Explain the Heuristics and the Naïve Bayesian Classification

[zzj0402]

Hi all,
I am working on a code classification task for my doctorate research. Github Linguist offers an exciting solution to detect coding language as explained here https://github.com/github/linguist/blob/master/docs/how-linguist-works.md . However, I need more details on the Heuristics and on the Naïve Bayesian Classification. Please elaborate:

• What is the Heuristics?
• What are the feature vectors for the Naïve Bayesian Classification?
• What is the training data if any?
• What is the result of the evaluation if any?

Cheers from New Zealand.

[lildude]

• What is the Heuristics?

They're regular expressions. See lib/linguist/heuristics.yml which are then used by lib/linguist/heuristics.rb to find matches in the file in question.

• What are the feature vectors for the Naïve Bayesian Classification?
• What is the training data if any?

You can answer both of these by looking at the code in the classifier lib/linguist/classifier.rb

• What is the result of the evaluation if any?

A programming language 😁

In short the classification works like a funnel. We passing in all languages Linguist knows about (lib/linguist/languages.yml) and then use different strategies to whittle that list down to a single language. The only "machine learning" involved is in the classifier which is a last resort and does quite a mediocre job at it.

We have a PR open to switch out the bayesian classifier for a centroid-based classifier in #5103 which will come at some point in the future.