def fetch_alpha_img():
#conv.conv.print_test()
# Fetch data
t1, l1, files = lp.batch_load_alpha(0, 1, True, [], LOAD.ALPHA)
#dset = lp.get_dataset(load_type=LOAD.ALPHA)
X_train = t1 #dset[0]
y_train = l1 #dset[1]
X_train = np.reshape(X_train, (-1, 1, 28, 28))
y_train = np.array(y_train)
# Downsample training data
n_train_samples = 1000 #3000
return X_train, y_train
def fetch_alpha_img():
#conv.conv.print_test()
# Fetch data
t1, l1, files = lp.batch_load_alpha( 0, 1, True, [], LOAD.ALPHA)
#dset = lp.get_dataset(load_type=LOAD.ALPHA)
X_train = t1 #dset[0]
y_train = l1 #dset[1]
X_train = np.reshape(X_train, (-1, 1, 28, 28))
y_train = np.array(y_train)
# Downsample training data
n_train_samples = 1000 #3000
return X_train, y_train
def run(max_iter=10, n_train_samples=300):
name = "alpha"
print str(datetime.datetime.now())
def signal_handler(signal, frame) :
print(" \n...you want to exit!")
for layer in nn.layers:
if (isinstance(layer, lnnet.ParamMixin)) : print "len: " + str(len(layer.W))
if True :
print("save weights.")
sys.stdout.flush()
nn.save_file(name=name)
sys.exit(0)
signal.signal(signal.SIGINT,signal_handler)
#n_train_samples = 3000 #3000
# Fetch data
#dset = lp.get_dataset(load_type=LOAD.ALPHA)
t1, l1, files = lp.batch_load_alpha( 0, n_train_samples, True, [], LOAD.ALPHA)
X_train = t1 #dset[0]
y_train = l1 #dset[1]
X_train = np.reshape(X_train, (-1, 1, 28, 28))
y_train = np.array(y_train)
'''
train_idxs = np.random.random_integers(0, len(dset[0])-1, n_train_samples)
#train_idxs = np.array([i for i in range(n_train_samples)])
X_train = X_train[train_idxs, ...]
y_train = y_train[train_idxs, ...]
'''
# Downsample training data
n_classes = len(lp.ascii_ymatrix(LOAD.ALPHA))
# Setup convolutional neural network
nn = cnnet.NeuralNetwork(
layers=[
conv.Conv(
n_feats=12,
filter_shape=(5, 5),
strides=(1, 1),
weight_scale=0.1,
weight_decay=0.001,
),
lnnet.Activation('relu'),
conv.Pool(
pool_shape=(2, 2),
strides=(2, 2),
mode='max',
),
conv.Conv(
n_feats=16,
filter_shape=(5, 5),
strides=(1, 1),
weight_scale=0.1,
weight_decay=0.001,
),
lnnet.Activation('relu'),
conv.Flatten(),
#lnnet.Linear(
# n_out=500,
# weight_scale=0.1,
# weight_decay=0.02,
#),
#lnnet.Activation('relu'),
lnnet.Linear(
n_out=n_classes,
weight_scale=0.1,
weight_decay=0.02,
),
lnnet.LogRegression(),
],
)
# Train neural network
t0 = time.time()
nn.set_android_load(False)
if n_train_samples != 300 and False : nn.set_interrupt(True)
if max_iter < 0 :
X = X_train
Y = y_train
Y_one_hot = one_hot(Y , load_type=LOAD.ALPHA)
nn.set_name(name)
nn._setup(X, Y_one_hot)
nn.load_file(name=name)
nn.status(-1,X,Y,Y_one_hot) ##end
else:
nn.fit(X_train, y_train, learning_rate=0.05, max_iter=max_iter, batch_size=64, name=name, load_type = LOAD.ALPHA)
t1 = time.time()
print('Duration: %.1fs' % (t1-t0))
if False:
# Evaluate on test data
error = nn.error(X_test, y_test)
print('Test error rate: %.4f' % error)
def run(max_iter=10, n_train_samples=300):
name = "alpha"
print str(datetime.datetime.now())
def signal_handler(signal, frame):
print(" \n...you want to exit!")
for layer in nn.layers:
if (isinstance(layer, lnnet.ParamMixin)):
print "len: " + str(len(layer.W))
if True:
print("save weights.")
sys.stdout.flush()
nn.save_file(name=name)
sys.exit(0)
signal.signal(signal.SIGINT, signal_handler)
#n_train_samples = 3000 #3000
# Fetch data
#dset = lp.get_dataset(load_type=LOAD.ALPHA)
t1, l1, files = lp.batch_load_alpha(0, n_train_samples, True, [],
LOAD.ALPHA)
X_train = t1 #dset[0]
y_train = l1 #dset[1]
X_train = np.reshape(X_train, (-1, 1, 28, 28))
y_train = np.array(y_train)
'''
train_idxs = np.random.random_integers(0, len(dset[0])-1, n_train_samples)
#train_idxs = np.array([i for i in range(n_train_samples)])
X_train = X_train[train_idxs, ...]
y_train = y_train[train_idxs, ...]
'''
# Downsample training data
n_classes = len(lp.ascii_ymatrix(LOAD.ALPHA))
# Setup convolutional neural network
nn = cnnet.NeuralNetwork(
layers=[
conv.Conv(
n_feats=12,
filter_shape=(5, 5),
strides=(1, 1),
weight_scale=0.1,
weight_decay=0.001,
),
lnnet.Activation('relu'),
conv.Pool(
pool_shape=(2, 2),
strides=(2, 2),
mode='max',
),
conv.Conv(
n_feats=16,
filter_shape=(5, 5),
strides=(1, 1),
weight_scale=0.1,
weight_decay=0.001,
),
lnnet.Activation('relu'),
conv.Flatten(),
#lnnet.Linear(
# n_out=500,
# weight_scale=0.1,
# weight_decay=0.02,
#),
#lnnet.Activation('relu'),
lnnet.Linear(
n_out=n_classes,
weight_scale=0.1,
weight_decay=0.02,
),
lnnet.LogRegression(),
], )
# Train neural network
t0 = time.time()
nn.set_android_load(False)
if n_train_samples != 300 and False: nn.set_interrupt(True)
if max_iter < 0:
X = X_train
Y = y_train
Y_one_hot = one_hot(Y, load_type=LOAD.ALPHA)
nn.set_name(name)
nn._setup(X, Y_one_hot)
nn.load_file(name=name)
nn.status(-1, X, Y, Y_one_hot) ##end
else:
nn.fit(X_train,
y_train,
learning_rate=0.05,
max_iter=max_iter,
batch_size=64,
name=name,
load_type=LOAD.ALPHA)
t1 = time.time()
print('Duration: %.1fs' % (t1 - t0))
if False:
# Evaluate on test data
error = nn.error(X_test, y_test)
print('Test error rate: %.4f' % error)