Integrate https://github.com/RubixML/RubixML in Tiki. Rubix ML is a "high-level machine learning and deep learning library"
What is machine learning? Please see: https://docs.rubixml.com/en/latest/what-is-machine-learning.htmlTiki already has mature data management tools
Started in 2002, its source code has been contributed to by more than 350 individuals from multiple organizations.. Now, in close collaboration with the Rubix ML community, we will add the necessary tools and interfaces to become a complete machine learning platform (managing data, choosing a model, training, evaluation, etc.) accessible to power users, like the rest of Tiki.
Both Tiki and Rubix ML are written in PHP, which will facilitate the integration. This is major: Most of the alternatives to Rubix ML are written in Java or Python. We could use them but since they would not be built-in, only a tiny fraction of the Tiki community would have access.
- We are coordinating on the Rubix ML chat room, powered by Telegram: https://t.me/RubixML
- Telegram client apps are Open Source: https://telegram.org/
- Permits multiple new features.
- See "Related" section below for some examples. Many of the these features have been desired for years but we didn't have a clean solution. Both Rubix ML and Tiki have a large feature set and a "one stop shop" philosophy.
- See various sections at https://github.com/RubixML/RubixML/tree/master/src like Anomaly Detection, Classification, Clustering, etc.
On what types of data?
- On Tiki system data (ex.: logins logs) so will be providing insight for all Tiki instances!
- Spammy registrations
- On standard features like forums, wiki page, comments, email, etc.
- Email classification, Spam detection
- On ad hoc data structures made with https://tikitrackers.org/
- Marc (instigator)
- Andrew (Lead dev of Rubix ML) is providing guidance)
- Roberto (developer) will coordinate the project
- Victor (Back-end code) will do initial integration
- Jonny (Front-end code)
- Alain Désilets (advisor)
- Ricardo Melo (advisor)
- Amna Bilal (advisor)
- Simon (junior dev)
- Kevin (junior dev)
- Najia (junior dev)
- Michael I. (tester/requirements for a multilingual project)
We'll start with some simple use cases, like reproducing some of the "Project Spotlight" on https://rubixml.com/, but directly within Tiki.
Performance is very important to train the model. Here is Andrew DalPino, the founder of Rubix ML:
- Mid-July, we'll have an online meeting the Rubix ML and Tiki communities to discuss.
- Read all the documentation
- Do a quick review of all the code base
- Run at least one of the tutorials: https://docs.rubixml.com/en/latest/#tutorials-example-projects
- Read all the open issues
- Contribute https://github.com/RubixML/RubixML/blob/master/CONTRIBUTING.md
Reading all the documentation and even a quick scan of all the source code is an unrealistic goal because the project is huge. So just focus at first on Tiki Trackers
- Read all the content and watch all the videos at https://tikitrackers.org/
- Install Tiki
- You can get source from https://gitlab.com/tikiwiki/tiki instead of tarball/zip
- Explore Tiki features for a few hours
- Build a simple tracker for yourself
- Contribute code to Tiki: Git Workflow
- Think about how we could add a graphical user interface (GUI) to Tiki to leverage of Rubix ML.
- Think about how the Rubix ML demos could be handled within Tiki
- Think about how we can have something like MLT without Elasticsearch: https://github.com/RubixML/RubixML/issues/75
- Machine Learning
- Naive Bayes classifier
- Natural language processing
- Optical character recognition
- Text Mining
- IRC QA Mining
- Natural Language Generation
- Use Cases for NLP and IR in Tiki
- Follow up about "Is PHP Now Suitable For Machine Learning?"