1. Documentation Reading

2. Create Random Tensor

  • create random tensor with shape (7, 7)；

import torch

tensor = torch.rand(size=(7, 7))
tensor,tensor.shape
(tensor([[0.8076, 0.0608, 0.7226, 0.3354, 0.5350, 0.7117, 0.7979],
         [0.2785, 0.8947, 0.6694, 0.8950, 0.4479, 0.4788, 0.3541],
         [0.0467, 0.7471, 0.7821, 0.4296, 0.1128, 0.8413, 0.4978],
         [0.3593, 0.7470, 0.0869, 0.4811, 0.4923, 0.7392, 0.3945],
         [0.0290, 0.7696, 0.6214, 0.0891, 0.4282, 0.0289, 0.7090],
         [0.3982, 0.1633, 0.7222, 0.2019, 0.7982, 0.8446, 0.0180],
         [0.1051, 0.7829, 0.1152, 0.6189, 0.7437, 0.5098, 0.2611]]),
 torch.Size([7, 7]))

3. Matrix Multiplication

  • perform matrix multiplication from random tensor with
    shape (1, 7)；hint提示：we need transpose the second tensor；

import torch

ta = torch.rand(size=(7, 7))
tb = torch.rand(size=(1, 7))

# transpose tb to avoid shape issue error
tc = torch.matmul(ta,tb.T) # no error because of transpose
ta,tb,tb.T,tc,tc.shape
(tensor([[0.0505, 0.2425, 0.9124, 0.7002, 0.6695, 0.0461, 0.7386],
         [0.7351, 0.6666, 0.4273, 0.8781, 0.3929, 0.7245, 0.2822],
         [0.0415, 0.7760, 0.5181, 0.3222, 0.7059, 0.8465, 0.2380],
         [0.0596, 0.1614, 0.5427, 0.3571, 0.9552, 0.1087, 0.6462],
         [0.0545, 0.0906, 0.1459, 0.4479, 0.8151, 0.0839, 0.2796],
         [0.2882, 0.2222, 0.9025, 0.6714, 0.3078, 0.0557, 0.7026],
         [0.5068, 0.6830, 0.7884, 0.8747, 0.2235, 0.3706, 0.7427]]),
 tensor([[0.8883, 0.6329, 0.6733, 0.8659, 0.8234, 0.1796, 0.8460]]),
 tensor([[0.8883],
         [0.6329],
         [0.6733],
         [0.8659],
         [0.8234],
         [0.1796],
         [0.8460]]),
 tensor([[2.6033],
         [2.8152],
         [2.0904],
         [2.1823],
         [1.5146],
         [2.4437],
         [3.0496]]),
 torch.Size([7, 1]))

4. Random Seed

  • set random seed to 0，do step create random tensor
    and Perform Matrix Multiplication over again；

import torch

# set manual seed
torch.manual_seed(0)

ta = torch.rand(size=(7, 7))
tb = torch.rand(size=(1, 7))

# matrix multiply tensor
tc = torch.matmul(ta,tb.T)
tc,tc.shape
(tensor([[1.8542],
         [1.9611],
         [2.2884],
         [3.0481],
         [1.7067],
         [2.5290],
         [1.7989]]),
 torch.Size([7, 1]))

5. Random Seed GPU

  • CPU：torch.manual_seed(666)

  • GPU：torch.cuda.manual_seed(666)

6. Send Tensor To GPU

  • 创建两个形状为(2,3)的随机张量，并将它们发送给GPU(含随机种子)；

import torch

# set random seed
torch.manual_seed(1234)

# check for access to GPU
device = "cuda" if torch.cuda.is_available() else "cpu"
print(f"Device: {device}")

# create two random tensors on GPU
ta = torch.rand(size=(2,3)).to(device)
tb = torch.rand(size=(2,3)).to(device)
ta,tb
Device: cuda

(tensor([[0.0290, 0.4019, 0.2598],
         [0.3666, 0.0583, 0.7006]], device='cuda:0'),
 tensor([[0.0518, 0.4681, 0.6738],
         [0.3315, 0.7837, 0.5631]], device='cuda:0'))

7. GPU Matrix Multiplication

import torch

# set random seed
torch.manual_seed(1234)

# check for access to GPU
device = "cuda" if torch.cuda.is_available() else "cpu"
print(f"Device: {device}")

# create two random tensors on GPU
ta = torch.rand(size=(2,3)).to(device)
tb = torch.rand(size=(2,3)).to(device)

# perform matmul on ta and tb
tc = torch.matmul(ta,tb.T)
tc, tc.shape
Device: cuda

(tensor([[0.3647, 0.4709],
         [0.5184, 0.5617]], device='cuda:0'),
 torch.Size([2, 2]))

8. Maximum and Minimum

import torch

# set random seed
torch.manual_seed(1234)

# check for access to GPU
device = "cuda" if torch.cuda.is_available() else "cpu"
print(f"Device: {device}")

# create two random tensors on GPU
ta = torch.rand(size=(2,3)).to(device)
tb = torch.rand(size=(2,3)).to(device)

# perform matmul on ta and tb
tc = torch.matmul(ta,tb.T)

max = torch.max(tc)
min = torch.min(tc)
arg_max = torch.argmax(tc)
arg_min = torch.argmin(tc)

print("Tensor:{}\nShape:{}\nMinimum:{}\nMaximum:{}\nArgMin:{}\nArgMax:{}"
    .format(tc,tc.shape,min,max,arg_min,arg_max))
Device: cuda
Tensor:tensor([[0.3647, 0.4709],
        [0.5184, 0.5617]], device='cuda:0')
Shape:torch.Size([2, 2])
Minimum:0.3647301495075226
Maximum:0.5617256760597229
ArgMin:0
ArgMax:3

9. Tensor Squeeze

  • 使用种子为7创建形状为(1,1,1,10)的随机张量，然后创建新的张量，
    去掉所有1维，留下shape (10)的张量，打印两个张量和它的形状；

import torch

torch.manual_seed(7)
ta = torch.rand(size=(1, 1, 1, 10))

# remove single dimension
tb = ta.squeeze()

# print out tensor
print("Tensor A:{}\n Tensor A Shape:{}\nTensor B:{}\nTensor B Shape:{}"
    .format(ta,ta.shape,tb,tb.shape))
Tensor A:tensor([[[[0.5349, 0.1988, 0.6592, 0.6569, 0.2328, 0.4251, 0.2071, 0.6297,
           0.3653, 0.8513]]]])
 Tensor A Shape:torch.Size([1, 1, 1, 10])
Tensor B:tensor([0.5349, 0.1988, 0.6592, 0.6569, 0.2328, 0.4251, 0.2071, 0.6297, 0.3653,
        0.8513])
Tensor B Shape:torch.Size([10])