wjtung / homework4-forests

NTU CSIE DSA 2015 Spring Homework 4, Decision Trees and Random Forests

dsa14hw4

Data Structure and Algorithm Homework 4

The first argument of both tree_predictor.cpp and forest_predictor.cpp is the test (prediction) data in the format same as training data, and the output is the accuray of the predictor on the test data.

For example,

./tree_predictor heart.test

Usage of the Makefile.ta:

1. make run_tree FNAME=heart.train EPSILON=0.1:

   run the decision tree on data heart.train with epsilon=0.1.

2. make run_forest FNAME=heart.train T=30:

   run the random forest with 30 trees on data heart.train.

3. make predict_tree PNAME=heart.test:

   make prediction by the decision tree model on data heart.test.

4. make predict_forest PNAME=heart.test:

   make prediction by random forest model on data heart.test.

You can play around with different FNAME, PNAME and T.

We provide two sample data, heart and a1a. The heart data set describes diagnosing of cardiac Single Proton Emission Computed Tomography (SPECT) images. The label means two catogories of patients : normal and abnormal. The features are

1. F1R: continuous (count in ROI (region of interest) 1 in rest)

2. F1S: continuous (count in ROI 1 in stress)

3. F2R: continuous (count in ROI 2 in rest)

4. F2S: continuous (count in ROI 2 in stress)

5. F3R: continuous (count in ROI 3 in rest)

6. F3S: continuous (count in ROI 3 in stress)

7. F4R: continuous (count in ROI 4 in rest)

8. F4S: continuous (count in ROI 4 in stress)

...

The a1a data set describes whether each example (an adult) is of income exceeding $50K/yr.

Both data sets are from

http://www.csie.ntu.edu.tw/~cjlin/libsvmtools/datasets/binary.html

Please feel free to play your trees with more data sets there.