from mnist import MNIST from mnist import save2disk from mnist import open4disk import numpy import pickle import bz2 # bzip2 import gzip # gzip if __name__ == "__main__": mn = MNIST("./data") print(type(mn.get_training()[0]), type(mn.get_training()[1])) print(mn.get_training()[0].shape, mn.get_training()[1].shape) print(type(mn.get_testing()[0]), type(mn.get_testing()[1])) print(mn.get_testing()[0].shape, mn.get_testing()[1].shape) print(type(mn.get_validation()[0]), type(mn.get_validation()[1])) print(mn.get_validation()[0].shape, mn.get_validation()[1].shape) print(type(mn.get_training()[0][1, :])) print(mn.get_training()[0][1, :].shape) mn.info save2disk(mn, compression='bzip2')
print(mn.train_images.shape) validation_images = numpy.asarray(mn.train_images) validation_labels = numpy.asarray(mn.train_labels) cantidad_validacion = 10000 validation_images = mn.train_images[0:cantidad_validacion] validation_labels = mn.train_labels[0:cantidad_validacion] print(validation_images.shape) print(validation_labels.shape) """ mn.load_data() print(type(mn.get_training()[0]), type(mn.get_training()[1])) print(mn.get_training()[0].shape, mn.get_training()[1].shape) print(type(mn.get_testing()[0]), type(mn.get_testing()[1])) print(mn.get_testing()[0].shape, mn.get_testing()[1].shape) print(type(mn.get_validation()[0]), type(mn.get_validation()[1])) print(mn.get_validation()[0].shape, mn.get_validation()[1].shape) ''' # esta re balanceada la base de datos, si tomo los primeros 10000 # estan uniformemente repartidos de los digitos 0..9; # sino probar ;) ## elijo uniformemente (en el monton mas o menos elegi un poco "igual" de cada una) ## etiquetas al azar, y de ahi saco para formar mi conjunto de validacion ## osea voy tachando de aca