home | copyright ©2016, tim@menzies.us
overview |
syllabus |
src |
submit |
chat
Take a data set with at least 3 classes.
Build a data generator that randomly
- spits out 1000 rows with the 3 classes at probability 50,50,0%
- then switches to spit out 1000 more rows at probability 10,30,60% (i.e. the third class magically "appears" after 1000 rows).
Build an incremental data reader that reads data in eras of 100 instances at a time.
- When run on the above data, print out the recall at eras 1,2,3,4,5,..,20 (note: at era 11, there should be a dip in the performance).
Write an Naive Bayes classifier
- that learns on eras 1..i
- then tests on era i+1.
- and prints out each predicted class and the log of the likelihood of that prediction
Using the A12 test, can you automatically raise an alert when the log of the likelihoods
of era j are different and worse to era `j-1'?