chengyaowang/ee660-machinelearning

Source Code for Machine Learning Project

EE660 Project

Affiliated course: USC EE660, Machine learning: Foundations & Methods, given in 19 fall semester

Used Dataset: Breast Cancer Wisconsin (Diagnostic)

Problem Type: Binary Classification

Methods Involved:

• Preprocessing & Visualization:
  • Scatter / Box plots features with the top-5 CV
  • Standardization
  • Feature correlation heat matrix
  • Per-feature class distribution plotting
  • Linear / Kernel PCA
• Part1, Finding Best Performance Models:
  • K-nearest Neighbor
  • AdaBoost & Boosting Forest (hierarchical grid search)
  • Elastic Net Logistic Regression (grid search)
  • K Means Clustering + Logistic Regression (hierachical)
• Part2, Experimental Explorations:
  • Monte Carlo Semi-supervised l1 support vector machine
  • Spectral Clustering (grid search)

Code Organization:

• Main.py: Highest level script for generating results
• dataGen.py: Class "dataGenerator" for data reading, preprocessing, visualization & data fetching
• modules.py: Class "models" containing all training & testing process of models used in this project
• utils.py: Class "util" for all graph plotting, dataset generating / fetching for models, result / log printing.

Dependencies: Python: 3.7.4

Scikit-learn: 0.21.3

Pandas: 0.25.1

Seaborn: 0.9.0

Other: numpy, datetime, shutil, os, statistics, matplotlib.pyplot, random