def test_lenet(self):
MNIST_must_converge('lenet', lenet.LeNet,
optimizers.AdamOptimizer,
# optimizers.MomentumOptimizer,
initial_learning_rate=1e-3,
batch_size=32,
epochs=5)
def test_resnet_must_converge_on_MNIST(self):
MNIST_must_converge("Resnet", resnet.ResNet,
optimizers.AdamOptimizer,
batch_size=32,
epochs=3,
initial_learning_rate=0.001,
)
def test_lenet(self):
model = lenet.LeNet
MNIST_must_converge(model,
optimizers.MomentumOptimizer,
batch_size=100,
epochs=90)
def test_single_layer(self):
model = mlp.MultiLayerPerceptron
MNIST_must_converge('mlpx1', model,
optimizers.RMSPropOptimizer,
initial_learning_rate=0.1,
batch_size=128,
epochs=epochs)
def test_vgg_must_converge_on_MNIST(self):
MNIST_must_converge("vgg16", vgg.VGG16,
optimizers.AdamOptimizer,
batch_size=16,
epochs=3,
initial_learning_rate=1e-3,
summaries=False,
use_debug_session=False,
)
def test_mlp_2048_2048_Momentum(self):
hidden_layers = [2048, 2048, 2048]
dropout = [0.2, 0.5, 0.5]
model = partial(mlp.MultiLayerPerceptron,
hidden_layers=hidden_layers,
dropout=dropout)
MNIST_must_converge(model,
optimizers.MomentumOptimizer,
batch_size=100,
epochs=90)
def xxx_test_mlp_2048_2048_No_Dropout_Gradient(self):
hidden_layers = [2048, 2048, 2048]
dropout = []
model = partial(mlp.MultiLayerPerceptron,
hidden_layers=hidden_layers,
dropout=dropout)
MNIST_must_converge(model,
optimizers.GradientDescentOptimizer,
batch_size=100,
epochs=90)
def xxx_test_mlp_layer_with_dropout(self):
hidden_layers = [1024, 1024]
dropout = [0.2, 0.5]
model = partial(mlp.MultiLayerPerceptron,
hidden_layers=hidden_layers,
dropout=dropout)
MNIST_must_converge(model,
optimizers.GradientDescentOptimizer,
batch_size=100,
epochs=90)
def test_mlp_2048_2048_momentum(self):
hidden_layers = [2048, 2048, 2048]
dropout = [0.2, 0.5, 0.5]
model = partial(mlp.MultiLayerPerceptron,
hidden_layers=hidden_layers,
dropout=dropout)
MNIST_must_converge('mlpx2048x2048x2048', model,
optimizers.RMSPropOptimizer,
initial_learning_rate=0.1,
batch_size=128,
epochs=epochs)
def test_mlp_2048_2048_no_dropout_gradient(self):
hidden_layers = [2048, 2048, 2048]
dropout = []
model = partial(mlp.MultiLayerPerceptron,
hidden_layers=hidden_layers,
dropout=dropout)
MNIST_must_converge('mlpx2048x2048x2048xNoDropout', model,
optimizers.RMSPropOptimizer,
initial_learning_rate=0.1,
batch_size=32,
epochs=epochs)
def test_mlp_layer_with_dropout(self):
hidden_layers = [1024, 1024]
dropout = [0.2, 0.5]
model = partial(mlp.MultiLayerPerceptron,
hidden_layers=hidden_layers,
dropout=dropout, activation_fn="tanh")
MNIST_must_converge('mlpx1024x1024', model,
optimizers.GradientDescentOptimizer,
initial_learning_rate=0.1,
batch_size=128,
epochs=epochs)
def xxx_test_single_layer(self):
model = mlp.MultiLayerPerceptron
MNIST_must_converge(model,
optimizers.GradientDescentOptimizer,
batch_size=100,
epochs=90)
