Pattern recognition
projects
Nov. 1st, 2018
Data
• 6 datasets, dataset01-dataset06

• 3 artificial problems (1-3), with large number of samples, files
*.npz (use numpy.load to read the files); 

• within each file, there is a matrix “X” (samples by rows)
and a vector “y” (for labels)

• 3 real-world problems (4-6): for each problem, there are 2
files (*.dat) with the data matrix X (*-train-X.dat) and the
labels (*-train-y.dat). Use numpy.loadtxt for reading. Data
matrix dimensions: 400x14,883; 450x14,883; 450x14,883
(samples by rows)
Rules
• Model each problem separately, using the classifier(s) of your choice

• Once you believe you have the “right” model, train the final classifier and estimate
its performance in terms of (at least): error rate, sensitivity and specificity and
provide for each 95% confidence intervals

• The test sets will be released on Dec. 6th: by the end of that day:

• you should send me a brief description (1 page/slide per model - best: an
IPython notebook file *.ipynb) of your modeling approach (a schema would
do) and the claimed performance (and 95% CIs)

• you should send me your predictions for each of the 6 problems in the form of
a text file with 1 label per row

• On Dec. 13th I will release the evaluations of your models: performance and how
well you estimated it.