[ 0.5905, -0.0414, -1.8667],
[ 1.5337, 0.0957, 1.1172]])
"""
torch.nn.Conv1d
函数原型：torch.nn.Conv1d(in_channels, out_channels, kernel_size, stride=1, padding=0, dilation=1, groups=1, bias=True)
作用：一维卷积
参数：
-
in_channels: 输入通道数
-
out_channels: 输出通道数
-
kernel_size: 卷积核大小
-
stride：滑动步长
-
padding：
-
dilation：
-
groups：
-
bias：
示例
import torch
import torch.nn as nn
# 一维卷积：输入通道=4，输出通道=6（卷积核个数），核大小=3
conv1d = nn.Conv1d(in_channels=4, out_channels=6, kernel_size=3, padding=1)
# 输入：batch=2，通道=4，序列长度=5
x = torch.randn(2, 4, 5)
print("输入数据:\n", x)
print("输入形状:", x.shape)
output = conv1d(x)
print("输出数据:\n", output)
print("输出形状:", output.shape)
print("卷积核形状:", conv1d.weight.shape)
print("偏置项形状:", conv1d.bias.shape)
"""
输入数据:
tensor([[[-1.9907, 0.3726, -0.5296, 0.3108, 0.4288],
[ 0.5699, -0.2702, -0.4008, 0.9082, -0.3383],
[-0.4370, -0.6798, 1.2663, 0.9417, -0.2399],
[-1.0636, -0.4160, 0.0577, 0.0892, 0.2607]],
[[-1.6195, 0.6700, -0.2367, 1.2948, 0.4731],
[-0.4735, 1.4904, 0.4031, -0.7842, -0.5952],
[-0.4210, 0.2534, -0.6936, 0.7000, 1.4875],
[ 1.9850, 1.4336, 1.0419, -0.2482, 0.6426]]])
输入形状: torch.Size([2, 4, 5])
输出数据:
tensor([[[ 1.1034, -0.3813, 0.5577, 0.4728, 0.0519],
[ 0.6731, -0.4322, -0.3775, -0.0852, -0.0534],
[ 0.4972, -0.8872, -0.3269, 0.0670, -0.1191],
[-0.6457, -0.5456, 0.6170, -0.1733, -0.1201],
[-0.1629, 0.0907, 0.2409, 0.0179, 0.2075],
[-0.0758, -0.4469, -0.5576, -0.0177, -0.2723]],
[[-0.2226, -0.9915, 0.0894, -0.4626, 0.6914],
[-0.0640, 0.3878, 0.5609, -0.1710, -0.4539],
[ 0.0470, -1.0436, -0.0946, -0.8633, 0.0134],
[ 0.8971, 0.9505, -0.0975, 0.7095, 0.3819],
[ 0.2498, -0.5089, 0.4106, 0.1497, 0.1669],
[-0.8418, -0.2747, -0.1477, -0.3279, -0.3471]]],
grad_fn=<ConvolutionBackward0>)
输出形状: torch.Size([2, 6, 5])
卷积核形状: torch.Size([6, 4, 3])
偏置项形状: torch.Size([6])
"""
torch.reshape
函数原型：torch.reshape(input, shape)
作用：按照指定的形状来重新调整张量的视图，同时对于非-1列，相关数据采用采用内存连续来组织，具体见下方示例
参数：
-
input(Tensor): 输入张量。 -
shape(tuple 或 int): 目标形状。形状中的元素个数必须与输入张量的元素个数相同。可以使用-1让 PyTorch 自动推断维度。
示例：
import torch
x = torch.arange(1, 25)
A = x.reshape(-1, 6)
a = x.reshape(-1, 6).permute(1, 0)
b = x.reshape(6, -1)
print("原始数组 x:")
print(x)
print("x.shape =", tuple(x.shape))
print()
print("A = x.reshape(-1, 6):")
print(A)
print("A.shape =", tuple(A.shape), " strides =", A.stride(), " contiguous =", A.is_contiguous())
print()
print("a = x.reshape(-1, 6).permute(1, 0):")
print(a)
print("a.shape =", tuple(a.shape), " strides =", a.stride(), " contiguous =", a.is_contiguous())
print()
print("B = x.reshape(6, -1):")
print(b)
print("B.shape =", tuple(b.shape), " strides =", b.stride(), " contiguous =", b.is_contiguous())
"""
原始数组 x:
tensor([ 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18,
19, 20, 21, 22, 23, 24])
x.shape = (24,)
A = x.reshape(-1, 6):
tensor([[ 1, 2, 3, 4, 5, 6],
[ 7, 8, 9, 10, 11, 12],
[13, 14, 15, 16, 17, 18],
[19, 20, 21, 22, 23, 24]])
A.shape = (4, 6) strides = (6, 1) contiguous = True
a = x.reshape(-1, 6).permute(1, 0):
tensor([[ 1, 7, 13, 19],
[ 2, 8, 14, 20],
[ 3, 9, 15, 21],
[ 4, 10, 16, 22],
[ 5, 11, 17, 23],
[ 6, 12, 18, 24]])
a.shape = (6, 4) strides = (1, 6) contiguous = False
B = x.reshape(6, -1):
tensor([[ 1, 2, 3, 4],
[ 5, 6, 7, 8],
[ 9, 10, 11, 12],
[13, 14, 15, 16],
[17, 18, 19, 20],
[21, 22, 23, 24]])
B.shape = (6, 4) strides = (4, 1) contiguous = True
"""
【声明】本内容来自华为云开发者社区博主，不代表华为云及华为云开发者社区的观点和立场。转载时必须标注文章的来源（华为云社区）、文章链接、文章作者等基本信息，否则作者和本社区有权追究责任。如果您发现本社区中有涉嫌抄袭的内容，欢迎发送邮件进行举报，并提供相关证据，一经查实，本社区将立刻删除涉嫌侵权内容，举报邮箱：
cloudbbs@huaweicloud.com
- 点赞
- 收藏
- 关注作者
评论（0）