1
import numpy as np
2
import torch
3

4
a_np = np.array([2,3.3])
5
a_tensor = torch.from_numpy(a_np)
6
print("a_tensor: ", a_tensor)

1
import torch
2

3
a_list = [2,3.3]
4
a_torch = torch.tensor(a_list)
5
print("a_torch: ", a_torch)
6
print("a_torch.dtype: ", a_torch.dtype)

1
import torch
2

3
data = [1,2.1]
4

5
print("Tensor(data):\n", torch.Tensor(data))
6
print("Tensor(2,3):\n", torch.Tensor(2,3))
7
print("tensor(data):\n", torch.tensor(data))
8
try:
9
    print("tensor(2,3):\n", torch.tensor(2,3))
10
except TypeError as e:
11
    print("tensor(2,3)代码运行报错")
12
    print(e)

1
import torch
2

3
print("torch.empty([2,3]):\n", torch.empty([2,3]))
4
print("torch.empty(2,3):\n", torch.empty(2,3))
5
# 这里等价于 torch.tensor([1,2])
6
print("torch.FloatTensor([2,3]):\n", torch.FloatTensor([2,3]))
7
print("torch.FloatTensor(2,3):\n", torch.FloatTensor(2,3))

1
import torch
2

3
print(torch.tensor([1,2.3]).type())
4
torch.set_default_tensor_type(torch.DoubleTensor)
5
print(torch.tensor([1,2.3]).type())

1
import torch
2

3
def showMsg(t, tname):
4
    print("{}:\n{}".format(tname, t))
5
    print("{}平均值:{}\n".format(tname, t.sum() / t.numel()))
6

7
# rand函数会随机产生[0, 1)的均匀分布数值
8
a = torch.rand(3, 3)
9
showMsg(a, "rand")
10

11
# 根据输入的shape生成一个[0,1)均匀分布矩阵
12
a_like = torch.rand_like(a)
13
showMsg(a_like, "a_like")
14

15
# 输入最小值 min，最大值 max 和矩阵阶数 shape
16
# 生成符合均匀分布的范围为[min, max)的整形矩阵
17
a_int = torch.randint(1, 10, [3, 3])
18
showMsg(a_int, "a_int")
19

20
a_int_like = torch.randint_like(a_int, 1, 10)
21
showMsg(a_int_like, "a_int_like")

1
import torch
2

3
def showMsg(t, tname):
4
    print("{}:\n{}".format(tname, t))
5
    mean = t.sum() / t.numel()
6
    print("{}平均值:{}".format(tname, mean))
7
    print("{}方差:{}\n".format(tname, ((t - mean) ** 2).sum() / t.numel() ))
8

9
# 生成N(0,1)的标准正态分布
10
showMsg(torch.randn(300,300), "标准正态分布")
11

12
'''
13
生成符合一定条件的正态分布矩阵
14
torch.normal(mean, standard_Deviation, shape)
15
mean: 均值
16
standard_Deviation: 标准差
17
shape: 生成矩阵的阶数
18
'''
19
a = torch.normal(1, 10 ,(300, 300))
20
showMsg(a, "正态分布")

1
import torch
2

3
def showMsg(t, tname):
4
    print("{}:\n{}\n".format(tname, t))
5

6
showMsg(torch.full([2, 3], 0), "full0")
7
showMsg(torch.full([3, 3], 1), "full1")
8
showMsg(torch.full([], 2), "full2")
9
showMsg(torch.full([1], 3), "full3")

1
import torch
2

3
def showMsg(t, tname):
4
    print("{}:\n{}\n".format(tname, t))
5

6
# 生成[0, 10)的等差数列，默认步长为1
7
showMsg(torch.arange(0, 10), "arange0-10")
8
showMsg(torch.arange(0, 10,2), "arange0-10,step=2")
9

10
# 生成[0, 10]的等差数列，默认步长为1
11
showMsg(torch.range(0, 10), "range0-10")
12
showMsg(torch.range(0, 10, 2), "range0-10,step=2")

1
import torch
2

3
def showMsg(t, tname):
4
    print("{}:\n{}\n".format(tname, t))
5

6
# 生成[0, 10]的等差数列，默认个数为100
7
showMsg(torch.linspace(0, 10), "linspace默认数量")
8
showMsg(torch.linspace(0, 10, steps=4), "linspace, steps=4")
9

10
# 生成[0, 10]的等差数列L1，然后以L1为幂进行运算，并返回运算结果
11
# 默认底为 10，生成数量为100
12
# 下面代码的结果等价于[10^0, 10^0.1010, ... , 10^1]
13
showMsg(torch.logspace(0, 1), "logspace")
14

15
# [2^0, 2^0.33, 2^0.66, 2^1]
16
showMsg(torch.logspace(0, 1, base=2, steps=4), "logspace,base=2, steps=4")

1
import torch
2

3
def showMsg(t, tname):
4
    print("{}:\n{}\n".format(tname, t))
5

6
a = torch.ones(3, 3)
7
showMsg(a, "全1矩阵")
8
showMsg(torch.ones_like(a), "全1矩阵_like")
9
showMsg(torch.zeros(3, 3), "全0矩阵")
10
showMsg(torch.zeros_like(a), "全0矩阵_like")

1
import torch
2

3
def showMsg(t, tname):
4
    print("{}:\n{}\n".format(tname, t))
5

6
showMsg(torch.eye(3,3), "单位矩阵1")
7
showMsg(torch.eye(3,4), "单位矩阵2")
8
showMsg(torch.eye(3), "单位矩阵3")

1
import torch
2

3
def showMsg(t, tname):
4
    print("{}:\n{}\n".format(tname, t))
5

6
# 生成[0, 10)的整数序列并打散
7
# 可用此来打散索引
8
showMsg(torch.randperm(10), "randperm")
9

10
# 简单使用例子
11
math_scores = [
12
    ["张三", 90],
13
    ["李四", 85]
14
]
15

16
english_scores = [
17
    ["张三", 80],
18
    ["李四", 90]
19
]
20

21
idx = torch.randperm(2)
22
print("姓名\t数学\t英语")
23
for i in idx:
24
    print("{}\t{}\t{}".format(math_scores[i][0], math_scores[i][1], english_scores[i][1]))