Two Approaches to Extract Logical Rules for Mushroom Edibility: Neural Networks and Genetic Algorithm

Project for ANU COMP4660/8420 (Bio-inspired Computing: Applications and Interfaces), Semester 1 2018.

By Yuanbo Han, 2018-05-31. See the project report here.

Environment

• Python 3.6.3
  • numpy 1.14.3
  • matplotlib 2.2.2
  • pandas 0.22.0
  • torch 0.4.0
  • sklearn 0.19.1
  • pydotplus 2.0.2
• graphviz 2.40.1

Note that the above are just versions during experiment, not the least requirements.

Data Set

Mushroom Data Set/agaricus-lepiota.data.csv

Original source: UCI Machine Learning Repository

Codes

• bpNN.py
• decisionTree.py
• displayWeight.py
• GATree.py
• load_data.py

BP Neural Network

Run bpNN.py. It will read in the data, perform discretization, train a back-propagation neural network, and generate a file called "net_weights" which stores the weights in the model. To adjust parameters, see line 14~26. To change the network structure, see line 29~35.

Run displayWeight.py. It will read "net_weights" file and print the network weights for attribute values.

GA + Decision Tree

Run GATree.py. It will read in the data, perform Genetic Algorithm for feature selection, and generate a "tree.pdf" which is the diagram of the final Decision Tree. Control parameters can be adjusted in line 6~13.