def main():
# Get data
(x_train, y_train), (x_test, y_test) = datasets.mnist.load_data()
# Pre-processing
# Normalize
x_train = x_train / 255.0
x_test = x_test / 255.0
# Classes to categorical
num_classes = 10
y_train = utils.to_categorical(y_train, num_classes=num_classes)
y_test = utils.to_categorical(y_test, num_classes=num_classes)
# Shuffle dataset
x_train, y_train = utils.shuffle_dataset(x_train, y_train)
x_test, y_test = utils.shuffle_dataset(x_test, y_test)
# Params *********************************
batch_size = int(len(x_train) / 10)
epochs = 10
# Define architecture
l_in = Input(shape=x_train[0].shape)
l = Reshape(l_in, shape=(28 * 28, ))
l = PRelu(Dense(l, 1024))
l = LeakyRelu(Dense(l, 1024), alpha=0.1)
l_out = Softmax(Dense(l, num_classes))
# Build network
mymodel = Net()
mymodel.build(
l_in=[l_in],
l_out=[l_out],
optimizer=Adam(lr=0.001),
losses=[losses.CrossEntropy()],
metrics=[[metrics.CategoricalAccuracy()]],
debug=False,
smart_derivatives=False,
)
# Print model
mymodel.summary()
# Train
mymodel.fit([x_train], [y_train],
x_test=[x_test],
y_test=[y_test],
batch_size=batch_size,
epochs=epochs,
evaluate_epoch=False,
print_rate=1)
def test_iris_model1(self):
from sklearn import datasets
# Get data
iris = datasets.load_iris()
X = iris.data # we only take the first two features.
Y = iris.target
# Pre-processing
# Standarize
X = (X - np.mean(X, axis=0)) / np.std(X, axis=0)
# Classes to categorical
num_classes = 3
Y = utils.to_categorical(Y, num_classes=num_classes)
# Shuffle dataset
idxs = np.arange(len(X))
np.random.shuffle(idxs)
X, Y = X[idxs], Y[idxs]
# Select train/test
c = 1 # 0.8
tr_size = int(len(X) * c)
x_train, y_train = X[:tr_size], Y[:tr_size]
# Define architecture
l_in = Input(shape=x_train[0].shape)
l = Dense(l_in, 20)
l = Relu(l)
l = Dense(l, 15)
l = Relu(l)
l = Dense(l, num_classes)
l_out = Softmax(l)
# Build network
model = Net()
model.build(l_in=[l_in],
l_out=[l_out],
optimizer=Adam(lr=10e-2),
losses=[losses.CrossEntropy()],
metrics=[[metrics.CategoricalAccuracy()]],
debug=False)
# Check gradient
passed = gradient_check(model, [x_train], [y_train],
batch_size=int(len(x_train) / 10),
max_samples=25)
self.assertTrue(passed)
def test_mnist_model_mlp(self):
from keras import datasets
# Get data
(x_train, y_train), (_, _) = datasets.mnist.load_data()
# Pre-processing
# Normalize
x_train = x_train / 255.0
# Classes to categorical
num_classes = 10
y_train = utils.to_categorical(y_train, num_classes=num_classes)
# Shuffle dataset
x_train, y_train = utils.shuffle_dataset(x_train, y_train)
l_in = Input(shape=x_train[0].shape)
l = Reshape(l_in, shape=(28 * 28, ))
l1 = Relu(Dense(l, 512))
# l2 = Relu(Dense(l1, 512))
# l = Add([l1, l2])
# l = BatchNorm(l)
l = Relu(Dense(l1, 512))
l_out = Softmax(Dense(l, num_classes))
# Build network
model = Net()
model.build(
l_in=[l_in],
l_out=[l_out],
optimizer=Adam(lr=10e-3),
losses=[losses.CrossEntropy()],
metrics=[[metrics.CategoricalAccuracy()]],
debug=False,
smart_derivatives=False,
)
# Check gradient
passed = gradient_check(model, [x_train], [y_train],
batch_size=int(len(x_train) / 10),
max_samples=5,
verbose=2)
self.assertTrue(passed)
def main():
# Get data
iris = datasets.load_iris()
X = iris.data # we only take the first two features.
Y = iris.target
# Pre-processing
# Standarize
X = (X - np.mean(X, axis=0)) / np.std(X, axis=0)
# Classes to categorical
num_classes = 3
Y = utils.to_categorical(Y, num_classes=num_classes)
# Shuffle dataset
idxs = np.arange(len(X))
np.random.shuffle(idxs)
X, Y = X[idxs], Y[idxs]
# Select train/test
c = 0.8
tr_size = int(len(X) * c)
x_train, y_train = X[:tr_size], Y[:tr_size]
x_test, y_test = X[tr_size:], Y[tr_size:]
# Params *********************************
batch_size = int(len(x_train) / 5)
epochs = 500
# Define architecture
l_in = Input(shape=x_train[0].shape)
l = Dense(l_in,
20,
kernel_regularizer=L2(lmda=0.01),
bias_regularizer=L1(lmda=0.01))
l = Relu(l)
l = Dense(l, 15)
l = BatchNorm(l)
l = Dropout(l, 0.1)
l = Relu(l)
l = Dense(l, num_classes)
l_out = Softmax(l)
# Build network
mymodel = Net()
mymodel.build(l_in=[l_in],
l_out=[l_out],
optimizer=Adam(lr=10e-2),
losses=[losses.CrossEntropy()],
metrics=[[metrics.CategoricalAccuracy()]],
debug=False)
# Print model
mymodel.summary()
# Train
mymodel.fit([x_train], [y_train],
x_test=[x_test],
y_test=[y_test],
batch_size=batch_size,
epochs=epochs,
evaluate_epoch=True,
print_rate=10)
# Save mode
mymodel.save("./trained/trained_iris.pkl", save_grads=True)
# Evaluate
print("\n----------------------")
print("Evaluation:")
lo, me = mymodel.evaluate([x_test], [y_test], batch_size=batch_size)
str_eval = mymodel._format_eval(lo, me)
print(f"- Losses[{', '.join(str_eval[0])}]")
print(f"- Metrics[{'; '.join(str_eval[1])}]")
def test_mnist_model_conv(self):
from keras import datasets
# Get data
(x_train, y_train), (_, _) = datasets.mnist.load_data()
# Pre-processing
# Add channel dimension
x_train = np.expand_dims(x_train, axis=1)
# Normalize
x_train = x_train / 255.0
# Classes to categorical
num_classes = 10
y_train = utils.to_categorical(y_train, num_classes=num_classes)
# Shuffle dataset
x_train, y_train = utils.shuffle_dataset(x_train, y_train)
# Define architecture
l_in = Input(shape=x_train[0].shape)
l = l_in
l = Conv2D(l,
filters=2,
kernel_size=(3, 3),
strides=(1, 1),
padding="none")
l = MaxPool2D(l, pool_size=(3, 3), strides=(2, 2), padding="none")
l = Relu(l)
# l = GaussianNoise(l, stddev=0.1)
# l = Conv2D(l, filters=4, kernel_size=(3, 3), strides=(1, 1), padding="same")
# l = MaxPool2D(l, pool_size=(3, 3), strides=(2, 2), padding="none")
# l = Relu(l)
# l = DepthwiseConv2D(l, kernel_size=(3, 3), strides=(1, 1), padding="none")
# l = PointwiseConv2D(l, filters=1)
# l = MaxPool2D(l, pool_size=(3, 3), strides=(2, 2), padding="none")
# l = Relu(l)
l = Reshape(l, shape=(-1))
l = Dense(l,
num_classes,
kernel_initializer=initializers.RandomUniform())
l_out = Softmax(l)
# Build network
model = Net()
model.build(
l_in=[l_in],
l_out=[l_out],
optimizer=Adam(lr=10e-2),
losses=[losses.CrossEntropy()],
metrics=[[metrics.CategoricalAccuracy()]],
debug=False,
smart_derivatives=False,
)
# Check gradient
passed = gradient_check(model, [x_train], [y_train],
batch_size=int(len(x_train) / 10),
max_samples=5,
verbose=2)
self.assertTrue(passed)
def main():
# Get data
(x_train, y_train), (x_test, y_test) = datasets.mnist.load_data()
# Pre-processing
# Normalize
x_train = x_train / 255.0
x_test = x_test / 255.0
# Classes to categorical
num_classes = 10
y_train_label = y_train.reshape((-1, 1))
y_test_label = y_test.reshape((-1, 1))
y_train = utils.to_categorical(y_train, num_classes=num_classes)
y_test = utils.to_categorical(y_test, num_classes=num_classes)
# Shuffle dataset
x_train, y_train = utils.shuffle_dataset(x_train, y_train)
x_test, y_test = utils.shuffle_dataset(x_test, y_test)
# Params *********************************
batch_size = int(len(x_train) / 10)
epochs = 10
# Define architecture
l_in1 = Input(shape=x_train[0].shape)
l_in2 = Input(shape=x_train[0].shape)
l_in3 = Input(shape=x_train[0].shape)
l1 = Reshape(l_in1, shape=(28 * 28, ))
l2 = Reshape(l_in2, shape=(28 * 28, ))
l3 = Reshape(l_in2, shape=(28 * 28, ))
l1 = Tanh(Dense(l1, 300))
l2 = Relu(Dense(l2, 300))
l3 = Sigmoid(Dense(l3, 300))
l_mid1 = Add([l1, l2])
l_mid2 = Add([l1, l3])
l1 = Relu(Dense(l_mid1, 300))
l2 = Relu(Dense(l_mid2, 300))
l2 = Add([l1, l2])
l3 = Add([l2, l3])
l1 = BatchNorm(l1)
l_out1 = Softmax(Dense(l1, num_classes))
l_out2 = Softmax(Dense(l2, num_classes))
l_out3 = Relu(Dense(l3, 1))
# Build network
mymodel = Net()
mymodel.build(
l_in=[l_in1, l_in2, l_in3],
l_out=[l_out1, l_out2, l_out3],
optimizer=Adam(lr=0.001),
losses=[losses.CrossEntropy(),
losses.CrossEntropy(),
losses.MSE()],
metrics=[[metrics.CategoricalAccuracy()],
[metrics.CategoricalAccuracy()],
[metrics.MSE(), metrics.MAE()]],
debug=False,
smart_derivatives=True,
)
# Print model
mymodel.summary()
# Train
mymodel.fit([x_train, x_train, x_train], [y_train, y_train, y_train_label],
x_test=[x_test, x_test, x_test],
y_test=[y_test, y_test, y_test_label],
batch_size=batch_size,
epochs=epochs,
evaluate_epoch=False,
print_rate=1)
def main():
# Get data
(x_train, y_train), (x_test, y_test) = datasets.mnist.load_data()
# Pre-processing
# Add channel dimension
x_train = np.expand_dims(x_train, axis=1)
x_test = np.expand_dims(x_test, axis=1)
# Normalize
x_train = x_train / 255.0
x_test = x_test / 255.0
# Classes to categorical
num_classes = 10
y_train = utils.to_categorical(y_train, num_classes=num_classes)
y_test = utils.to_categorical(y_test, num_classes=num_classes)
# Shuffle dataset
x_train, y_train = utils.shuffle_dataset(x_train, y_train)
x_test, y_test = utils.shuffle_dataset(x_test, y_test)
# Params *********************************
batch_size = int(len(x_train) / 1)
epochs = 10
# Define architecture
l_in = Input(shape=x_train[0].shape)
l = l_in
l = Conv2D(l,
filters=2,
kernel_size=(3, 3),
strides=(1, 1),
padding="none")
l = MaxPool2D(l, pool_size=(3, 3), strides=(2, 2), padding="none")
l = LeakyRelu(l, alpha=0.0)
l = GaussianNoise(l, stddev=0.1)
# l = Conv2D(l, filters=4, kernel_size=(3, 3), strides=(1, 1), padding="same")
# l = MaxPool2D(l, pool_size=(3, 3), strides=(2, 2), padding="none")
# l = Relu(l)
# l = DepthwiseConv2D(l, kernel_size=(3, 3), strides=(1, 1), padding="none")
# l = PointwiseConv2D(l, filters=1)
# l = MaxPool2D(l, pool_size=(3, 3), strides=(2, 2), padding="none")
# l = Relu(l)
l = Reshape(l, shape=(-1))
l = Dense(l, num_classes, kernel_initializer=initializers.RandomUniform())
l_out = Softmax(l)
# Build network
mymodel = Net()
mymodel.build(
l_in=[l_in],
l_out=[l_out],
optimizer=Adam(lr=0.01),
losses=[losses.CrossEntropy()],
metrics=[[metrics.CategoricalAccuracy()]],
debug=False,
smart_derivatives=True,
)
# Print model
mymodel.summary()
# Train
mymodel.fit([x_train], [y_train],
x_test=[x_test],
y_test=[y_test],
batch_size=batch_size,
epochs=epochs,
evaluate_epoch=False,
print_rate=1)