---

Neural network library written in Python completely from scratch, using only numpy.

Usage

from pygrad.tensor import Tensor

a, b = Tensor(np.random.randn(3, 1)), Tensor(np.random.randn(1, 3))
c = a.mul(b)
d = c.exp()
d.backward()

print(a.grad, b.grad, c.grad)
# should output something like
#  [[-16.51232262 -16.51232262 -16.51232262]]
#  [[-53.88103349 -53.88103349 -53.88103349]]
#  [[ 3.50417648  3.50417648  3.50417648]
#   [15.67424587 15.67424587 15.67424587]
#   [16.0131011  16.0131011  16.0131011 ]]

References

• https://github.com/geohot/tinygrad/
• https://github.com/karpathy/micrograd/

Copying

This is free and unencumbered software released into the public domain.