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
