API Overview

jamgrad exposes a compact API centered on Tensor plus a small neural-network helper module.

Import surface

from jamgrad import Tensor
from jamgrad import nn

Tensor (jamgrad.tensor)

Core features:

• Arithmetic with autograd support: +, -, *, /, **, unary -
• Matrix multiplication: @
• Unary ops: .exp(), .log(), .sum()
• Backpropagation entrypoint: .backward()
• Graph labeling and visualization: .set_label(), .to_dot()

nn module (jamgrad.nn)

• Linear(in_features, out_features)
• relu(x)
• softmax(x)
• cross_entropy_loss(predictions, targets)

Quick example

from jamgrad import Tensor
from jamgrad.nn import Linear, relu

x = Tensor([[1.0, 2.0]], requires_grad=False)
layer = Linear(2, 1)
y = relu(layer(x))