Architecture

The Machine Learning model is broken up into a number of key components, derived from freely-available helper apps within Akumen. These components are:

1. Delimited Data Connector: retrieves the data from the web and automatically decomposes it into an Akumen-consumable resultset;
2. Customised PETL: takes the resultset and performs a few modifications (mostly as an example of how heavier transformations can be performed);
3. Auto-SKLearn: uses the resultset with some configuration items to perform automated machine learning on the dataset and build a trained model; and
4. SKLearn-Executor: loads the trained model and, given a correct input set, returns the answer.

Not every machine learning application needs these four components: some can be three. If you’re training a model to determine accuracy and not execute on an ongoing basis, you can use two.