def loadDataset(self):
print 'load MNIST dataset'
dataset = data.load_mnist_data()
dataset['data'] = dataset['data'].astype(np.float32)
dataset['data'] /= 255 # 最終的にnpの行列データを取得
dataset['target'] = dataset['target'].astype(np.int32)
return dataset
def test8_evaluate(self):
(train_data, valid_data) = load_mnist_data()
net = nnet.Network([784, 5, 10], debug=True)
nt = net.evaluate(train_data[:100])
self.assertEqual(nt, 14)
nv = net.evaluate(valid_data[9000:])
self.assertEqual(nv, 102)
def test_load_data(self):
mnist = data.load_mnist_data()
self.assertEqual(len(mnist['data']), 70000)
self.assertEqual(len(mnist['data'][0]), 784)
self.assertEqual(mnist['data'][0][0], 0)
self.assertEqual(mnist['data'][0][156], 126)
self.assertEqual(len(mnist['target']), 70000)
self.assertEqual(mnist['target'][0], 5)
def test9_train(self):
(train_data, valid_data) = load_mnist_data()
# reduce data sets for faster speed:
train_data = train_data[:1000]
valid_data = valid_data[:1000]
net = nnet.Network([784, 12, 10], debug=True)
net.train(train_data, valid_data, epochs=1, mini_batch_size=8, alpha=5)
nv = net.evaluate(valid_data)
self.assertEqual(nv, 503)
def test7_update_minibatch(self):
(train_data, _) = load_mnist_data()
train_data_vec = [(x, nnet.unit(y, 10)) for x, y in train_data]
mini_batch = train_data_vec[10:20]
net = nnet.Network([784, 15, 10], debug=True)
net.update_mini_batch(mini_batch, 5.5)
self.assertEqual(net.biases[1].shape[0], 10)
self.assertEqual(net.biases[1].shape[1], 1)
self.assertAlmostEqual(net.biases[1][9, 0], -0.537912934538857)
self.assertEqual(net.weights[0].shape[0], 15)
self.assertEqual(net.weights[0].shape[1], 784)
self.assertAlmostEqual(net.weights[0][9, 333], 0.099262147771176)
def read_dataset(self):
print('load MNIST dataset')
mnist = data.load_mnist_data()
mnist['data'] = mnist['data'].astype(np.float32)
mnist['data'] /= 255
mnist['target'] = mnist['target'].astype(np.int32)
# Number of data
self.N = 60000
self.x_train, self.x_test = np.split(mnist['data'], [self.N])
self.y_train, self.y_test = np.split(mnist['target'], [self.N])
# Number of test data
self.N_test = self.y_test.size
def main():
# Load data
device = torch.device("cuda:0" if (
torch.cuda.is_available() and ngpu > 0) else "cpu")
dataloader = load_mnist_data(classes=[0, 1],
batch_size=batch_size,
image_size=image_size)
# Plot some training images
#plot_data_batch(device, dataloader)
# Create the generator
netG = Generator_Quantum(n_qubits, q_depth, q_delta).to(device)
# Create the Discriminator
netD = Discriminator_FCNN(image_size * image_size, ngpu).to(device)
train(device, netG, netD, dataloader)
from chainer import optimizers
import numpy as np
from pylab import *
import time
import data
# 学習のパラメータ
batchsize = 100
n_epoch = 20
# データセットの準備
print ("load MNIST dataset")
mnist = data.load_mnist_data()
mnist["data"] = mnist["data"].astype(np.float32)
mnist["data"] /= 255
mnist["target"] = mnist["target"].astype(np.int32)
# 訓練データとテストデータに分割
N = 60000
x_train, x_test = np.split(mnist["data"], [N])
y_train, y_test = np.split(mnist["target"], [N])
N_test = y_test.size
# 画像を (nsample, channel, height, width) の4次元テンソルに変換
# MNISTはチャンネル数が1なのでreshapeだけでOK
X_train = x_train.reshape((len(x_train), 1, 28, 28))
X_test = x_test.reshape((len(x_test), 1, 28, 28))
if __name__ == "__main__":
parser = argparse.ArgumentParser(description="Chainer-BriCA integration")
parser.add_argument("--gpu", "-g", default=-1, type=int, help="GPU ID")
args = parser.parse_args()
use_gpu = False
if args.gpu >= 0:
print "Using gpu: {}".format(args.gpu)
use_gpu = True
cuda.get_device(args.gpu).use()
batchsize = 100
mnist = data.load_mnist_data()
mnist['data'] = mnist['data'].astype(np.float32)
mnist['data'] /= 255
mnist['target'] = mnist['target'].astype(np.int32)
N_train = 60000
x_train, x_test = np.split(mnist['data'], [N_train])
y_train, y_test = np.split(mnist['target'], [N_train])
N_test = y_test.size
f = open('sda.pkl', 'rb')
stacked_autoencoder = pickle.load(f)
f.close()
scheduler = brica1.VirtualTimeSyncScheduler()
agent = brica1.Agent(scheduler)
from data import load_mnist_data
import nnet
import numpy as np
from matplotlib import pyplot as plt
def show(x):
""" visualize a single training example """
im = plt.imshow(np.reshape(1 - x, (28, 28)))
im.set_cmap('gray')
print("loading MNIST dataset")
(train_data, valid_data) = load_mnist_data()
# reduce data sets for faster speed:
train_data = train_data
valid_data = valid_data
# to see a training example, uncomment:
#x, y = train_data[123]
#show(x)
#plt.title("label = %d" % y)
# some initial params, not necessarily good ones
net = nnet.Network([784, 80, 10])
print("training")
net.train(train_data, valid_data, epochs=10, mini_batch_size=8, alpha=0.5)
def __init__(self):
self._data = data.load_mnist_data()
self._data['data'] = self._data['data'].astype(np.float32)
self._data['data'] /= 255
self._data['target'] = self._data['target'].astype(np.int32)
self._last = 0
help='Use tiny datasets for quick tests')
args = parser.parse_args()
batchsize = args.batchsize
n_epoch = args.epoch
n_latent = args.dimz
print('GPU: {}'.format(args.gpu))
print('# dim z: {}'.format(args.dimz))
print('# Minibatch-size: {}'.format(args.batchsize))
print('# epoch: {}'.format(args.epoch))
print('')
# Prepare dataset
print('load MNIST dataset')
mnist = data.load_mnist_data(args.test)
mnist['data'] = mnist['data'].astype(np.float32)
mnist['data'] /= 255
mnist['target'] = mnist['target'].astype(np.int32)
if args.test:
N = 30
else:
N = 60000
x_train, x_test = np.split(mnist['data'], [N])
y_train, y_test = np.split(mnist['target'], [N])
N_test = y_test.size
# Prepare VAE model, defined in net.py
model = net.VAE(784, n_latent, 500)
def main():
# Reproducibility
seed = settings["seed"]
random.seed(seed)
np.random.seed(seed)
torch.manual_seed(seed)
torch.cuda.manual_seed_all(seed)
torch.backends.cudnn.deterministic = True
torch.backends.cudnn.benchmark = False
print("Loading datasets...")
circle_train_input, circle_train_target, \
circle_test_input, circle_test_target = generate_circle_classification_dataset()
mnist_train_input, mnist_train_target, mnist_test_input, mnist_test_target = load_mnist_data()
fashion_mnist_train_input, fashion_mnist_train_target, \
fashion_mnist_test_input, fashion_mnist_test_target = load_fashion_mnist_data()
print("Done!")
results_dir = settings["results_dir"]
# Boolean variable, whether to run the mini-batch size and learning rate experiment or was it already completed
# WARNING: Takes a very long time
run_mini_batch_size_lr_experiments = False
if run_mini_batch_size_lr_experiments:
print("Running the mini-batch size and learning rate experiments...")
circle_experiment_log, circle_training_logs = \
run_mini_batch_size_experiment("circle", circle_train_input, circle_train_target,
circle_test_input, circle_test_target)
circle_experiment_log.to_csv(results_dir + "circle_mini_batch_size_lr_experiment_log.csv",
sep=",", header=True, index=False, encoding="utf-8")
circle_training_logs.to_csv(results_dir + "circle_mini_batch_size_lr_training_logs.csv",
sep=",", header=True, index=False, encoding="utf-8")
mnist_experiment_log, mnist_training_logs = \
run_mini_batch_size_experiment("mnist", mnist_train_input, mnist_train_target,
mnist_test_input, mnist_test_target)
mnist_experiment_log.to_csv(results_dir + "mnist_mini_batch_size_lr_experiment_log.csv",
sep=",", header=True, index=False, encoding="utf-8")
mnist_training_logs.to_csv(results_dir + "mnist_mini_batch_size_lr_training_logs.csv",
sep=",", header=True, index=False, encoding="utf-8")
fashion_mnist_experiment_log, fashion_mnist_training_logs = \
run_mini_batch_size_experiment("fashion_mnist", fashion_mnist_train_input, fashion_mnist_train_target,
fashion_mnist_test_input, fashion_mnist_test_target)
fashion_mnist_experiment_log.to_csv(results_dir + "fashion_mnist_mini_batch_size_lr_experiment_log.csv",
sep=",", header=True, index=False, encoding="utf-8")
fashion_mnist_training_logs.to_csv(results_dir + "fashion_mnist_mini_batch_size_lr_training_logs.csv",
sep=",", header=True, index=False, encoding="utf-8")
print("Done!")
else:
print("Loading the first experiment logs...")
circle_experiment_log = pd.read_csv(results_dir + "circle_mini_batch_size_lr_experiment_log.csv",
sep=",", header=0, index_col=None, encoding="utf-8")
circle_training_logs = pd.read_csv(results_dir + "circle_mini_batch_size_lr_training_logs.csv",
sep=",", header=0, index_col=None, encoding="utf-8")
mnist_experiment_log = pd.read_csv(results_dir + "mnist_mini_batch_size_lr_experiment_log.csv",
sep=",", header=0, index_col=None, encoding="utf-8")
mnist_training_logs = pd.read_csv(results_dir + "mnist_mini_batch_size_lr_training_logs.csv",
sep=",", header=0, index_col=None, encoding="utf-8")
fashion_mnist_experiment_log = pd.read_csv(results_dir + "fashion_mnist_mini_batch_size_lr_experiment_log.csv",
sep=",", header=0, index_col=None, encoding="utf-8")
fashion_mnist_training_logs = pd.read_csv(results_dir + "fashion_mnist_mini_batch_size_lr_training_logs.csv",
sep=",", header=0, index_col=None, encoding="utf-8")
print("Done!")
print("Computing combined score for experiment conditions...")
circle_experiment_log = compute_combined_score_for_experiment_conditions(circle_experiment_log)
mnist_experiment_log = compute_combined_score_for_experiment_conditions(mnist_experiment_log)
fashion_mnist_experiment_log = compute_combined_score_for_experiment_conditions(fashion_mnist_experiment_log)
print("Done!")
print("Finding optimal mini-batch sizes and learning rates...")
circle_best_mini_batch_sizes, circle_best_lrs = \
extract_optimal_parameters_from_experiment_log(circle_experiment_log)
mnist_best_mini_batch_sizes, mnist_best_lrs = \
extract_optimal_parameters_from_experiment_log(mnist_experiment_log)
fashion_mnist_best_mini_batch_sizes, fashion_mnist_best_lrs = \
extract_optimal_parameters_from_experiment_log(fashion_mnist_experiment_log)
print("Done!")
def merge_experiment_data_from_different_datasets(circle_data, mnist_data, fashion_mnist_data):
"""
Helper function to merge experiment results from the different datasets into one combined log
:param circle_data: results from an experiment on the 2D circle dataset, pandas.Dataframe
:param mnist_data: results from an experiment on the MNIST dataset, pandas.Dataframe
:param fashion_mnist_data: results from an experiment on the FashionMNIST dataset, pandas.Dataframe
:returns: merged experiment log, pandas.Dataframe
"""
return pd.concat((circle_data, mnist_data, fashion_mnist_data), axis=0, ignore_index=True)
print("Generating the plots for the first experiment...")
experiment_log = merge_experiment_data_from_different_datasets(circle_experiment_log,
mnist_experiment_log,
fashion_mnist_experiment_log)
training_logs = merge_experiment_data_from_different_datasets(circle_training_logs,
mnist_training_logs,
fashion_mnist_training_logs)
visualize_results_first_experiment(experiment_log, training_logs)
print("Done!")
# Boolean variable, whether to run the convergence region experiment or was it already completed
# WARNING: Takes a long time
run_convergence_region_experiments = False
if run_convergence_region_experiments:
print("Running the convergence region experiments...")
circle_experiment_log = run_convergence_region_experiment("circle", circle_train_input, circle_train_target,
circle_best_mini_batch_sizes, circle_best_lrs)
circle_experiment_log.to_csv(results_dir + "circle_convergence_region_experiment_log.csv",
sep=",", header=True, index=False, encoding="utf-8")
mnist_experiment_log = run_convergence_region_experiment("mnist", mnist_train_input, mnist_train_target,
mnist_best_mini_batch_sizes, mnist_best_lrs)
mnist_experiment_log.to_csv(results_dir + "mnist_convergence_region_experiment_log.csv",
sep=",", header=True, index=False, encoding="utf-8")
fashion_mnist_experiment_log = \
run_convergence_region_experiment("fashion_mnist", fashion_mnist_train_input, fashion_mnist_train_target,
fashion_mnist_best_mini_batch_sizes, fashion_mnist_best_lrs)
fashion_mnist_experiment_log.to_csv(results_dir + "fashion_mnist_convergence_region_experiment_log.csv",
sep=",", header=True, index=False, encoding="utf-8")
print("Done!")
else:
print("Loading the second experiment logs...")
circle_experiment_log = pd.read_csv(results_dir + "circle_convergence_region_experiment_log.csv",
sep=",", header=0, index_col=None, encoding="utf-8")
mnist_experiment_log = pd.read_csv(results_dir + "mnist_convergence_region_experiment_log.csv",
sep=",", header=0, index_col=None, encoding="utf-8")
fashion_mnist_experiment_log = pd.read_csv(results_dir + "fashion_mnist_convergence_region_experiment_log.csv",
sep=",", header=0, index_col=None, encoding="utf-8")
print("Done!")
print("Generating the plots for the second experiment...")
experiment_log = merge_experiment_data_from_different_datasets(circle_experiment_log,
mnist_experiment_log,
fashion_mnist_experiment_log)
visualize_results_second_experiment(experiment_log)
print("Done!")
def train(trainL, trainA, testL, testA):
trainLoss = []
trainAccuracy = []
testLoss = []
testAccuracy = []
l1w = []
l2w = []
l3w = []
print('load MNIST dataset')
mnist = data.load_mnist_data()
mnist['data'] = mnist['data'].astype(np.float32)
mnist['data'] /= 255
mnist['target'] = mnist['target'].astype(np.int32)
N = 1000
lsizes = [784, 50, 50, 10]
x_train, x_test = np.split(mnist['data'], [N])
y_train, y_test = np.split(mnist['target'], [N])
N_test = y_test.size
# Prepare multi-layer perceptron model, defined in net.py
if args.net == 'simple':
#model = net.MnistMLP(lsizes)
model = net.MnistMLP(layer_sizes=lsizes)
if args.gpu >= 0:
cuda.get_device(args.gpu).use()
model.to_gpu()
xp = np if args.gpu < 0 else cuda.cupy
elif args.net == 'parallel':
cuda.check_cuda_available()
model = L.Classifier(net.MnistMLPParallel(784, n_units, 10))
xp = cuda.cupy
# Setup optimizer
optimizer = optimizers.Adam()
optimizer.setup(model)
# Init/Resume
if args.initmodel:
print('Load model from', args.initmodel)
serializers.load_npz(args.initmodel, model)
if args.resume:
print('Load optimizer state from', args.resume)
serializers.load_npz(args.resume, optimizer)
# Pretrain loop
print("start pretrain")
epo = p_epoch
for j in six.moves.range(1, len(lsizes)):
if j == len(lsizes) - 1:
model.setfinetuning()
print("start finetuning")
epo = n_epoch
for epoch in six.moves.range(1, epo + 1):
print('layer ', j, 'p_epoch ', epoch)
perm = np.random.permutation(N)
sum_accuracy = 0
sum_loss = 0
for i in six.moves.range(0, N, batchsize):
x = chainer.Variable(xp.asarray(x_train[perm[i:i +
batchsize]]))
t = chainer.Variable(xp.asarray(y_train[perm[i:i +
batchsize]]))
optimizer.update(model, x, t, j)
sum_loss += float(model.loss.data) * len(t.data)
if not model.pretrain:
sum_accuracy += float(model.accuracy.data) * len(t.data)
if model.pretrain:
print('Pretrain: train mean loss={}'.format(sum_loss / N))
else:
print('Finetune: train mean loss={}, accuracy={}'.format(
sum_loss / N, sum_accuracy / N))
trainLoss.append(sum_loss / N)
trainAccuracy.append(sum_accuracy / N)
# evaluation
sum_accuracy = 0
sum_loss = 0
model.train = False
for i in six.moves.range(0, N_test, batchsize):
x = chainer.Variable(xp.asarray(x_test[i:i + batchsize]),
volatile='on')
t = chainer.Variable(xp.asarray(y_test[i:i + batchsize]),
volatile='on')
loss = model(x, t, j)
sum_loss += float(loss.data) * len(t.data)
if not model.pretrain:
sum_accuracy += float(model.accuracy.data) * len(t.data)
if model.pretrain:
print('Pretrain: test mean loss={}'.format(sum_loss / N_test))
else:
print('Finetune: test mean loss={}, accuracy={}'.format(
sum_loss / N_test, sum_accuracy / N_test))
testLoss.append(sum_loss / N_test)
testAccuracy.append(sum_accuracy / N_test)
model.train = True
# Save the model and the optimizer
savecsv(trainLoss, trainL)
savecsv(trainAccuracy, trainA)
savecsv(testLoss, testL)
savecsv(testAccuracy, testA)
print('save the model')
serializers.save_npz('mlp.model', model)
print('save the optimizer')
serializers.save_npz('mlp.state', optimizer)
def __init__(self):
self._data = data.load_mnist_data()
self._data['data'] = self._data['data'].astype(np.float32)
self._data['data'] /= 255
self._data['target'] = self._data['target'].astype(np.int32)
self._last = 0
def load_mnist_binary_data(self):
mnist = data.load_mnist_data()
mnist['data'] = mnist['data'].astype(np.float32)
mnist['data'] /= 255
mnist['target'] = mnist['target'].astype(np.int32)
return mnist