emlearn VS kondo

> My big gripe, and for obvious reasons, is that we need to step away from cloud-based inference, and it doesn't seem like anyone's working on that.

I think there are steps being taken in this direction (check out [1] and [2] for interesting lightweight transpile / ad-hoc training projects) but there is a lack of centralized community for these constrained problems.

[1] https://github.com/emlearn/emlearn

Examples: Gaussian Mixture Models (GMM) for anomaly detection or clustering Mahalanobis distance (EllipticEnvelope) for anomaly detection Decision trees and tree ensembles (Random Forest, ExtraTrees) Feed-forward Neural Networks (Multilayer Perceptron, MLP) for classification Gaussian Naive Bayes for classification

https://github.com/emlearn/emlearn is a ML inference engine for microcontrollers and embedded systems, allowing to deploy models to any platform with a C99 compiler. Has also been used for network traffic analysis as a Linux kernel module, and embedded in Android apps.

The C64 has 64 kB of RAM. That is more than many contemporary microcontrollers. Using something like https://github.com/emlearn/emlearn allows to generate portable C code of ML models for such targets. Should be able to classify digits (MNIST) no problem on such hardware. Assuming there is a workable C compiler available.

Disclosure: Maintainer of emlearn project.

I built emlearn, a Machine Learning inference engine for microcontrollers and embedded systems. It allows converting traditional ML models to simple and portable C99, following best practices in embedded software (no dynamic allocations etc). https://github.com/emlearn/emlearn

You can use cargo sweep or kondo to clean up unused files.

I like to use kondo for that

Blast em https://github.com/tbillington/kondo

Sorry for spoiling your fun, but did you give kondo a try?

This is cool :) I'll have to poke around how you did Async, I've been dragging my feet on implementing it in kondo.