Create feature files

Create feature files.

Before this script is run, the download.py should get executed to generate a handwriting_datasets.pickle with exactly those symbols that should also be present in the feature files and only raw_data that might get used for the test-, validation- and training set.

hwrt.create_ffiles.get_parser()

Return the parser object for this script.

hwrt.create_ffiles.get_sets(path_to_data)

Get a training, validation and a testset as well as a dictionary that maps each formula_id to an index (0...nr_of_formulas).

Parameters:

path_to_data :

a pickle file that contains a list of datasets

hwrt.create_ffiles.main(feature_folder, create_learning_curve=False)

main function of create_ffiles.py

hwrt.create_ffiles.make_hdf5(dataset_name, feature_count, data, output_filename, create_learning_curve)

Create the hdf5 file.

Parameters:

filename :

name of the file that hdf5_create will use to create the hdf5 file.

feature_count : integer

number of features

data : list of tuples

data format (‘feature_string’, ‘label’)

hwrt.create_ffiles.prepare_dataset(dataset, formula_id2index, feature_list, is_traindata, do_normalization=False)

Transform each instance of dataset to a (Features, Label) tuple.

hwrt.create_ffiles.training_set_multiplication(training_set, mult_queue)

Multiply the training set by all methods listed in mult_queue.

Parameters:

training_set :

set of all recordings that will be used for training

mult_queue :

list of all algorithms that will take one recording and generate more than one.

Returns:

mutliple recordings