Logistic Regression from scratch with PyTorch

logistic_regression.py

import torch

x = torch.tensor([1.5,2.8,15.,19.5])
y = torch.tensor([0.,0.,1.,1.])

w = torch.tensor([-.3], requires_grad=True)
b = torch.tensor([0.1232154], requires_grad=True)


def model(x):
    return torch.sigmoid(x*w + b)


lr = 0.1

for i in range(100):
    y_pred = model(x)
    print(y_pred)

    errors = - (y*torch.log(y_pred) + (1-y) * torch.log(1 - y_pred))
    loss = errors.mean()
    loss.backward()
    w = torch.tensor([w - lr * w.grad], requires_grad=True)
    b = torch.tensor([b - lr * b.grad], requires_grad=True)

    print(f'loss: {loss.item():.4f}')
    print(f'w: {w.item():.4f}')
    print(f'b: {b.item():.4f}')
    print()


print(model(x))
print(model(x) > 0.5)
print(y > 0.5)