Quickstart

Install

git clone https://github.com/Armangb1/pygorch.git
cd pygorch
pip install -e .            # runtime (NumPy)
pip install -e ".[dev]"     # + pytest, pytest-cov, ruff for development

Tensors and gradients

import numpy as np
import gorch

x = gorch.Tensor(np.array([[0.0], [1.0]]), requires_grad=True)
W = gorch.Tensor(np.random.randn(1, 1), requires_grad=True)

y = (x @ W).tanh()
y.sum().backward()

print(W.grad.value)  # shape matches W.value

Scalar outputs can call backward() with no arguments. For non-scalar outputs, pass an explicit gradient seed:

y.backward(gorch.Tensor(np.ones_like(y.value)))

Train a small network

model = gorch.nn.Sequential(
    gorch.nn.Linear(2, 8),
    gorch.nn.Tanh(),
    gorch.nn.Linear(8, 2),
    gorch.nn.Softmax(),
)
loss_fn = gorch.nn.CrossEntropyLoss()
opt = gorch.optim.Adam(model.parameters(), lr=0.05)

x = gorch.Tensor(X)
target = gorch.Tensor(Y)

for _ in range(800):
    opt.zero_grad()
    pred = model(x)
    loss = loss_fn(pred, target)
    loss.backward()
    opt.step()

Save and load

state = model.state_dict()     # dict of parameter names -> numpy arrays
model.load_state_dict(state)   # restore weights in place

model.save("model.pkl")        # pickles the state_dict
model.load("model.pkl")        # and back

Run the examples

python examples/xor_mlp.py        # MLP learning the XOR function
python examples/neural_control.py # Kalman filtering + EKF/RLS system identification

Run the tests

pip install -e ".[dev]"
pytest