def main():
""" load training data"""
inputs = np.loadtxt("../handwriting/X2_100samples.dat")
targets = np.loadtxt("../handwriting/y2_100samples.dat")
""" define network topology """
conec = mlgraph((inputs.shape[1], 10, 1))
# reg = 0.1
reg = False
net = ffnet(conec)
system = NNSystem(net, inputs, targets, reg=reg)
database = system.create_database(
# db="/home/ab2111/machine_learning_landscapes/neural_net/db_ffnet_100samples_reg"+str(reg) +".sqlite"
db="../db/db_ffnet_100samples.sqlite")
run_gui(system, database)
def RunSimulation(self, real_time_rate=1.0):
'''
The Princess Diaries was a good movie.
'''
builder = DiagramBuilder()
scene_graph = builder.AddSystem(SceneGraph())
# object_file_path = FindResourceOrThrow(
# "drake/examples/manipulation_station/models/061_foam_brick.sdf")
# sdf_file = FindResourceOrThrow("drake/multibody/benchmarks/acrobot/acrobot.sdf")
# urdf_file = FindResourceOrThrow("drake/multibody/benchmarks/acrobot/acrobot.urdf")
sdf_file = "assets/acrobot.sdf"
urdf_file = "assets/acrobot.urdf"
plant = builder.AddSystem(MultibodyPlant())
plant.RegisterAsSourceForSceneGraph(scene_graph)
Parser(plant, scene_graph).AddModelFromFile(sdf_file)
plant.Finalize(scene_graph)
assert plant.geometry_source_is_registered()
builder.Connect(
plant.get_geometry_poses_output_port(),
scene_graph.get_source_pose_port(plant.get_source_id()))
builder.Connect(scene_graph.get_query_output_port(),
plant.get_geometry_query_input_port())
# Add
nn_system = NNSystem(self.pytorch_nn_object)
builder.AddSystem(nn_system)
# NN -> plant
builder.Connect(nn_system.NN_out_output_port,
plant.get_actuation_input_port())
# plant -> NN
builder.Connect(plant.get_continuous_state_output_port(),
nn_system.NN_in_input_port)
# Add meshcat visualizer
meshcat = MeshcatVisualizer(scene_graph)
builder.AddSystem(meshcat)
# builder.Connect(scene_graph.GetOutputPort("lcm_visualization"),
builder.Connect(scene_graph.get_pose_bundle_output_port(),
meshcat.GetInputPort("lcm_visualization"))
# build diagram
diagram = builder.Build()
meshcat.load()
# time.sleep(2.0)
RenderSystemWithGraphviz(diagram)
# construct simulator
simulator = Simulator(diagram)
# context = diagram.GetMutableSubsystemContext(
# self.station, simulator.get_mutable_context())
simulator.set_publish_every_time_step(False)
simulator.set_target_realtime_rate(real_time_rate)
simulator.Initialize()
sim_duration = 5.
simulator.StepTo(sim_duration)
print("stepping complete")
def main():
""" load training data"""
inputs = np.loadtxt("../handwriting/X2_100samples.dat")
targets = np.loadtxt("../handwriting/y2_100samples.dat")
""" define network topology """
conec = mlgraph((inputs.shape[1],10,1))
# reg = 0.1
reg=False
net = ffnet(conec)
system = NNSystem(net, inputs, targets, reg=reg)
database = system.create_database(
# db="/home/ab2111/machine_learning_landscapes/neural_net/db_ffnet_100samples_reg"+str(reg) +".sqlite"
db="../db/db_ffnet_100samples.sqlite"
)
run_gui(system, database)
def RunSimulation(self, real_time_rate=1.0):
'''
Here we test using the NNSystem in a Simulator to drive
an acrobot.
'''
sdf_file = "assets/acrobot.sdf"
urdf_file = "assets/acrobot.urdf"
builder = DiagramBuilder()
plant, scene_graph = AddMultibodyPlantSceneGraph(builder)
parser = Parser(plant=plant, scene_graph=scene_graph)
parser.AddModelFromFile(sdf_file)
plant.Finalize(scene_graph)
# Add
nn_system = NNSystem(self.pytorch_nn_object)
builder.AddSystem(nn_system)
# NN -> plant
builder.Connect(nn_system.NN_out_output_port,
plant.get_actuation_input_port())
# plant -> NN
builder.Connect(plant.get_continuous_state_output_port(),
nn_system.NN_in_input_port)
# Add meshcat visualizer
meshcat = MeshcatVisualizer(scene_graph)
builder.AddSystem(meshcat)
# builder.Connect(scene_graph.GetOutputPort("lcm_visualization"),
builder.Connect(scene_graph.get_pose_bundle_output_port(),
meshcat.GetInputPort("lcm_visualization"))
# build diagram
diagram = builder.Build()
meshcat.load()
# time.sleep(2.0)
RenderSystemWithGraphviz(diagram)
# construct simulator
simulator = Simulator(diagram)
# context = diagram.GetMutableSubsystemContext(
# self.station, simulator.get_mutable_context())
simulator.set_publish_every_time_step(False)
simulator.set_target_realtime_rate(real_time_rate)
simulator.Initialize()
sim_duration = 5.
simulator.StepTo(sim_duration)
print("stepping complete")
dg.plot()
dg.label_minima(labels)
print labels
plt.show()
# dg.savefig("/home/ab2111/machine_learning_landscapes/neural_net/dg.png")
from NNSystem import NNSystem
""" load training data"""
inputs = np.loadtxt("../handwriting/X2_100samples.dat")
targets = np.loadtxt("../handwriting/y2_100samples.dat")
from ffnet_validation import get_validation_data
vinputs, vtargets = get_validation_data()
""" define network topology """
conec = mlgraph((inputs.shape[1],10,1))
print inputs.shape
# exit()
net = ffnet(conec)
system = NNSystem(net, inputs, targets)
database = system.create_database(db="../db/db_ffnet_100samples.sqlite")
# make_disconnectivity_graph(system, database, vinputs, vtargets)
# plt.plot(ts.coords,'x')
# plt.plot(ts.eigenvec,'o')
# plt.show()
make_validation_disconnectivity_graph(system, database)
def create_nn_policy_system(kNetConstructor, params_list):
net = create_nn(kNetConstructor, params_list)
net.eval()
nn_policy = NNSystem(pytorch_nn_object=net)
return nn_policy