def test_run_topology_in_ui(self):
def callback():
callback.called = True
callback.called = False
ObserverSystem.initialized = False
observer_system = ObserverSystemVoid(None)
run_topology_with_ui(TopologyStub(callback), max_steps=1, observer_system=observer_system)
assert callback.called
def run_task1_conv():
bt = SeT1ConvTopologicalGraph()
run_topology_with_ui(bt)
def run_task1():
bt = SeT1Bt()
run_topology_with_ui(bt)
def run_task0_conv(use_dataset: bool, save_gpu_memory: bool):
topology = SeT0ConvTopology(use_dataset=use_dataset, save_gpu_memory=save_gpu_memory)
run_topology_with_ui(topology)
def run_task0_convSP(use_dataset: bool):
topology = SeT0ConvSPTopology(use_dataset=use_dataset)
run_topology_with_ui(topology)
def run_task0_narrow(use_dataset: bool):
topology = SeT0NarrowHierarchy(use_dataset=use_dataset)
run_topology_with_ui(topology, create_non_persisting_observer_system())
def run_task0(use_dataset: bool):
bt = SeT0BasicTopology(use_dataset=use_dataset)
run_topology_with_ui(bt, create_non_persisting_observer_system())
def main():
parser = argparse.ArgumentParser()
parser.add_argument("--model", default="se_dataset_navigation")
parser.add_argument("--seed", type=int, help="Global random seed.")
parser.add_argument("--random-seed",
action='store_true',
help="Use random seed. Overrides the seed argument.")
args = parser.parse_args()
# UI persistence storage file - set to None to turn off the persistence
storage_file = 'observers.yaml'
if args.random_seed:
seed = None
else:
seed = args.seed if args.seed is not None else 1337
def topology_factory(key) -> Topology:
if key == "SP":
# works
model = SpatialPoolerTopology()
elif key == "SPTP":
# not updated
model = SpatialTemporalPoolerTopology()
elif key == "seq":
# not updated
model = SequenceTopology()
elif key == "noise":
# works
model = RandomNoiseTopology()
elif key == "noise_only":
# works
model = RandomNoiseOnlyTopology()
elif key == "random_num":
# works
model = RandomNumberTopology()
elif key == "switch":
# works
model = SwitchTopology()
elif key == "mnist":
# works
model = MnistTopology()
elif key == "mnist_seq":
# works
model = SequenceMnistTopology()
elif key == "SP_hierarchy":
# works
model = SpatialPoolerHierarchy()
elif key == "bouncing_ball":
# works
model = BouncingBallTopology()
elif key == "debug_agent":
# works (probably)
model = DebugAgentTopology()
elif key == "se_ta":
# WIP
model = SeToyArchDebugTopology()
elif key == "expert":
# works
model = ExpertTopology()
elif key == "expert_hierarchy":
# works
model = ExpertHierarchyTopology()
elif key == "se_dataset_navigation":
model = SEDatasetSPRepresentationTopology()
elif key == "se_dataset_objects":
model = SeDatasetObjectsTopology()
elif key == "grid_world":
model = GridWorldTopology()
elif key == "looping":
model = LoopingTopology()
elif key == "conv_sp":
model = ConvSpatialPoolerTopology()
elif key == "mse":
model = MseDemoTopology()
elif key == "actions_grid_world":
model = TaActionsGridWorldTopology()
elif key == "rtx_narrow":
model = Task0NarrowTopology()
elif key == "sample_collection":
model = SEDatasetSampleCollectionTopology()
elif key == "nnet":
model = NNetTopology()
elif key == "l3_conv_topology":
model = L3ConvTopology()
elif key == "receptive_field":
model = ReceptiveFieldTopology()
elif key == "context_test":
model = ContextTestTopology()
elif key == "exploration_grid_world":
model = TaExplorationGridWorldTopology()
elif key == "bottom_up_attention":
model = BottomUpAttentionTopology()
elif key == "lrf_object_detection_topology":
model = LrfObjectDetectionTopology()
elif key == "disentangled_world_node_topology":
model = DisentangledWorldNodeTopology()
elif key == "net":
model = NetworkFlockTopology()
elif key == "goal_directed":
model = GoalDirectedTopology()
elif key == "goal_directed_narrow":
model = GoalDirectedNarrowHierarchyTopology()
elif key == "gl_nn":
model = GlNnTopology()
elif key == "gl_nn_fg":
model = GlFakeGateNnTopology()
elif key == "super_hard":
model = SingleExpertExperimentalTopology()
elif key == "multi_dataset_alphabet":
model = MultiDatasetAlphabetTopology()
elif key == "nn-gradual-learning-demo":
model = setup_demo_model()
else:
raise NotImplementedError
model.assign_ids()
return model
# Create simulation, it is registers itself to observer_system
with observer_system_context(storage_file) as observer_system:
# run_topology_factory_with_ui(topology_factory=topology_factory,
# topology_params={'key': args.model},
# seed=seed,
# max_steps=0,
# auto_start=False,
# observer_system=observer_system)
# A simpler version which doesn't recreate the topology, just restarts.
run_topology_with_ui(topology=topology_factory(args.model),
seed=seed,
max_steps=0,
auto_start=False,
observer_system=observer_system)
print('Running simulation, press enter to quit.')
input()
flock_lr=0.01,
gate_lr=0.05, # 0.005 worked, 0.001 is safer
coefficients_min_max=0.3, # 0.001 used
predictor_hidden_s=30,
predictor_n_layers=3,
use_teleport=True)
else:
model = SwitchableWorldTopology(temporal_class_definitions=classes, sx=sx, sy=sy)
return model
if __name__ == '__main__':
# configure the experiment
model = setup_demo_model()
model.assign_ids()
seed = 1123
# Create simulation, it is registers itself to observer_system
with observer_system_context('observers.yaml') as observer_system:
run_topology_with_ui(topology=model,
seed=seed,
max_steps=0,
auto_start=False,
observer_system=observer_system)
print('Running simulation, press enter to quit.')
input()
def main():
parser = argparse.ArgumentParser()
parser.add_argument("--model", default="gl_nn")
parser.add_argument("--sequences", default="s0", nargs='*')
parser.add_argument("--seed", type=int, help="Global random seed.")
parser.add_argument("--n-predictors",
type=int,
default=2,
help="Number of predictors.")
parser.add_argument("--random-seed",
action='store_true',
help="Use random seed. Overrides the seed argument.")
args = parser.parse_args()
# UI persistence storage file - set to None to turn off the persistence
storage_file = 'observers.yaml'
if args.random_seed:
seed = None
else:
seed = args.seed if args.seed is not None else 1337
def topology_factory(key, seq_list) -> Topology:
fourth_sequence_excluded = np.array([[0.8, 0.1, 0.1, 0.0],
[0.1, 0.8, 0.1, 0.0],
[0.1, 0.1, 0.8, 0.0],
[1.0, 0.0, 0.0, 0.0]])
cycle_except_third = np.array([[0.0, 1.0, 0.0, 0.0],
[0.0, 0.0, 0.0, 1.0],
[0.0, 0.0, 0.0, 1.0],
[1.0, 0.0, 0.0, 0.0]])
sequence_table = {
"s0":
SequenceGenerator([
[1, 2, 3, 1, 2, 3, 1, 2, 3],
[5, 4, 3, 5, 4, 3, 5, 4, 3],
[1, 2, 3, 1, 2, 3, 1, 2, 3],
[5, 4, 3, 5, 4, 3, 5, 4, 3],
], diagonal_transition_matrix(4, 0.8)),
"s1":
SequenceGenerator([
[1, 2, 3, 1, 2, 3, 1, 2, 3],
[2, 1, 3, 2, 1, 3, 2, 1, 3],
[1, 2, 3, 1, 2, 3, 1, 2, 3],
[5, 4, 3, 5, 4, 3, 5, 4, 3],
], diagonal_transition_matrix(4, 0.8)),
"s2":
SequenceGenerator(
[[1, 2, 3, 1, 2, 3, 1, 2, 3], [2, 1, 3, 2, 1, 3, 2, 1, 3],
[1, 2, 3, 1, 2, 3, 1, 2, 3], [1]], fourth_sequence_excluded),
"test":
SequenceGenerator([
[1, 2, 3, 1, 2, 3, 1, 2, 3],
[2, 1, 3, 2, 1, 3, 2, 1, 3],
[1, 2, 3, 1, 2, 3, 1, 2, 3],
[5, 4, 3, 5, 4, 3, 5, 4, 3],
], cycle_except_third)
}
seqs = [sequence_table[seq] for seq in seq_list]
if key == "gl_nn":
model = GlNnTopology(seqs, args.n_predictors)
elif key == "gl_nn_fg":
model = GlFakeGateNnTopology()
else:
raise NotImplementedError
model.assign_ids()
return model
# Create simulation, it is registers itself to observer_system
with observer_system_context(storage_file) as observer_system:
run_topology_with_ui(topology=topology_factory(args.model,
args.sequences),
seed=seed,
max_steps=0,
auto_start=False,
observer_system=observer_system)
print('Running simulation, press enter to quit.')
input()