def test_affine_wrapper(backend_default):
"""
Verify that the Affine wrapper constructs the right layer objects.
"""
nout = 11
aff = Affine(nout, Uniform())
assert isinstance(aff, list)
assert len(aff) == 1
assert isinstance(aff[0], Linear)
assert aff[0].nout == nout
aff = Affine(nout, Uniform(), bias=Uniform())
assert isinstance(aff, list)
assert len(aff) == 2
assert isinstance(aff[0], Linear)
assert isinstance(aff[1], Bias)
aff = Affine(nout, Uniform(), activation=Rectlin())
assert isinstance(aff, list)
assert len(aff) == 2
assert isinstance(aff[0], Linear)
assert isinstance(aff[1], Activation)
aff = Affine(nout, Uniform(), bias=Uniform(), activation=Rectlin())
assert isinstance(aff, list)
assert len(aff) == 3
assert isinstance(aff[0], Linear)
assert isinstance(aff[1], Bias)
assert isinstance(aff[2], Activation)
def test_conv_wrapper(backend_default):
"""
Verify that the Conv wrapper constructs the right layer objects.
"""
conv = Conv((4, 4, 3), Uniform())
assert isinstance(conv, list)
assert len(conv) == 1
assert isinstance(conv[0], Convolution)
conv = Conv((4, 4, 3), Uniform(), bias=Uniform())
assert isinstance(conv, list)
assert len(conv) == 1
assert isinstance(conv[0], Convolution_bias)
conv = Conv((4, 4, 3), Uniform(), activation=Rectlin())
assert isinstance(conv, list)
assert len(conv) == 2
assert isinstance(conv[0], Convolution)
assert isinstance(conv[1], Activation)
conv = Conv((4, 4, 3), Uniform(), bias=Uniform(), activation=Rectlin())
assert isinstance(conv, list)
assert isinstance(conv[0], Convolution_bias)
assert isinstance(conv[1], Activation)
assert len(conv) == 2
def create_layers(self, max_action_no):
layers = []
initializer = self.get_initializer(input_size=4 * 8 * 8)
layers.append(
Conv(fshape=(8, 8, 32),
strides=4,
init=initializer,
bias=initializer,
activation=Rectlin()))
initializer = self.get_initializer(input_size=32 * 4 * 4)
layers.append(
Conv(fshape=(4, 4, 64),
strides=2,
init=initializer,
bias=initializer,
activation=Rectlin()))
initializer = self.get_initializer(input_size=64 * 3 * 3)
layers.append(
Conv(fshape=(3, 3, 64),
strides=1,
init=initializer,
bias=initializer,
activation=Rectlin()))
initializer = self.get_initializer(input_size=7 * 7 * 64)
layers.append(
Affine(nout=512,
init=initializer,
bias=initializer,
activation=Rectlin()))
initializer = self.get_initializer(input_size=512)
layers.append(
Affine(nout=max_action_no, init=initializer, bias=initializer))
return layers
def test_conv_wrapper(backend_default):
"""
Verify that the Conv wrapper constructs the right layer objects.
"""
conv = Conv((4, 4, 3), Uniform())
assert isinstance(conv, list)
assert len(conv) == 1
assert isinstance(conv[0], Convolution)
conv = Conv((4, 4, 3), Uniform(), bias=Uniform())
assert isinstance(conv, list)
# temp roll back conv_bias
if False and conv[0].be.is_mkl():
assert len(conv) == 1
assert isinstance(conv[0], Convolution_bias)
else:
assert len(conv) == 2
assert isinstance(conv[0], Convolution)
assert isinstance(conv[1], Bias)
conv = Conv((4, 4, 3), Uniform(), activation=Rectlin())
assert isinstance(conv, list)
assert len(conv) == 2
assert isinstance(conv[0], Convolution)
assert isinstance(conv[1], Activation)
conv = Conv((4, 4, 3), Uniform(), bias=Uniform(), activation=Rectlin())
assert isinstance(conv, list)
# temp roll back conv_bias
if False and conv[0].be.is_mkl():
assert isinstance(conv[0], Convolution_bias)
assert isinstance(conv[1], Activation)
assert len(conv) == 2
else:
assert isinstance(conv[0], Convolution)
assert isinstance(conv[1], Bias)
assert isinstance(conv[2], Activation)
assert len(conv) == 3
from neon.layers import Conv, Affine, Pooling
from neon.initializers import Uniform
from neon.transforms.activation import Rectlin, Softmax
from neon.models import Model
from neon.initializers import Kaiming
from neon.optimizers import Adadelta
from neon.layers import GeneralizedCost
from neon.transforms import CrossEntropyMulti
from neon.optimizers import GradientDescentMomentum, RMSProp
from neon.callbacks.callbacks import Callbacks
from neon.data import ArrayIterator
from Readfile import DataSet, readfile
be = gen_backend(backend='cpu', batch_size=30)
init_uni = Uniform(low=-0.1, high=0.1)
layers = [
Conv(fshape=(4, 4, 16), init=init_uni, activation=Rectlin()),
Pooling(fshape=2, strides=2),
Conv(fshape=(4, 4, 32), init=init_uni, activation=Rectlin()),
Pooling(fshape=2, strides=2),
Conv(fshape=(4, 4, 32), init=init_uni, activation=Rectlin()),
Pooling(fshape=2, strides=2),
Affine(nout=500, init=init_uni, activation=Rectlin()),
Affine(nout=11, init=init_uni, activation=Softmax())
]
model = Model(layers)
model.load_params('model.pkl')
data = readfile('PreImage', 'label.csv')
X_test = data.test_data
test_set = ArrayIterator(X_test, None, nclass=11, lshape=(1, 200, 200))
test_set = ArrayIterator(Xtest, y_test, nclass=2, lshape=(1, 12, 12))
from neon.layers import Conv, Affine, Pooling, Dropout
from neon.initializers import Uniform, Constant, Gaussian
from neon.transforms.activation import Rectlin, Softmax
init_uni = Uniform(low=-0.1, high=0.1) # try Gaussian(loc=0, scale=0.1)
init_cst = Constant(0.1) #to avoid dead neurons
layers = []
layers.append(
Conv(fshape=(2, 2, 32),
init=init_uni,
bias=init_cst,
padding=0,
activation=Rectlin()))
layers.append(Pooling(fshape=2, strides=2))
layers.append(Affine(nout=2, init=init_uni, activation=Softmax()))
from neon.models import Model
model = Model(layers)
from neon.layers import GeneralizedCost
from neon.transforms import CrossEntropyBinary
cost = GeneralizedCost(costfunc=CrossEntropyBinary())
from neon.optimizers import GradientDescentMomentum
optimizer = GradientDescentMomentum(0.1, momentum_coef=0.9)
from neon.callbacks.callbacks import Callbacks
callbacks = Callbacks(model, train_set)