1. Load Packages
1
2
3
4
5
import torch
from torch import nn, optim
import torch.nn.functional as F
torch.manual_seed(1)
2. Prepare Dataset
1
2
3
4
5
6
X_train = torch.FloatTensor([[73, 80, 75],
[93, 88, 93],
[89, 91, 80],
[96, 98, 100],
[73, 66, 70]])
y_train = torch.FloatTensor([[152], [185], [180], [196], [142]])
1) Without Using torch.nn module
3. Initialize Weights
1
2
W = torch.ones((3, 1), requires_grad=True)
b = torch.zeros(1, requires_grad=True)
4. Gradient Descent
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
params = [W, b]
optimizer = optim.SGD(params, lr=1e-5)
for epoch in range(10):
# 1) hypothesis 정의
hypothesis = X_train @ W + b
# 2) cost 정의
cost = torch.mean((hypothesis - y_train)**2)
# 3) optimizer 설정
optimizer.zero_grad() # gradient를 0으로 초기화
cost.backward() # gradient 계산
optimizer.step() # W, b 업데이트
print(f"Epoch {epoch:2}/100 | Cost: {cost.item():.4f}")
2) Using torch.nn module
3. Gradient Descent
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
model = nn.LazyLinear(1, bias=True)
optimizer = optim.SGD(model.parameters(), lr=1e-5)
for epoch in range(10):
# 1) hypothesis 정의
hypothesis = model(X_train)
# 2) cost 정의
cost = F.mse_loss(hypothesis, y_train)
# 3) optimizer 설정
optimizer.zero_grad() # gradient를 0으로 초기화
cost.backward() # gradient 계산
optimizer.step() # W, b 업데이트
print(f"Epoch {epoch:>2}/100 | Cost: {cost.item():.4f}")
3) Using torch.nn module - Subclass API
3. Gradient Descent
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
class Model(nn.Module):
def __init__(self, output_dim):
super().__init__()
self.linear = nn.LazyLinear(output_dim)
def forward(self, x):
return self.linear(x)
model = Model(3)
optimizer = optim.SGD(model.parameters(), lr=1e-5)
for epoch in range(10):
# 1) hypothesis 정의
hypothesis = model(X_train)
# 2) cost 정의
cost = F.mse_loss(hypothesis, y_train)
# 3) optimizer 설정
optimizer.zero_grad() # gradient를 0으로 초기화
cost.backward() # gradient 계산
optimizer.step() # W, b 업데이트
print(f"Epoch {epoch:>2}/100 | Cost: {cost.item():.4f}")
PREVIOUSEtc