Week 11: Machine Learning Methods and Final Presentation

This week, was adding some more Machine Learning Methods, and a final Presentation for the work I’ve done through this period!

Adding More Machine Learning Support

A set of other

A set of other training methods were also added. GLM was decided to be split into some of its more commonly used components. This expands the support to also include:

  1. (Regression) Random Forest
  2. Binomial Logistic Regression
  3. GLM
  4. DBScan
More model types!

This, is rather easy to do now that we have added declarations; there is a standard ML_Core.dhsql memory file, and this provides a nice way to provide built-in declarations.

Going from here, following the declaration syntax, we can add in a lot of the methods:

Here is a declaration for Classification forest

Predict-only methods

DBScan is a good one, as it is an unsupervised Machine Learning technique that does not fit into the traditional ‘train-predict’ idealogy. Here, the idea is have a variant of the predict statement that captures a syntax similar to train. Let’s model the statement as:

predict from <ind> method <methodtype> [ADD ORDER] [<trainoptions>];

Adding in the grammar, as most of the work is similar to train, we can add this in, and voila! A new statement for this form of machine learning!


Final Presentation

This is where a lot of the time was spent; I’ve been gathering and making examples for HSQL, and how it can be used, and on Thursday, we had a presentation on HSQL, what’s new and how it can be useful! (I will post more about it when I can!)

Wrapping up

As week 11 comes to an end, there are some things that may require documentation, and those are being looked at; and that’s about it! We still have to add some examples for the easy to use machine learning statements, and maybe also look at some code generation suggestions. (and perhaps even try to implement one!)