Timmygrad is a scalar value gradient descent optimizer for Python. It is designed to be simple and easy to use, with a focus on readability and understandability. It is not designed for performance, but rather for educational purposes.
this is timmy btw
Here is a simple Linear Regression example using Timmygrad:
m = Value(0.0)
c = Value(0.0)
alpha = 0.01 # learning rate
epochs = 200
for epoch in range(epochs):
for x, y in zip(X, Y):
# forward pass
y_pred = m * x + c
# compute loss
loss = (y - y_pred) ** 2
# backward pass
loss.backward()
# update weights
m.data -= alpha * m.grad
c.data -= alpha * c.grad
# reset gradients
m.grad = 0
c.grad = 0
