Determine the components that need to be implemented

[mikesiers]

Issue Overview

Decision tree algorithms are comprised of several independent components such as pruning and splitting criteria. Additionally, the same is true for decision tree data structures. For example, splitting a parent into children and specifying leaf nodes.

This is the first work that will be done on wattle. Therefore, it is important to clearly define what all of these components are now. After these have been identified, they should also be represented visually in some sort of illustration.

Tasks to Close Issue

• [x] Identify all the required components. They can be posted in a reply to this issue.
• [x] Generate the illustration. yEd is suitable for this purpose. A .graphml and .png should be added to the repo in a documentation folder.

[mikesiers]

Components

Initially, I thought that it would be helpful to think of the required components in two different categories. Data structure components and algorithm components. However, after determining the structure of these components, I'm convinced that this separation of implementation is not useful. Therefore, I have written the components as follows:

• Node
  • flag describing whether it is a leaf.
  • flag describing whether it is the root.
  • a reference to the training data points which belong to this node.
  • function which takes a test as input and creates children based off the node + test. The test is a function.
  • function for pruning which takes a pruning criterion as input.
  • function for getting the class supports.
  • a toString function.
• Branch
  • a test represented as a string. For display and record keeping purposes only.
  • a reference to parent
  • a reference to child
  • a toString function.
• Tree
  • A reference to the root node.
  • A node count.
  • Function for pruning which takes a pruning criterion as input.
  • Function for finding a node based on logic rule object.
  • Function for getting all logic rules. Logic rules are represented as strings, they are in the form: "IF condition1 AND condition2 ... AND conditionX THEN (support1, support2, ... , supportY)"
  • Function which takes data as input and builds the decision tree. This function should also take a flag as input which describes whether or not the tree should be pruned. This function should also take a dictionary of custom arguments such as {'min_leaf_records' : 10, 'min_gain_ratio' : 0.9}.
  • A toString function.

[mikesiers]

The work in commit 59a4812 was enough to close this issue. It should have been closed via the commit message.