satoshizpy/cto

semi-supervised learning

###Optimally combining classifiers for semi-supervised learning

We propose a new semi-supervised method combing Xgboost and transductive support vector machine.

Experiments on 14 UCI data

Requirements

• Python 3.6+
• pandas
• matplotlib
• numpy
• Pycharm

Train

Obtain the visualization of the real data by T-SNE

python DataDistribution.py

Compare the diversity of Xgboost, TSVM, DecisionTree

python compare_diversity.py

Run the proposed method for 14 real data

python new_algorithm_test

Results (Accuracy)

Data	cjs	hill	segment	wdbc	steel	analcat	synthetic
Accuracy	0.98	0.499	0.925	0.954	0.649	0.993	0.92

Results (Accuracy)

Data	vehicle	german	gina	madelon	texture	gas-grift	dna
Accuracy	0.625	0.716	0.857	0.543	0.953	0.965	0.911