def get_data():
"""
>>> x_test, t_test = get_data()
>>> np.shape(x_test)
(10000, 784)
>>> np.shape(t_test)
(10000,)
"""
(x_train, t_train), (x_test, t_test) = \
load_mnist(normalize=True, flatten=True, one_hot_label=False)
return x_test, t_test
#!/usr/bin/env python3
import numpy as np
from lib.mnist import load_mnist
from chap6.multi_layer_net import MultiLayerNet
from chap6.overfit_weight_decay import train
from optimizers import SGD
(x_train, t_train), (x_test, t_test) = load_mnist(normalize=True,
one_hot_label=True)
x_train = x_train[:300]
t_train = t_train[:300]
optimizer = SGD(lr=0.01)
max_epochs = 201
train_size = x_train.shape[0]
batch_size = 100
# train
network = MultiLayerNet(input_size=784,
hidden_size_list=[100, 100, 100, 100, 100, 100],
output_size=10)
(train_loss_list, train_acc_list, test_acc_list) = train(network)
network = MultiLayerNet(input_size=784,
hidden_size_list=[100, 100, 100, 100, 100, 100],
output_size=10,
use_dropout=True,
dropout_ratio=0.2)
(train_loss_list_decay, train_acc_list_decay,
test_acc_list_decay) = train(network)
# draw out
return acc / x.shape[0]
def _test():
import doctest
doctest.testmod()
if __name__ == "__main__":
_test()
print("start mnist training")
from chap6.optimizers import SGD
from lib.mnist import load_mnist
(x_train, t_train), (x_test, t_test) = load_mnist(one_hot_label=True,
flatten=False)
optimizer = SGD()
max_epochs = 201
train_size = x_train.shape[0]
batch_size = 100
def train(network):
train_loss_list = []
train_acc_list = []
test_acc_list = []
iter_per_epoch = max(train_size / batch_size, 1)
print(iter_per_epoch)
epoch_cnt = 0
import numpy as np from lib.mnist import load_mnist from lib.network import TwoLayerNet (x_train, t_train), (x_test, t_test) = load_mnist(path='dataset', one_hot_label='True', flatten='True') network = TwoLayerNet(input_size=784, hidden_size=50, output_size=10) network.restore_network() print(np.argmax(network.predict(x_test[0]))) print(t_test[0])
failures = []
for idx in range(x_test.shape[0]):
if y[idx] != t_test[idx]:
failures.append((x_test[idx], y[idx], t_test[idx]))
for i in range(min(len(failures), 60)):
img, y, _ = failures[i]
if (i % 10 == 0): print()
print(y, end=", ")
img = img.reshape(28, 28)
plt.subplot(6, 10, i + 1)
plt.imshow(img, cmap='gray')
print()
plt.show()
(x_train, t_train), (x_test, t_test) = load_mnist()
options = {
'weight_decay_lambda': 0,
'learning_rate': 0.01,
'beta1': 0.9,
'beta2': 0.999
}
overfitmask = np.random.choice(x_train.shape[0], 300)
x_train = x_train[overfitmask]
t_train = t_train[overfitmask]
network = Network2()
network.add(Dense(784, 50, options, 'sgd', 'he'))
network.add(Relu())
network.add(Dense(50, 10, options, 'sgd', 'xavier'))
network.add(SoftmaxWithLoss())
