def get_data(conf):
"""Return data to be used in this session.
Args:
conf: Configuration dictionary
Returns:
X_train: numpy array of floats with shape [input_dimension, num train examples] in [0, 1].
Y_train: numpy array of integers with shape [output_dimension, num train examples].
X_devel: numpy array of floats with shape [input_dimension, num devel examples] in [0, 1].
Y_devel: numpy array of integers with shape [output_dimension, num devel examples].
X_test: numpy array of floats with shape [input_dimension, num test examples] in [0, 1].
Y_test: numpy array of integers with shape [output_dimension, num test examples].
"""
data_dir = os.path.join(conf['data_root_dir'], conf['dataset'])
if conf['dataset'] == 'cifar10':
conf['input_dimension'] = 32*32*3
conf['output_dimension'] = 10
X_train, Y_train, X_devel, Y_devel, X_test, Y_test = import_data.load_cifar10(
data_dir, conf['devel_size'])
elif conf['dataset'] == 'mnist':
conf['input_dimension'] = 28*28*1
conf['output_dimension'] = 10
X_train, Y_train, X_devel, Y_devel, X_test, Y_test = import_data.load_mnist(
data_dir, conf['devel_size'])
elif conf['dataset'] == 'svhn':
conf['input_dimension'] = 32*32*3
conf['output_dimension'] = 10
X_train, Y_train, X_devel, Y_devel, X_test, Y_test = import_data.load_svhn(
data_dir, conf['devel_size'])
conf['layer_dimensions'] = ([conf['input_dimension']] +
conf['hidden_dimensions'] +
[conf['output_dimension']])
if conf['verbose']:
print("Train dataset:")
print(" shape = {}, data type = {}, min val = {}, max val = {}".format(X_train.shape,
X_train.dtype,
np.min(
X_train),
np.max(X_train)))
print("Development dataset:")
print(" shape = {}, data type = {}, min val = {}, max val = {}".format(X_devel.shape,
X_devel.dtype,
np.min(
X_devel),
np.max(X_devel)))
print("Test dataset:")
print(" shape = {}, data type = {}, min val = {}, max val = {}".format(X_test.shape,
X_test.dtype,
np.min(
X_test),
np.max(X_test)))
return X_train, Y_train, X_devel, Y_devel, X_test, Y_test
def get_data(conf):
"""
Function for loading data from import_data.py and specifying correct dimensions.
This function returns training, development and test data.
"""
data_dir = os.path.join(conf["data_root_dir"], conf["dataset"])
print(data_dir)
if conf["dataset"] == "cifar10":
conf["channels_x"] = 3
conf["height_x"] = 32
conf["width_x"] = 32
conf["input_dimension"] = conf["channels_x"] * conf["height_x"] * conf[
"width_x"]
conf["output_dimension"] = 10
X_train, Y_train, X_devel, Y_devel, X_test, Y_test = import_data.load_cifar10(
conf, data_dir, conf["devel_size"])
elif conf["dataset"] == "mnist":
conf["channels_x"] = 1
conf["height_x"] = 28
conf["width_x"] = 28
conf["input_dimension"] = conf["channels_x"] * conf["height_x"] * conf[
"width_x"]
conf["output_dimension"] = 10
X_train, Y_train, X_devel, Y_devel, X_test, Y_test = import_data.load_mnist(
conf, data_dir, conf["devel_size"])
elif conf["dataset"] == "svhn":
conf["channels_x"] = 3
conf["height_x"] = 32
conf["width_x"] = 32
conf["input_dimension"] = conf["channels_x"] * conf["height_x"] * conf[
"width_x"]
conf["output_dimension"] = 10
X_train, Y_train, X_devel, Y_devel, X_test, Y_test = import_data.load_svhn(
conf, data_dir, conf["devel_size"])
conf["layer_dimensions"] = [
conf["input_dimension"]
] + conf["hidden_dimensions"] + [conf["output_dimension"]]
if conf["verbose"]:
print("Train dataset:")
print(
" shape = {}, data type = {}, min val = {}, max val = {}".format(
X_train.shape, X_train.dtype, np.min(X_train),
np.max(X_train)))
print("Development dataset:")
print(
" shape = {}, data type = {}, min val = {}, max val = {}".format(
X_devel.shape, X_devel.dtype, np.min(X_devel),
np.max(X_devel)))
print("Test dataset:")
print(
" shape = {}, data type = {}, min val = {}, max val = {}".format(
X_test.shape, X_test.dtype, np.min(X_test), np.max(X_test)))
return X_train, Y_train, X_devel, Y_devel, X_test, Y_test
print('Only numerics are legal')
print(e)
else:
dims = [784, 128, 10]
conf = {
'layer_dimensions': dims,
'learning_rate': float(options.learningrate),
'epochs': int(options.epoch),
'batch_size': int(options.batchsize),
'activation_function': 'relu'
}
model = net(layerdimensions=conf['layer_dimensions'])
X, y, X_test, y_test = load_mnist('data_mnist')
b = len(X) // conf['batch_size']
print('\nModel')
print('-' * 50)
print('Layer dimentions: {}'.format(conf['layer_dimensions']))
print('Learning rate: {}'.format(conf['learning_rate']))
print('Epochs: {}'.format(conf['epochs']))
print('Batch size: {}\n'.format(conf['batch_size']))
print('-' * 50)
print('One hot encoding target values..\n')
y = one_hot_encode(y)
y_test = one_hot_encode(y_test)
print('Scaling pixel data..\n')
import numpy as np
import os
from import_data import obtain, load_mnist
from VAEs import VAE_q2
# In[Load and importantion code]
# Import binarized MNIST if it was not already done
exists = os.path.isfile(os.path.join(os.getcwd(), 'binarized_mnist'))
if os.path.isdir(os.path.join(os.getcwd(), 'binarized_mnist')) == False:
obtain(os.path.join(os.getcwd(), 'binarized_mnist'))
batch_size = 32
valid_batch_size = 200
test_batch_size = 200
train_MNISt, val_MNIST, test_MNIST = load_mnist(batch_size, valid_batch_size,
test_batch_size)
# In[Train function]
def train(epochs, l_r):
model = VAE_q2()
if torch.cuda.is_available():
print("Using the GPU")
device = torch.device("cuda")
cuda_available = True
else:
print(
"WARNING: You are about to run on cpu, and this will likely run out \
of memory. \n You can try setting batch_size=1 to reduce memory usage")
device = torch.device("cpu")