Work
- Implemented the loading and saving of a .properties file holding the constants for the algorithms. One default file is loaded from the root of the project folder and can be changed / replaced. For each experiment a folder is created named after the experiment name and a subfolder named after the current time stamp (so that every result of identical experiments are "unique"). This folder is populated with the .properties file used for that experiment, a text file holding the output generated while running the experiment and possibly other output files (like .csv files).
- I looked again into the splitting criteria and read that the best option is to only find the best test and do an f-test to check whether the split is significant (no need to compare two tests to find the most significant test). For this no Hoeffding bound is needed.
- I'm still not sure if checking for a significant split is correctly understood and implemented.
- Running the Six-Hump Camel back does show some splitting but does not do what it should do.
- I should ask if the values for the constants I'm using are any good.
No comments:
Post a Comment