serendipious / nodejs-decision-tree

NodeJS Implementation of Decision Tree using ID3 Algorithm

Decision Tree for Node.js

This Node.js module implements a Decision Tree using the ID3 Algorithm

Installation

npm install decision-tree

Usage

Import the module

var DecisionTree = require('decision-tree');

Prepare training dataset

var training_data = [
  {"color":"blue", "shape":"square", "liked":false},
  {"color":"red", "shape":"square", "liked":false},
  {"color":"blue", "shape":"circle", "liked":true},
  {"color":"red", "shape":"circle", "liked":true},
  {"color":"blue", "shape":"hexagon", "liked":false},
  {"color":"red", "shape":"hexagon", "liked":false},
  {"color":"yellow", "shape":"hexagon", "liked":true},
  {"color":"yellow", "shape":"circle", "liked":true}
];

Prepare test dataset

var test_data = [
  {"color":"blue", "shape":"hexagon", "liked":false},
  {"color":"red", "shape":"hexagon", "liked":false},
  {"color":"yellow", "shape":"hexagon", "liked":true},
  {"color":"yellow", "shape":"circle", "liked":true}
];

Setup Target Class used for prediction

var class_name = "liked";

Setup Features to be used by decision tree

var features = ["color", "shape"];

Create decision tree and train the model

var dt = new DecisionTree(class_name, features);
dt.train(training_data);

Alternately, you can also create and train the tree when instantiating the tree itself:

var dt = new DecisionTree(training_data, class_name, features);

Predict class label for an instance

var predicted_class = dt.predict({
  color: "blue",
  shape: "hexagon"
});

Evaluate model on a dataset

var accuracy = dt.evaluate(test_data);

Export underlying model for visualization or inspection

var treeJson = dt.toJSON();

Create a decision tree from a previously trained model

var treeJson = dt.toJSON();
var preTrainedDecisionTree = new DecisionTree(treeJson);

Alternately, you can also import a previously trained model on an existing tree instance, assuming the features & class are the same:

var treeJson = dt.toJSON();
dt.import(treeJson);

nodejs-decision-tree's Issues

Support Continuous Feature & Target Variables

• Establish types for Feature & Target variables
• Determine best split for feature variables using modern techniques

The package don't allow import the model already trained

We cannot load one model already trained. A cool feature is to import a already trained model.

Implement support for Random Forests

Decision Tree Error

There's an error here.

decision-tree.js:20
root = childNode.child;
^

TypeError: Cannot read property 'child' of undefined
at Object.ID3.predict (/nodejs-decision-tree-id3-master/lib/decision-tree.js:20:23)
at Object. (/nodejs-decision-tree-id3-master/example.js:22:34)
at Module._compile (module.js:434:26)
at Object.Module._extensions..js (module.js:452:10)
at Module.load (module.js:355:32)
at Function.Module._load (module.js:310:12)
at Function.Module.runMain (module.js:475:10)
at startup (node.js:117:18)
at node.js:951:3

Implement input validation

Offload tree node storage off-memory

Move training to a separate method

Implement support for online/incremental learning

Break decision-tree source into multiple components

• Break plain objects into classes
• Move methods into separate modules
• Use classes to parse and validate training and test data-tables

Saving then importing a model gives an error

When saving a model with .toJSON, and later importing it, it generates the following error:

TypeError: Cannot read property 'length' of undefined

I triple checked multiple times, this error is very real.

Thank you 😍

Just want to say thank you, you made my life easier! God bless. It's a beautiful piece of code! 🙏🏻

Export tree for visualisation

How would I go about exporting the underlying decision tree so that I could visualise it in d3 or something similar?

Build de tree

Hi, I would like to know how can I build a visual representation of tree.
I saw how to predict but I can't understand how can I build the tree.

Subsequent declarations screw up the previous models

Hello, when I have 1 model, it works great (dynamic_dt below predicts correctly), but when I declare multiple ones (dynamic_dt, mde_dt, ...etc) the results are not accurate. What is happening here, how can I fix this?

Missing data

How does this algorithm handle cases of missing data in the dataset?