ITree is an innovative web tool designed to facilitate interactive decision-making using classification trees. This intuitive platform supports manual, semi-automatic, and automatic induction of decision trees, catering to both educational and professional needs in the biomedical field. ITree stands out with its ability to allow users to manually modify tree structures, apply various splitting algorithms, and instantly view the impact of these changes on prediction models and testing datasets. Whether you're a researcher, educator, or someone passionate about decision tree induction, ITree offers a versatile, user-driven experience. The full working application can be accessed and tested at https://itree.wi.pb.edu.pl.

Introduction
Installation and Setup
Usage
Technical Details
Contributing
Authors and Acknowledgment
License
References and Links
Appendix/FAQ

Installation and Setup

Prerequisites

Before installing ITree, ensure you have the following prerequisites installed:

Node.js
npm (comes with Node.js)

Installation Steps

Clone the Repository
Navigate to the Project Directory
Install Dependencies

Starting the App

To run the app in the development mode, use:
Open http://localhost:3000 to view it in your browser.

Usage

ITree is an interactive web-based platform that facilitates the induction and manipulation of decision trees using a user-provided dataset. Here's how to harness the full potential of ITree's features:

Initial Configuration

File Upload: Start by uploading your dataset in a CSV format. You can also upload skeleton of decision tree based on our model in JSON. Take a look at script in Python below.
Algorithm Selection: Choose from algorithms like C4.5, TSP, or WTSP for node splitting.
Decision Attribute Configuration: Select the attribute that will be used for classification.
Parameter Settings: Configure essential settings such as minimum node size, maximum tree depth, and entropy threshold.
Draw Tree: With your settings in place, click 'Draw' to generate your initial decision tree.

Python script

In example we used well-known iris set, function is based on 'clf' object so here you can assing your model.

When you upload csv file with your data and json file with your skeleton then aplication will show your tree and will spread the samples over the tree.

-**REMINDER**: Please be aware that both files must have the same attribute names. In other way distribution won't work.

import matplotlib.pyplot as plt
from sklearn import tree
from sklearn.datasets import load_iris
from sklearn.tree import DecisionTreeClassifier
import json

# Load data
iris = load_iris()
X = iris.data
y = iris.target

# Create tree model
clf = DecisionTreeClassifier()
clf.fit(X, y)

# Function for generating json
def node_to_dict(node, feature_names, target_names):
  result = {}
  # Leaf
  if clf.tree_.children_left[node] == -1:
    result['category'] = target_names[clf.tree_.value[node].argmax()]
  # Node
  else:
    feature = feature_names[clf.tree_.feature[node]]
    threshold = round(clf.tree_.threshold[node], 3)
    predicate = "==" if isinstance(threshold, str) else ">="
    weight = clf.tree_.weight[node] if hasattr(clf.tree_, 'weight') else None
    result = {
      'attr2': feature,
      'pivot': threshold,
      'predicateName': predicate,
      'weight': weight,
      'match': node_to_dict(clf.tree_.children_right[node], feature_names, target_names),
      'notMatch': node_to_dict(clf.tree_.children_left[node], feature_names, target_names)
    }
  return result

# Convert root of tree to json
tree_json = node_to_dict(0, iris.feature_names, iris.target_names)

# Save structure of tree to json file
with open('decision_tree.json', 'w') as json_file: json.dump(tree_json, json_file, indent=2)

# Show plot with tree
fig = plt.figure(figsize=(25,20))
_ = tree.plot_tree(clf, feature_names=iris.feature_names, class_names=iris.target_names, filled=True)
plt.show()

Interactive Decision Tree Viewer

Training and Testing Data View: The main view displays the decision tree generated from your training set. You can upload a testing set to evaluate the prediction model's performance.
Tree Node Interaction: Click on any node to fold or unfold branches, allowing for manual pruning or expansion.
Statistics and Metrics: View real-time updates on accuracy and confusion matrix as you interact with the tree.

Manual Adjustments and Tests

Node Modification: Directly adjust the splits in your decision tree by selecting nodes and choosing alternative split strategies.
Test Selection: For each node, you can pick among different tests, like C4.5, TSP, or WTSP, to determine how to best split your data.

Real-time Data Analysis

Confusion Matrix: Access the confusion matrix from the interface to understand the accuracy of your tree.
Data Viewer: Inspect the distribution of your data and the classification results within the nodes of your decision tree.

Experimentation and Learning

Interactive Learning: ITree serves as an educational tool, allowing users to experiment with different decision tree configurations and learn about their effects on data classification.

Technical Details

Code Organization

The ITree codebase is structured to facilitate easy navigation and understanding. Key components are organized in separate directories.

Available Scripts

npm start: Runs the app in development mode. Access it at http://localhost:3000.
npm test: Launches the test runner in interactive watch mode.
npm run build: Builds the app for production to the build folder. It optimizes the build for performance.
npm run eject: Removes the single build dependency from your project, copying all configuration files and transitive dependencies into your project.

Dependencies

ITree utilizes several key libraries and frameworks, including React. The detailed list of dependencies can be found in the package.json file.

Contributing

We warmly welcome contributions to the ITree project. Whether it's feature requests, bug reports, or code contributions, here's how you can contribute:

Guidelines

Reporting Issues: Please use the GitHub issue tracker to report bugs or suggest features.
Submitting Pull Requests: For code contributions, please fork the repository and use a new branch for your work. Pull requests are eagerly awaited.

Contact Info

For major changes or discussions in context of research/development please contact m.czajkowski@pb.edu.pl.
For major changes or discussions in context of software please contact hubert.sokool@gmail.com to align with the project's goals and standards.

README.md

ITree: Interactive Decision-Making with Classification Trees

Introduction

Table of Contents