jedhsu/tensor

Library for tensor manipulation on top of torch.

tensor

tensor{.:python} is a tensor manipulation library.

It provides the following features:

1. Patterns on the cell relationships are parametrized by a Topology{.:python}.

Consider checkerboard pattern. Note how this is a minimal syntax to describe the crux of a pattern, which is what "sets are open".

Consider a convolutional layer. (Run trhough).

2. References at the cell-level for network graph construction.

Features to add: Framework for network architecture, framed as a network topology.

type annotation for arrays

[TODO} try and integrate in networkx complex graph structs

modularization of array operators. Every operator is a class. This enables symbolic expression computation at the type-level (what else?)

Tensor strengthens the type system the operator of numpy arrays. * embeds tensors with a dependent type for their geometry. This can be later extended so a static checker can check that geometry of tensor operators is well-formed. * provides type classes for all the

Helpful because the file system structure helps infer the structure of the operators.

Geometry[11111]

Simple example on dualities.