my-autograd

Documentation at https://lawjarp-a.github.io/my-autograd/

Install

pip install my_autograd

How to use

Derive from the class Value. Use that to define the scalar values that you will use to implement backpropagation. Use the backward() method to compute the gradients.

# Define a Linear layer
w = Value(3.0, label='w')
b = Value(2.0, label='b')
x = Value(1.0, label='x')

z = (w * x) + b; z.label = 'z'

y = z.relu(); y.label = 'y'

# Visualize the graph
draw_dag(y)

# Call the backward method to compute gradients
y.backward()

# Visualize the graph again
draw_dag(y)