Bart Massey
This file contains the instance data and writeup for the heart anomaly homework given in my Intro AI class, together with the writeup and the original paper describing the dataset.
-
heart.pdf: Assignment writeup -
spect.pdf: Original paper -
spect-orig.*.csv: Binarized training and test data from the original paper. The dependent variable is in the first column. -
spect-resplit.*.csv: Original binarized data resplit proportionally to prevalence in the sample. Not clear that this is the same as prevalence in the population. Training / test instances are split 2::1. -
spect-itg.*.csv: Features and class are taken from the continuous version of the original data, and binarized for maximized information-theoretic gain. Same training and test instances as original. Both the training and test instances are used in the binarization, which perhaps improves it unfairly. -
spect-resplit-itg.*.csv: Features and class are taken from the continuous version of the original data, and binarized for maximized information-theoretic gain. Resplit as above.
The original resources are also available:
-
SPECT: Binarized data from the original paper. This repo contains a copy of the data.
-
SPECTF: Continuous version of the data from the original paper. This repo contains a copy of the data.
The binarization was via itg.py. The resplit was via resplit.py.