189 8069 5689
import numpy as np
import torch# 等差数列(开始，结束，个数)
torch.linspace(0.1,1,5)tensor([0.1000, 0.3250, 0.5500, 0.7750, 1.0000])# 空矩阵
print(torch.empty([3,4]))
# 找了一块不用的内存，不在乎里面的数据，可能是之前遗留下来的(返回未初始化的数据)
print(torch.empty(3,5))tensor([[0., 0., 0., 0.],
[0., 0., 0., 0.],
[0., 0., 0., 0.]])
tensor([[1.0286e-38, 1.0194e-38, 9.6429e-39, 9.2755e-39, 9.1837e-39],
[9.3674e-39, 1.0745e-38, 1.0653e-38, 9.5510e-39, 1.0561e-38],
[1.0194e-38, 1.1112e-38, 1.0561e-38, 9.9184e-39, 1.0653e-38]])# 以base为底，(2,6)五等份，从2的2次方到2的6次方
torch.logspace(2,6,5,base = 2)tensor([ 4., 8., 16., 32., 64.])# (m,n)全a的矩阵
torch.full((3,4),2)tensor([[2, 2, 2, 2],
[2, 2, 2, 2],
[2, 2, 2, 2]])# 二维单位矩阵(E)
torch.eye(5)tensor([[1., 0., 0., 0., 0.],
[0., 1., 0., 0., 0.],
[0., 0., 1., 0., 0.],
[0., 0., 0., 1., 0.],
[0., 0., 0., 0., 1.]])# 高斯分布：torch.normal(mean, std, out=None)
print(torch.normal(2, 3, size=(1,5)))
# 一一对应生成一个
print(torch.normal(mean = torch.arange(1.,11.), std = torch.arange(1,0,-0.1)))tensor([[6.0627, 3.5489, 3.7053, 0.9814, 3.8663]])
tensor([-0.7955, 1.0690, 3.7225, 3.0768, 4.5076, 5.9667, 6.9369, 8.6122,
8.8240, 9.8426])# 标准正态分布
torch.randn((3,4))tensor([[-0.1535, -1.9643, -0.8889, -0.4511],
[ 0.4905, 0.1066, 0.5044, 1.4474],
[ 1.2457, -0.8081, 2.6163, -1.1373]])# 整数均匀分布(均匀分布 =>范围内的数被取到的概率一样 =>low，high范围内的整数随机数) [low,high)
torch.randint(1, 10, (2,3))tensor([[3, 7, 8],
[9, 7, 8]])# 生成从0到n-1的随机排列(随机调整数据序列)
torch.randperm(5)tensor([2, 0, 1, 4, 3])# 伯努利分布(0-1分布)
s = torch.empty(3, 3).uniform_(0,1) # 生成(0,1)之间的随机数
print(s)
print(torch.bernoulli(s))tensor([[0.8420, 0.6974, 0.6081],
[0.2279, 0.3899, 0.1904],
[0.9900, 0.6994, 0.6235]])
tensor([[1., 1., 1.],
[1., 1., 0.],
[1., 1., 0.]])# torch.cat:按dim指定维度拼接，0为行，1为列
x = torch.randint(1, 10, (2, 3))
torch.cat((x, x, x),dim = 0)tensor([[7, 3, 7],
[7, 4, 7],
[7, 3, 7],
[7, 4, 7],
[7, 3, 7],
[7, 4, 7]])# torch.stack:新创建一个维度，拼接(在高一维拼接)
y = torch.randint(1, 10, (2,3))
torch.stack((x,y),1)tensor([[[7, 3, 7],
[9, 1, 8]],
[[7, 4, 7],
[9, 5, 9]]])# torch.split:为int时，表示每一份的长度；为list时，按list元素切分
a = torch.arange(10).reshape(5,2)
print('a:',a)
print('为int时:',torch.split(a, 1)) # 返回tensor列表
print('为list时:',torch.split(a,[2,3]))a: tensor([[0, 1],
[2, 3],
[4, 5],
[6, 7],
[8, 9]])
为int时: (tensor([[0, 1]]), tensor([[2, 3]]), tensor([[4, 5]]), tensor([[6, 7]]), tensor([[8, 9]]))
为list时: (tensor([[0, 1],
[2, 3]]), tensor([[4, 5],
[6, 7],
[8, 9]]))# torch.index_select:按dim指定，截取；indices参数必须是tensor数组的格式
x = torch.randn(3, 4)
print(x)
indices = torch.tensor([0, 2])
torch.index_select(x, 1, indices)tensor([[-0.3880, 0.3159, -1.5236, -1.0030],
[ 1.2080, 0.5500, 0.0146, -1.0543],
[-1.7705, 0.6398, 2.2544, -2.0102]])
tensor([[-0.3880, -1.5236],
[ 1.2080, 0.0146],
[-1.7705, 2.2544]])# torch.masked_select
x = torch.randn(3,4)
print(x)
mask = x.ge(0.5)# 返回True or False 以input为界
print(mask)
torch.masked_select(x, mask) # 按True进行索引，返回一维数组tensor([[ 0.0288, -0.0047, 1.6408, 0.2148],
[-0.7997, -0.3897, 1.1881, -0.5227],
[-0.5981, 1.8946, 0.1165, 0.7575]])
tensor([[False, False, True, False],
[False, False, True, False],
[False, True, False, True]])
tensor([1.6408, 1.1881, 1.8946, 0.7575])# reshape 10 = 2 * 5
torch.arange(10).reshape(2,5)tensor([[0, 1, 2, 3, 4],
[5, 6, 7, 8, 9]])# torch.transpose:按维度转置
x = torch.randint(1,10,(3,4))
print(x)
y = torch.transpose(x, 0, 1)# dim=1和dim=0交换
print(y)
# print(x.T) 二维的矩阵转置可以直接x.T得到tensor([[2, 1, 9, 7],
[9, 4, 5, 8],
[6, 9, 9, 8]])
tensor([[2, 9, 6],
[1, 4, 9],
[9, 5, 9],
[7, 8, 8]])# torch.squeeze:压缩长度为1的维度
x = torch.zeros(2, 1, 2, 1, 2)
print(x.shape)
y = torch.squeeze(x)
print(y,'\n',y.shape)torch.Size([2, 1, 2, 1, 2])
tensor([[[0., 0.],
[0., 0.]],
[[0., 0.],
[0., 0.]]])
torch.Size([2, 2, 2])# torch.unsqueeze:按照dim扩展维度
x = torch.tensor([1, 2, 3, 4])
torch.unsqueeze(x, dim = 1)tensor([[1],
[2],
[3],
[4]])# 首先随机生成训练样本x,y
x = torch.rand(20,1) * 10
y = 2 * x + (torch.rand(20,1) + 6) # y = ax + b# 初始化线性回归参数w,b
w = torch.randn((1), requires_grad=True)
b = torch.randn((1), requires_grad=True)
# 设置学习率lr
lr = 0.05for iterator in range(100):
wx = torch.mul(w, x)# 向量相乘
y_pre = torch.add(wx,b)# 向量相加
# 均方误差损失loss=
loss = (0.5 * (y - y_pre) ** 2).mean()
# 损失函数求导，梯度下降法最小化loss
loss.backward()
# 更新参数
w.data.sub_(lr * w.grad) # '-' 表示 -=
b.data.sub_(lr * b.grad)
# 梯度清零（可以放到求导之前）
w.grad.data.zero_()
b.grad.data.zero_()
print(w.data, b.data)tensor([2.1755]) tensor([5.2428])你是否还在寻找稳定的海外服务器提供商？创新互联www.cdcxhl.cn海外机房具备T级流量清洗系统配攻击溯源，准确流量调度确保服务器高可用性，企业级服务器适合批量采购，新人活动首月15元起，快前往官网查看详情吧
