Usage: $ python test.py stateFilename outdir (testFile.dat | inkmldir lgdir)

test.py takes a filename for a serialized classifier, an output directory, and either a pickled testing data set or directories for inkml and lg files to test. The lg files are used to ensure that the files created in the output directory are as expected.

After it has run, it will calculate it's recognition rate, and the new lg files will be in outdir.

Usage: $ python train.py (-nn|-rf|-et|modelFilename) outFilename (inFilename.dat | inkmldir lgdir)

train.py takes either a pickled, untrained classifier or an arguemet telling it to create a 1-NN, random forest or extra trees cassifier with our set options. It also takes an output filename (we've been using '.mdl' with these) and either a pickled training data set or inkml and lg directories.

After it hasbeen run, a trained classifier while have been serialized to outFilename.

Usage: $ python split_inkmls.py (file.inkml | inkmlDirectory) lgdir trainingDir testingDir trainingLgDir testingLgDir [trainingPerc]]

OR

$ python split.py (file.inkml | inkmlDirectory) lgdir trainingFilename [testingFilename [trainingPerc]]

The former takes a directory full of inkml files (which may be in subfolders) and a directory of lg files (which are assumed to contain at least those for all the inkml files) and splits them up more or less evenly into seperate directories which really ought to start out empty.

The latter instead serializes them to a pair of files, one for training, one for testing.