pytorch学习（一）—Tensor

男娘i 2023-06-05 12:53 5阅读 0赞

#Tensor的创建及常用方法
    from __future__ import print_function
    import torch as t
    import numpy as np
    
    
    #创建一个3*3全1的矩阵
    x1 = t.ones(3,3)
    print(x1)
    
    #5*5全0矩阵
    x2 = t.zeros(5,5)
    print(x2)
    
    #与x1同维度的0矩阵
    x3 = t.zeros_like(x1)
    print(x3)
    
    #与x1同维度全0矩阵
    x4 = t.ones_like(x1)
    print(x4)
    
    #对角矩阵
    x5 = t.diag(t.from_numpy(np.array([1,2,3,4,5])))
    print(x5)
    
    #5*5随机矩阵
    x6 = t.rand(5,5)
    print(x6)
    
    #5*5 norm分布矩阵
    x7 = t.randn(5,5)
    print(x7)
    
    #创建一个empty Tensor
    x8 = t.empty(3,3)
    print(x8)
    
    #创建一个Tensor,给定数值
    x9 = t.Tensor([[1,0,0],[0,1,0],[0,0,1]])
    print(x9)
    
    #根据已有的Tensor创建一个新的Tensor
    x10 = t.rand_like(x9,dtype=t.float)
    print(x10)
    
    #获取Tensor的size，Tensor.size实际上是一个Tuple
    print(x10.size())
    
    ###函数名以_结尾的都是inplace方式，即会修改调用者自己的数据###
    #Tensor的 in place操作，会改变Tensor本身
    x11 = t.rand(3,3)
    print(x11.t_())
    print(x11.copy_(x10))
    
    #Tensor resize/reshape操作
    x12 = t.randn(4,4)
    print(x12)
    x13 = x12.view(16)
    print(x13)
    x14 = x13.view(-1,8)
    print(x14)
    
    #只有一个数的Tensor，使用xxx.item()得到python数值
    x15 = t.rand(1)
    print(x15)
    print(x15.item())
    
    #获取Tensor中元素的个数
    x16 = t.randn(1,2,3,4,5)
    print(x16)
    print(t.numel(x16))
    print(t.numel(t.zeros(4,5)))
    
    #判断一个对象是否是Tensor
    print(t.is_tensor(x16))
    print(t.is_tensor(np.array([1,2,3,4])))
    
    #判断一个对象是否为Pytorch storage object
    print(t.is_storage(t.empty(3,3)))
    print(t.is_storage(np.zeros(shape=(3,3))))
    
    #设置Tensor的数据类型，初始默认值为torch.float32
    print(t.Tensor([1,2,3]).dtype)
    t.set_default_dtype(t.float64)
    print(t.Tensor([1,2,3]).dtype)
    
    #获取默认的Tensor数据类型
    print(t.get_default_dtype())
    t.set_default_dtype(t.float32)
    print(t.get_default_dtype()) 
    
    print("*************")
    
    ###Tensor常见的操作###
    #加减乘除
    #矩阵运算
    #切片
    
    #pytorch矩阵的操作
    var = t.Tensor() #返回一个Tensor
    tensor1 = t.Tensor(3,3)
    tensor2 = t.Tensor(3,3)
    
    print(tensor1,tensor2)
    var2 = t.add(tensor1,tensor2)   #矩阵相加
    var3 = t.sub(tensor1,tensor2)   #减
    var4 = t.mul(tensor1,tensor2)   #乘
    var5 = t.div(tensor1,tensor2)   #矩阵点除
    var6 = t.mm(tensor1,tensor2)    #矩阵乘
    
    print(dir(t.Tensor))
    print(help(t.tensor))
    
    x17 = t.Tensor(5,3)   #构造一个5*3的Tensor
    x18 = t.rand(5,3)   #构造一个随机初始化的Tensor
    print(x17.size())
    print(x18.size())
    print(x17)
    print(x18)
    
    ###Tensor相加###
    #2个Tensor相加
    #方式一
    y = t.rand(5,3)
    print(y)
    var7 = t.add(x17,y)
    #方式二
    var8 = x18 + y 
    #方式三
    var9 = t.rand(5,3)
    t.add(x17,y,out=var9)
    print(var9)
    
    #Tensor相加，y.add_()会改变y的值
    y.add_(x18)
    print(y)
    
    ###Tensor相减###
    x19 = t.rand(5,5)
    x20 = t.rand(5,5)
    y2 = x19 - x20
    y3 = t.sub(x19,x20)
    print(x19.sub_(x20))
    
    ###Tensor相乘###
    x21 = t.rand(3,3)
    x22 = t.rand(3,3)
    y4 = x21 * x22
    y5 = t.mul(x21,x22)
    print(y5.mul_(x22))
    
    ###Tensor相除###
    x23 = t.rand(5,5)
    x24 = t.rand(5,5)
    y6 = x23 / x24
    y7 = t.div(x23,x24)
    print(x7.div_(x24))
    
    ###Tensor矩阵乘###
    x25 = t.rand(5,5)
    x26 = t.rand(5,5)
    y8 = t.mm(x25,x26)
    #print(x9.mm_(x26)) #错误用法
    
    #矩阵切片###
    var10 = x18[:,1]
    print(x18)
    print(var10)
    
    ####Tensor与numpy.ndarray的转换####
    #numpy.adarray表示一个高维的列表/容器。
    #在数值计算中numpy.ndarray表示一个高维矩阵
    print("################")
    x27 = t.Tensor(5,5)
    
    #将Tensor转换为numpy的ndarray
    var11 = x27.numpy()
    print(var11)
    
    #numpy的ndarray转为Tensor
    x28 = np.ndarray(shape=(5,5),dtype=np.int32)
    print(x28)
    var12 = t.from_numpy(x28)
    print(var12)