Starting with pendigits.data and pendigits.train

% cd pendigits
% ../iwal-split 0.1 pendigits.data

1) creates a random bootstrap set B containing 0.1 fraction of pendigits.data
	pendigits.bootstrap
2) resamples 10 .boot datasets from B, including each example in B 
with probability 2/3 (constants 10 and 2/3 are hardcoded)
	bootstrap/pendigits-N.boot where N ranges from 0 to 9
3) the rest of pendigits.data is put into pendigits.stream 
(9/10 of pendigits.data)

--------------------------------------------------------------------------
Prefix k- indicates the k-class version of the code (originally written
for binary classification)
--------------------------------------------------------------------------
Next steps (all scripted in k-driver, which takes a single argument, the name
of the data directory):

- Run weka on the 10 binary datasets to create 10 classifiers

% ./script pendigits

- Use these classifiers to rejection sample pendigits.stream

% ./k-iwal 10 pendigits/trees.J48/pendigits pendigits/pendigits > pendigits/active.errors

k-iwal runs k-script-costing to learn an importance-weighted classifier

- Comapre a passive learner trained on the entire pendigits.data to an active
  importance-weighted classifier

% ./passive pendigits/pendigits > pendigits/pendigits.errors
--------------------------------------------------------------------------

To run k-driver (in the main directory):

% ./k-driver pendigits
