🏴‍☠️ 🍠 ⚔️ Arghhhh....!!! Cpt. Postato says you're standing on the fuzzy plank 🏴‍☠️

Postato is a tool that identifies the position of your body with fuzzy logic.

Fuzzy inference system

The fuzzy rule set is built upon crisp data which is accepted under the form of a dataset where each tuple is in the format (xWrist, yWrist, zWrist, xThigh, yThigh, zThigh, activity). The data used is provided by the UCI data repo. It is first fuzzified with soft kMeans++ using Newton's gravity formula producing a super cluster of fuzzy clusters. Once they are obtained a fuzzy rule is generated from the fuzzy boundaries of each cluster. A rule consists of a mapping between a body position axis and a fuzzy number. Currently Postato supports triangular and gaussian fuzzy numbers.

But how is this valuable? 🤔

A fuzzy inferer is required to make use of a fuzzy rule set. Postato uses the one of Mamdani.

How to use?

At the moment only fuzzy rule set plotting is supported. It can be executed with:

# git clone this repo
# Fetch all dependencies.
go get ./...

# If you have direnv installed run `direnv allow` otherwise
source .envrc

# For gaussian fuzzy numbers
go run cmd/postato/main.go draw -d data/sample.csv -t gaussian

# For triangular fuzzy numbers
go run cmd/postato/main.go draw -d data/sample.csv -t triangular

All images are generated within the gen/image directory. There is an example dataset located at data/sample.csv.

Accuracy

It is tested using 10-fold-cross validation which can be tested like executed like this:

# For gaussian fuzzy numbers. Shows a success rate of ~82%.
go run cmd/postato/main.go test -d data/sample.csv -t gaussian

# For triangular fuzzy numbers. Shows a success rate of ~12% which is far worse.
go run cmd/postato/main.go test -d data/sample.csv -t triangular