Engine

API Documentation for the automatic differentiation engine of nanograd

Automatic differentiation engine

class nanograd_bgriebel._core.engine.Value

A wrapped float which can be used in computing gradients.

Parameters:

data (float) – Data to wrap in the Value.

backwards(self: nanograd_bgriebel._core.engine.Value) → None

Compute the gradients of a Value.

Uses backpropagation to calculate the dertivative of this Value with respect to any Vaues which were used to generate it.

Examples

>>> # Create some Values
>>> x = Value(3)
>>> y = Value(4)
>>> # Perform a computation with the Values
>>> z = x*y
>>> # Calculate the gradient of z wrt x and y
>>> z.backwards()
>>> print(x.grad)
4
>>> print(y.grad)
3

property data

Data wrapped by the Value.

Type:

float

property grad

The gradient associated with the Value

Type:

float

relu(self: nanograd_bgriebel._core.engine.Value) → nanograd_bgriebel._core.engine.Value

Calculate the output of a ReLU on the Value.

Returns:

The output of the ReLU operation wrapped into a Value.

Return type:

Value

zero_grad(self: nanograd_bgriebel._core.engine.Value) → None

Set the value of grad to 0.0