DecisionTree

A decision tree implemented by Hunt's Algorithm.

Dataset

• training set : adult.data
• evaluation set : adult.test
• Listing of attributes
  • label : >50K, <=50K
  • attribute: age, workclass, fnlwgt, education, education-num, marital-status, occupation, relationship, race,sex, capital-gain, capital-loss, hours-per-week, native-country, income

Usage

You can run the code with the following command :

python main.py

then you can train a new tree and evaluate this decision tree using adult.test by entering following operations and All results will be displayed on the terminal :

• T for training new tree and save
• E for using current tree to evaluate (which will be error if no tree load or train before)
• Q for quit

File Structure

Source Codes

• load.py : preprocess the raw data,remove specified columns and missing values,use binary to read from tree.pickle or write decision trees into tree.pickle
• train.py:implemente the Hunt's algorithm to create Decision tree, it can process both ordinal attributes and nominal attributes.
• evaluate.py:evaluate this decision tree using adult.test
• struture.py:define decision tree node
• main.py: execute and display result

Documentation

1. • tree.pickle:decision trees can be written into tree.pickle
   • output.txt: visualize binary decision tree and describe the decision tree built from the Adult training set
   • res.csv: the test dataset and it's evaluation result, the last column represents whether this row evaluated correctly
2. • the decision tree built from the Adult training set can be found in the output of train.py, it shows the best split attribute and its GINI value for each iteration.
   • evaluation correct number: 12694
   • evaluation correct rates: 0.8428950863213811