vdinovi / decision_tree Goto Github PK

This project is incomplete due to poor planning, coordination, and understanding of the lab spec. The single program evaluate.py runs cross-validation on the files from the ELECTIONS dataset reporting the following statistics:
- Parameters used (gain ratio, threshold, beta)
- Confusion Matrix
- Precision
- Recall
- F-measure
- Overall accuracy
- Average accuracy

example run: python evaluate.py --gratio --threshold 0.0001 --beta 0.8 ../dataset/domain.xml ../dataset/tree01-1000-numbers.csv

There is no random-forest classifier, it does not handle the IRIS dataset, and is unable to handle numerical attributes. It also does not write the model to a file, instead keeping building it in memory and using that for classification/cross-validation.

The program evaluate.py can be run with the options described below to perform 10-fold cross-validation on generated decision trees.
It reports various statistics based on the result of this cross validation.

The required command line parameters for evaluate.py are:
* [schema_file] xml file that contains the names for the elections dataset (domain.xml)
* [data_file] csv file that contains data in numbers format (tree0X-XXXX-numbers.csv)

The optional parameters that evaluate.py accepts are as follows:
* [-h] for help
* [--gratio] for using information gain-ratio else will use information gain.
* [--threshold VALUE] for specifying the threshold value to be used (DEFAULT=0.01)
* [--beta VALUE] for specifiying the value of beta to use when calculating f-measure (DEFAULT=1)
* [--plot] to plot various values of threshold against information gain and information gain ratio. Note that this generates plot/text files and requires matplotlib.