In [None]:
from google.colab import drive
drive.mount('/content/drive')1
Mounted at /content/drive
연결된 그래픽 카드와 CUDA 버전 확인하기
In [None]:
!nvidia-smi1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
Wed Apr 7 11:23:37 2021
+-----------------------------------------------------------------------------+
| NVIDIA-SMI 460.67 Driver Version: 460.32.03 CUDA Version: 11.2 |
|-------------------------------+----------------------+----------------------+
| GPU Name Persistence-M| Bus-Id Disp.A | Volatile Uncorr. ECC |
| Fan Temp Perf Pwr:Usage/Cap| Memory-Usage | GPU-Util Compute M. |
| | | MIG M. |
|===============================+======================+======================|
| 0 Tesla K80 Off | 00000000:00:04.0 Off | 0 |
| N/A 69C P8 34W / 149W | 0MiB / 11441MiB | 0% Default |
| | | N/A |
+-------------------------------+----------------------+----------------------+
+-----------------------------------------------------------------------------+
| Processes: |
| GPU GI CI PID Type Process name GPU Memory |
| ID ID Usage |
|=============================================================================|
| No running processes found |
+-----------------------------------------------------------------------------+
In [None]:
!nvcc -V1
2
3
4
5
nvcc: NVIDIA (R) Cuda compiler driver
Copyright (c) 2005-2020 NVIDIA Corporation
Built on Wed_Jul_22_19:09:09_PDT_2020
Cuda compilation tools, release 11.0, V11.0.221
Build cuda_11.0_bu.TC445_37.28845127_0
torch, torchvision, torchtext, 버전 확인
In [None]:
import torch
import torchvision
import torchtext
print(f'torch version: {torch.__version__}')
print(f'torchvision version: {torchvision.__version__}')
print(f'torchtext version: {torchtext.__version__}')1
2
3
torch version: 1.8.1+cu101
torchvision version: 0.9.1+cu101
torchtext version: 0.9.1
Basic 1. Tensor
Tensor vs. Ndarray
In [None]:
import torch
import numpy as npIn [None]:
# make random 2D-array data
import random as r
pick=lambda :r.randint(5,10)
R,C=pick(),pick()
base = [[r.randint(0,100) for __ in range(C)] for _ in range(R)]
print('random base data')
for row in base:
print(*row)1
2
3
4
5
6
7
8
9
random base data
46 13 8 58 43 78 52 97 29 10
15 51 70 14 3 16 88 29 71 58
86 63 96 83 2 80 56 18 63 70
48 88 52 84 79 70 72 50 48 97
75 67 59 41 89 18 67 85 46 12
39 31 16 33 91 0 14 46 41 27
65 18 73 77 63 69 42 42 90 49
90 49 8 19 47 60 77 64 29 93
In [None]:
as_ndarray = np.array(base)
as_tensor = torch.tensor(base,dtype=torch.int)
print("casted to ndarray:")
print(as_ndarray)
print()
print("="*50)
print()
print("casted to tensor:")
print(as_tensor)1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
casted to ndarray:
[[46 13 8 58 43 78 52 97 29 10]
[15 51 70 14 3 16 88 29 71 58]
[86 63 96 83 2 80 56 18 63 70]
[48 88 52 84 79 70 72 50 48 97]
[75 67 59 41 89 18 67 85 46 12]
[39 31 16 33 91 0 14 46 41 27]
[65 18 73 77 63 69 42 42 90 49]
[90 49 8 19 47 60 77 64 29 93]]
==================================================
casted to tensor:
tensor([[46, 13, 8, 58, 43, 78, 52, 97, 29, 10],
[15, 51, 70, 14, 3, 16, 88, 29, 71, 58],
[86, 63, 96, 83, 2, 80, 56, 18, 63, 70],
[48, 88, 52, 84, 79, 70, 72, 50, 48, 97],
[75, 67, 59, 41, 89, 18, 67, 85, 46, 12],
[39, 31, 16, 33, 91, 0, 14, 46, 41, 27],
[65, 18, 73, 77, 63, 69, 42, 42, 90, 49],
[90, 49, 8, 19, 47, 60, 77, 64, 29, 93]], dtype=torch.int32)
Dimension, Shape
np.ndim == torch.dim() np.shape == torch.shape
In [None]:
print(f'dimension of as_ndarray, {as_ndarray.ndim}')
print(f'dimension of as_tensor, {as_tensor.dim()}')
print()
print(f'shape of as_ndarray, {as_ndarray.shape}')
print(f'shape of as_tensor, {as_tensor.shape}') # or .size()1
2
3
4
5
dimension of as_ndarray, 2
dimension of as_tensor, 2
shape of as_ndarray, (8, 10)
shape of as_tensor, torch.Size([8, 10])
Broadcasting ndarray == Broadcasting Tensor
In [None]:
a=[[1,2,],
[3,4]]
b=[5,6]
na,nb=np.array(a),np.array(b)
nc=na+nb
print("Broadcasting in ndarray")
print(nc)
print()
print("="*50)
ta,tb=torch.tensor(a),torch.tensor(b)
tc=ta+tb
print("Broadcasting in tensor")
print(tc)1
2
3
4
5
6
7
8
Broadcasting in ndarray
[[ 6 8]
[ 8 10]]
==================================================
Broadcasting in tensor
tensor([[ 6, 8],
[ 8, 10]])
Numpy axis parameter == Torch dim paratemer
In [None]:
a=[[1.,2.,3.],
[4.,5.,6.]]
ta=torch.tensor(a)
print("dim=1 적용")
print(ta.mean(dim=1))
print()
print("dim=0 적용")
print(ta.sum(dim=0))
print()
print("dim=-1 적용 (마지막 차원이 삭제되는 효과라 생각하면 된다.)")
print(ta.max(dim=-1)) # tensor.max() 는 (max_value,arg max) 의 두 정보를 동시에 반환한다.
print(f'max values: {ta.max(dim=-1)[0]}')
print(f'max indexes: {ta.max(dim=-1)[1]}')1
2
3
4
5
6
7
8
9
10
11
12
dim=1 적용
tensor([2., 5.])
dim=0 적용
tensor([5., 7., 9.])
dim=-1 적용 (마지막 차원이 삭제되는 효과라 생각하면 된다.)
torch.return_types.max(
values=tensor([3., 6.]),
indices=tensor([2, 2]))
max values: tensor([3., 6.])
max indexes: tensor([2, 2])