torchhd

torchhd copied to clipboard

The current `soft_quantize` and `hard_quantize` need to be replaced in favor of a model specific implementation of the precision and quantization for each model.

mikeheddes

Another way to look at the thermometer codes is to take the first hypervector as the identity hypervector of the model and the last hypervector as the negative of the...

mikeheddes

It would be good to support TorchScript so HDC models can be exported to be run in a production environment.

mikeheddes

I think we should work on adding a getting started page to the documentation for people unfamiliar with Python or PyTorch since there are also many other researchers that use...

mikeheddes

We would like to implement as part of this library the `train`, `test`, and `predict` utility functions like those provided in: https://github.com/rishikanthc/torch-hd/blob/main/torch_hd/utils.py. This issue serves as the discussion thread for...

mikeheddes

The current implementation of the binary tree requires knowing and providing the path of a node. Look into classical computing APIs of binary trees to see how to design a...

mikeheddes

This issue is to discuss design improvements around the access to the exact previous hypervector value in the data structures. Right now we require the user to pass the previous...

mikeheddes

Since text data is often used, we can add a wrapper around the Random embeddings to store a mapping of characters to random hypervectors (`Text(chars, padding=" ")`).

mikeheddes

Based on [A Theoretical Perspective on Hyperdimensional Computing](https://www.jair.org/index.php/jair/article/view/12664) we could presumably calculate the ideal threshold so that the user won't have to specify it.

mikeheddes

Is it possible to "OR" two vectors?

9

I'm fairly new to HDC and trying to wrap my head around how it works. There's a really good example out there that shows binding with multiplication (SHAPE is CIRCLE,...

uler3161