Pytorch入门之基本操作
继TensorFlow、keras之后，开始学习新的深度学习框架——Pytorch。整理只是为了方便以后查找。
学习内容来自一个印度小哥哥写的一个在GitHub上的深度学习教程，附上学习链接。
整理人：陈振庭
qq：2621336811
Tensor basics
import numpy as np
import torch
产生一个张量：
x = torch.Tensor(3, 4)
print("Type: {}".format(x.type()))
print("Size: {}".format(x.shape))
print("Values: \n{}".format(x))
结果：
Type: torch.FloatTensor
Size: torch.Size([3, 4])
Values:
tensor([[1.1744e-35, 0.0000e+00, 2.8026e-44, 0.0000e+00],
[ nan, 0.0000e+00, 1.3733e-14, 4.7429e+30],
[1.9431e-19, 4.7429e+30, 5.0938e-14, 0.0000e+
产生一个随机张量：
x = torch.randn(2, 3) # normal distribution (rand(2,3) -> uniform distribution)
print (x)
结果：
tensor([[ 0.7434, -1.0611, -0.3752],
[ 0.2613, -1.7051, 0.9118]])
产生一个全0或者全1张量：
x = torch.zeros(2, 3)
print (x)
x = torch.ones(2, 3)
print (x)
结果：
tensor([[0., 0., 0.],
[0., 0., 0.]])
tensor([[1., 1., 1.],
[1., 1., 1.]])
列表转化为张量：
x = torch.Tensor([[1, 2, 3],[4, 5, 6]])
print("Size: {}".format(x.shape))
print("Values: \n{}".format(x))
结果：
Size: torch.Size([2, 3])
Values:
tensor([[1., 2., 3.],
[4., 5., 6.]])
Numpy数组转化为张量：
x = torch.from_numpy(np.random.rand(2, 3))
print("Size: {}".format(x.shape))
print("Values: \n{}".format(x))
结果：
Size: torch.Size([2, 3])
Values:
tensor([[0.0372, 0.6757, 0.9554],
[0.5651, 0.2336, 0.8303]], dtype=torch.float64)
转换张量的类型：
x = torch.Tensor(3, 4)
print("Type: {}".format(x.type()))
x = x.long()
print("Type: {}".format(x.type()))
结果：
Type: torch.FloatTensor
Type: torch.LongTensor
Tensor operations
张量相加：
x = torch.randn(2, 3)
y = torch.randn(2, 3)
z = x + y
print("Size: {}".format(z.shape))
print("Values: \n{}".format(z))
结果：
Size: torch.Size([2, 3])
Values:
tensor([[ 0.5650, -0.0173, 1.1263],
[ 3.4274, 1.3610, -0.9262]])
张量点乘：
x = torch.randn(2, 3)
y = torch.randn(3, 2)
z = torch.mm(x, y)
print("Size: {}".format(z.shape))
print("Values: \n{}".format(z))
结果：
Size: torch.Size([2, 2])
Values:
tensor([[ 1.3294, -2.4559],
[-0.4337, 4.9667]])
张量转置：
x = torch.randn(2, 3)
print("Size: {}".format(x.shape))
print("Values: \n{}".format(x))
y = torch.t(x)
print("Size: {}".format(y.shape))
print("Values: \n{}".format(y))
结果：
Size: torch.Size([2, 3])
Values:
tensor([[ 0.0257, -0.5716, -0.9207],
[-1.0590, 0.2942, -0.7114]])
Size: torch.Size([3, 2])
Values:
tensor([[ 0.0257, -1.0590],
[-0.5716, 0.2942],
[-0.9207, -0.7114]])
转换张量形状：
z = x.view(3, 2)
print("Size: {}".format(z.shape))
print("Values: \n{}".format(z))
结果：
Size: torch.Size([3, 2])
Values:
tensor([[ 0.0257, -0.5716],
[-0.9207, -1.0590],
[ 0.2942, -0.7114]])
改变张量的形状一定要仔细，不然可能带来意外的结果：
x = torch.tensor([
[[1,1,1,1], [2,2,2,2], [3,3,3,3]],
[[10,10,10,10], [20,20,20,20], [30,30,30,30]]
])
print("Size: {}".format(x.shape))
print("Values: \n{}\n".format(x))
a = x.view(x.size(1), -1)
print("Size: {}".format(a.shape))
print("Values: \n{}\n".format(a))
b = x.transpose(0,1).contiguous()
print("Size: {}".format(b.shape))
print("Values: \n{}\n".format(b))
c = b.view(b.size(0), -1)
print("Size: {}".format(c.shape))
print("Values: \n{}".format(c))
结果：
Size: torch.Size([2, 3, 4])
Values:
tensor([[[ 1, 1, 1, 1],
[ 2, 2, 2, 2],
[ 3, 3, 3, 3]],
[[10, 10, 10, 10],
[20, 20, 20, 20],
[30, 30, 30, 30]]])
Size: torch.Size([3, 8])
Values:
tensor([[ 1, 1, 1, 1, 2, 2, 2, 2],
[ 3, 3, 3, 3, 10, 10, 10, 10],
[20, 20, 20, 20, 30, 30, 30, 30]])
Size: torch.Size([3, 2, 4])
Values:
tensor([[[ 1, 1, 1, 1],
[10, 10, 10, 10]],
[[ 2, 2, 2, 2],
[20, 20, 20, 20]],
[[ 3, 3, 3, 3],
[30, 30, 30, 30]]])
Size: torch.Size([3, 8])
Values:
tensor([[ 1, 1, 1, 1, 10, 10, 10, 10],
[ 2, 2, 2, 2, 20, 20, 20, 20],
[ 3, 3, 3, 3, 30, 30, 30, 30]])
张量的维度操作，即指定某个维度的轴对张量进行操作：
x = torch.randn(2, 3)
print("Values: \n{}".format(x))
y = torch.sum(x, dim=0) # add each row's value for every column
print("Values: \n{}".format(y))
z = torch.sum(x, dim=1) # add each columns's value for every row
print("Values: \n{}".format(z))
结果：
Values:
tensor([[ 0.4295, 0.2223, 0.1772],
[ 2.1602, -0.8891, -0.5011]])
Values:
tensor([ 2.5897, -0.6667, -0.3239])
Values:
tensor([0.8290, 0.7700])
Indexing, Splicing and Joining
张量的索引：
x = torch.randn(3, 4)
print("x: \n{}".format(x))
print ("x[:1]: \n{}".format(x[:1]))
print ("x[:1, 1:3]: \n{}".format(x[:1, 1:3]))
结果：
x:
tensor([[-1.0305, 0.0368, 1.2809, 1.2346],
[-0.8837, 1.3678, -0.0971, 1.2528],
[ 0.3382, -1.4948, -0.7058, 1.3378]])
x[:1]:
tensor([[-1.0305, 0.0368, 1.2809, 1.2346]])
x[:1, 1:3]:
tensor([[0.0368, 1.2809]])
张量的切片：用维度选择：
x = torch.randn(2, 3)
print("Values: \n{}".format(x))
col_indices = torch.LongTensor([0, 2])
chosen = torch.index_select(x, dim=1, index=col_indices) # values from column 0 & 2
print("Values: \n{}".format(chosen))
row_indices = torch.LongTensor([0, 1])
chosen = x[row_indices, col_indices] # values from (0, 0) & (2, 1)
print("Values: \n{}".format(chosen))
结果：
Values:
tensor([[ 0.0720, 0.4266, -0.5351],
[ 0.9672, 0.3691, -0.7332]])
Values:
tensor([[ 0.0720, -0.5351],
[ 0.9672, -0.7332]])
Values:
tensor([ 0.0720, -0.7332])
张量的拼接：
x = torch.randn(2, 3)
print("Values: \n{}".format(x))
y = torch.cat([x, x], dim=0) # stack by rows (dim=1 to stack by columns)
print("Values: \n{}".format(y))
结果：
Values:
tensor([[-0.8443, 0.9883, 2.2796],
[-0.0482, -0.1147, -0.5290]])
Values:
tensor([[-0.8443, 0.9883, 2.2796],
[-0.0482, -0.1147, -0.5290],
[-0.8443, 0.9883, 2.2796],
[-0.0482, -0.1147, -0.5290]])
Gradients
张量求梯度：
x = torch.rand(3, 4, requires_grad=True)
y = 3*x + 2
z = y.mean()
z.backward() # z has to be scalar
print("Values: \n{}".format(x))
print("x.grad: \n", x.grad)
结果：
Values:
tensor([[0.7014, 0.2477, 0.5928, 0.5314],
[0.2832, 0.0825, 0.5684, 0.3090],
[0.1591, 0.0049, 0.0439, 0.7602]], requires_grad=True)
x.grad:
tensor([[0.2500, 0.2500, 0.2500, 0.2500],
[0.2500, 0.2500, 0.2500, 0.2500],
[0.2500, 0.2500, 0.2500, 0.2500]])
- y=3x+2
- z=∑y/N
- ∂(x)∂(z)=∂(y)∂(z)∂(x)∂(y)=N1∗3=121∗3=0.25
CUDA tensors
是否使用CUDA？：
print (torch.cuda.is_available())
结果：
True
产生一个全0张量：
x = torch.Tensor(3, 4).to("cpu")
print("Type: {}".format(x.type()))
结果：
Type: torch.FloatTensor
产生一个全0张量（CUDA下）：
x = torch.Tensor(3, 4).to("cuda")
print("Type: {}".format(x.type()))
结果：
Type: torch.cuda.FloatTensor