Platform: Ubuntu

There are two ways to run experiments described in the paper.

A) Use scripts:

1) Ensure that you have all dependencies to build vw.
  https://github.com/VowpalWabbit/vowpal_wabbit/wiki/Dependencies#ubuntu

2) Run run-me.sh.  This will
  a) build vw
  b) download and pre-process existing data sets
  c) generate additional synthetic data sets
  d) run experiments and save results in a results folder


B) Manually go through all the steps:

1) First, in order to create the data sets please download them from the following URL's and save them with the suggested names:

  https://www.openml.org/data/get_csv/150677/BNG_wisconsin.arff       --> BNG_wisconsin.csv
  https://www.openml.org/data/get_csv/150680/BNG_cpu_act.arff         --> BNG_cpu_act.csv
  https://www.openml.org/data/get_csv/150679/BNG_auto_price.arff      --> BNG_auto_price.csv
  https://www.openml.org/data/get_csv/21230845/file639340bd9ca9.arff  --> black_friday.csv
  https://www.openml.org/data/get_csv/5698591/file62a9329beed2.arff   --> zurich.csv

2) Please use the vowpal_wabbit/utl/continous_action/preprocess_data.py to create preprocessed data.

  Note: for the synethic data "ds" you just need to run vowpal_wabbit/utl/continous_action/create_synthetic_data.ipynb instead of the above two steps.

  You then need to put the created data sets in the following path:
  vowpal_wabbit/test/train-sets/regression

3) In order to save the results please create a folder: "vowpal_wabbit/results"

4) For running the online algorithms for CATS as well as the comparators and saving the progressive validation results for each data set you can run:
  vowpal_wabbit/scripts/online_$name.sh

  for running the CATS offline algorithm and saving the loss estimation in SRM and test error results for each data set you can run:
  vowpal_wabbit/offline_$name.sh

  where name = BNG_wisconsin or BNG_cpu_act or BNG_auto_price or black_friday or zurich or ds_5
