purpose

Learning machine learning with Ruby one failing spec at a time.

dataset

This experiment uses the problem of learning which names belong to humans and which do not. It attempts a few naive algorithms going from most to least in order:

• dunce - learns nothing ever.
• space_dunce - thinks that spaces are all of the rage
• spaces - thinks that spaces are cool as long as they match the range learned from data
• ngram - currently a 2 gram that analyzes provided names and builds a map of ngram => human likelihood score association and then uses this map to score new names

All of the learners ask for more input and then attempt to adjust based on the score that you as a human feel those names deserve. Except for the dunce; that one learns nothing.

building

Build this project image using Docker.

  docker build --tag human_or_not:latest .

running

Running an NGram learner with the dataset of names use the following command.

  docker run --rm -it human_or_not:latest  # would drop you into running an NGram learner with a dataset of names
  ruby run.rb --learner=ngram --dataset=names
  # or use a different data set
  ruby run.rb --learner=ngram --dataset=guildies
  # or use a different lolgarithm for learning
  ruby run.rb --learner=spaces --dataset=guildies

obviously data

machine learning heavily depends on the data that you provide.

Even the simplest case of a spaces learner if you provide data that classifies "abced asdf" as a 0.0 relevance then "Bob Hope" would also be 0.0 relevance.

In cases of using the ngram learner you should play around to see how the data you provide when you interact with the learner by running it (see: running)