Data Multiplication

The automatic data multiplication works like this:

for algorithm in multiplication_queue:
    new_training_set = []
    for recording in training_set:
        samples = algorithm(recording)
        for sample in samples:
            new_training_set.append(sample)
    training_set = new_training_set
return new_training_set

Warning

The create_ffile procedure replaces the current set of recordings by the set returned by the data multiplication steps.

Recording rotated by -30, 0, +30 degrees

Data multiplication algorithms.

Each algorithm works on the HandwrittenData class. They have to be applied like this:

>>> import data_multiplication as multiply
>>> a = HandwrittenData(...)
>>> multiplication_queue = [multiply.copy(10),
                           mulitply.rotate(-30, 30, 5)
                           ]
>>> x = a.multiply(multiplication_queue)
class hwrt.data_multiplication.Multiply(nr=1)

Copy the data n times.

class hwrt.data_multiplication.Rotate(minimum=-30.0, maximum=30.0, num=5)

Add rotational variants of the recording.

hwrt.data_multiplication.get_data_multiplication_queue(model_description_multiply)

Get features from a list of dictionaries

>>> l = [{'Multiply': [{'nr': 1}]},              {'Rotate': [{'minimum':-30}, {'maximum': 30}, {'step': 5}]}]
>>> get_data_multiplication_queue(l)
[Multiply (1 times), Rotate (-30.00, 30.00, 5.00)]

Previous topic

Preprocessing

Next topic

Features

This Page