Beta Distribution-based Cross Entropy

Beta Distribution-based Cross Entropy (BetaDCE) is a general framework for variable selection in high dimensional data. BetaDCE generalizes well without the requirement of resampling such as cross validation and it can also be used for prediction. Here is a python implementation which can take advantage of multiple CPU cores to significantly accelerate the running time.

Author	Weixing Dai
E-mail	1155038704@link.cuhk.edu.hk

Example

To train:

python betadceTrain.py --data 'data/leukemia_train.csv'

The result of training on leukemia data set:

To predict:

python betadcePredict.py --data 'data/leukemia_test.csv'

The result of prediction on leukemia test set by the trained model:

Usage

betadceTrain.py

Function: Training by Adaptive k-nearest neighbor

usage: betadceTrain.py [-h] [--data D] [--ne N] [--out O] [--cpu C]

optional arguments:
-h, --help　 show this help message and exit
--data D　 data to be trained, the last column is the labels (required)
--ne N 　 maximum number of feature subsets evaluated in each epoch. the default is 2000000 (optional)
--out O 　 filename to keep the output of training. the default is lbsmodel.npz (optional)
--cpu C 　 the number of CPUs to use. the default is to use all of CPUs available (optional)

betadcePredict.py

Function: Prediction on the test set by the trained model

usage: betadcePredict.py [-h] [--data D] [--model M] [--result R]

optional arguments:
-h, --help 　 show this help message and exit
--data D 　 data to be predicted, the last column is the labels (required)
--model M 　 filename of the model trained. the default is lbsmodel.npz(optional)
--result R 　 filename to keep the result of prediction. the default is Result.txt (optional)