The program currently works with './Datasets/66_SAMPLED.csv' and './Datasets/66_test.csv'.
To run: All the files are already in there expected directories, therefore, to run simply enter the algorithm's name i.e python algorithmname.py.
- preprocessing.py
- input file: datasets folder containing training.csv, test.csv
- output file: training_new.csv, test_new.csv'
For the classifier algorithms the output file is already formatted with IsBadBuy predictions and refIDs, so any of decision_res.csv, bagging_res.csv, and adaboost_res.csv can be submitted to Kaggle to obtain a score.
- decisiontree.py
- input file: datasets folder containing 66_SAMPLED.csv, 66_test.csv, and ids.csv
- output file: decision_tree_res.csv
- bagging.py
- input file: datasets folder containing 66_SAMPLED.csv, 66_test.csv, and ids.csv
- output file: bagging_res.csv
- adaboost.py
- input file: datasets folder containing 66_SAMPLED.csv, 66_test.csv, and ids.csv
- output file: adaboost_res.csv
Currently bagging.py and adaboost.py is set to a default 5 number of rounds. To change this, visit line 183 for adaboost.py or 134 for bagging.py, containing the line:
ret = train(training_data, attr_names, label_name, num_attr, 5)
Caveat:
My implementation handles both continuous and discrete values. I created a boolean array to note if an attribute is continuous or discrete. After looking at the data set, I hardcoded the indication of continuous or discrete values into the boolean array called 'num_attr' for all algorithm implementation. If it is numeric or continuous, the corresponding array value is set as true. My implementation doesn’t handle training data with missing attribute values, as my preprocessing step already replaced missing values with newly derived ones.
Finally, I didn’t implement tree pruning, therefore, my C4.5 algorithm doesn’t go back through the tree once it's been created and attempt to remove branches that do not help by replacing them with leaf nodes.