Classification Neuron

A demo to classify points in a graph to a given group using linear regression.

Commit HEAD compiled with Zig 0.10.1.

Compilation

Docker:

$ docker build -t neuron .
$ docker run --rm -it -v "$(pwd)/Demo:/mnt" neuron
neuron% # Demo/ is now mounted to /mnt
neuron% classifier # <args...>

Native (Zig compiler):

$ zig build
$ ./zig-out/bin/classifier # <args...>

Or alternatively, use the run step of the Zig build system:

$ zig build run -- # <args...>

Tests

Unit tests are included for the training implementation:

$ zig build test

Usage

See Demo/, Demo/dataset_0/ and Demo/run/points_0/ for details about the in-/output formats.

Train model

Train: from an annotated points (x/y/group) dataset, trains the model using linear regression and outputs the model weights to stdout.

classifier train <annotated points>

$ classifier train dataset_0/annotated.json # dry run
$ classifier train dataset_0/annotated.json > dataset/weights.json # save output

Graph rendering

SVG: from any type of points file, converts it to an SVG format and outputs it to stdout.

classifier svg <any points>

$ classifier svg run/points_0.json # dry run
$ classifier svg dataset_0/annotated.json # dry run, colours annotated points
$ classifier svg dataset_0/annotated.json > dataset_0/annotated.svg # save output

Model validation

Test: from a weights file generated with classifier train, runs a points file through the model and compares the calculated result with the associated annotation to see if it's correct. It outputs the log to stdout.

classifier test <weights> <annotated points>

$ classifier test dataset_0/weights.json dataset_0/annotated.json # through trained dataset
$ classifier test dataset_0/weights.json dataset_0/annotated.test.json # through unseen dataset

Model run

Run: from a weights file generated with classifier train, runs a directory of points files through the model and outputs it to <directory>/<dataset>/<image>/ along with the SVG of both in-/output.

classifier run <weights> <points directory>

$ classifier run dataset_0/weights.json run/ # output written to run/dataset_0/*/...

QSmally/Classification-Neuron