5. pytorch를 이용한 Linear Regression

import torch
import numpy as np
import matplotlib.pyplot as plt
%matplotlib inline
import torch.nn as nn

X = torch.linspace(1,50,50).reshape(-1,1) # 1부터 50까지 50 step
# 크기에 맞춰서 1열 형태로
X
'''

tensor([[ 1.],
        [ 2.],
        [ 3.],
        [ 4.],......
'''

torch.manual_seed(71)
e = torch.randint(-8,9,(50,1), dtype=torch.float) # error 값, -8부터 9 까지 50 x 1 크기로

y = 2*X + 1 + e #함수 정의

plt.scatter(X.numpy(),y.numpy())

 

torch.manual_seed(59)
model = nn.Linear(in_features = 1, out_feautures = 1) #선형 회귀로 입력 1 출력 1개씩

print(model.weight)#가중치
print(model.bias)#편향
'''
Parameter containing:
tensor([[0.1060]], requires_grad=True)
Parameter containing:
tensor([0.9638], requires_grad=True)
'''



#모델 생성
class Model(nn.Module):
	def __init__(self,in_features, out_features):
    	super().__init__()
        self.linear = nn.Linear(in_features, out_features) # fully connected layer
        
    def forward(self,x):
    	y_pred = self.linear(x)
        return y_pred
        


torch.manual_seed(59)

model = Model(1,1)

print(model.linear.weight)
print(model.linear.bias)
'''
Parameter containing:
tensor([[0.1060]], requires_grad=True)
Parameter containing:
tensor([0.9638], requires_grad=True)
'''

# 모델 파라메터 이름과 값 출력
for name, param in model.named_parameters():
	print(name,'\t',param.item())
    
'''
linear.weight 	 0.10597813129425049
linear.bias 	 0.9637961387634277
'''


#x값을 선언하고 모델에 적용
x = torch.tensor([2.0])
print(model.forward(x))
# tensor([1.1758], grad_fn=<AddBackward0>)



x1 = np.linspace(0.0,50.0,50)
w1 = 0.1059
b1 = 0.9637

y1 = w1 * x1 +b1

plt.scatter(X.numpy(), y.numpy())
plt.plot(x1,y1,'r')

criterion = nn.MSELoss() # 손실함수 - Mean Squared Error 사용
optimizer = torch.optim.SGD(model.parameters(), lr=0.001) # 모델의 파라메터 최적화


epochs = 50
losses = []

for i in range(epochs):
	i = i+1
    
    #순전파
    y_pred = model.foward(X)
    
    #손실 계산
    loss = criterion(y_pred, y)#예측값과 실제 값 차이
    
    losses.append(loss)
    print(f"epoch {i} loss : {loss.item()} weight : {model.linear.weight.item()} bias: {model.linear.bias.item()}")
    
    #매 반복회수마다 최적화 한 값 초기화
    optimizer.zero_grad()
    
    #역전파 수행
    loss.backward()
    #기울기 하강 한단계 수행
    optimizer.step()

plt.plot(range(epochs), losses)
plt.ylabel('MSE LOSS')
plt.xlabel('Epoch')

# 새 데이터 생성
x = np.linspace(0.0,50.0,50)
current_weight = model.linear.weight.item()
current_bias = model.linear.bias.item()

predicted_y = current_weight *x + current_bias

plt.scatter(X.numpy(),y.numpy())
plt.plot(x,predicted_y,'r')