def make_cluster_dataloader(self, X, y, shuffle):
# return a data loader based on ClusterGCN
dataset = ClusterDataset(X,
y,
self.edge_index,
self.edge_weight,
num_nodes=self.hparams.num_nodes,
batch_size=batch_size,
shuffle=shuffle)
return DataLoader(dataset, batch_size=None)
opt = parser.parse_args()
if opt.clusters < 0:
# === pre-clustered environment according to environment file ===========
from room_dataset import RoomDataset
testset = RoomDataset("test", training=False)
ensemble = ExpertEnsemble(testset.num_experts)
else:
# === large, connected environment, perform clustering ==================
from cluster_dataset import ClusterDataset
testset = ClusterDataset("test", num_clusters=opt.clusters, training=False)
ensemble = ExpertEnsemble(
testset.num_experts, gating_capacity=2
) # for clustering environments we use a gating network with higher capacity
testset_loader = torch.utils.data.DataLoader(testset,
shuffle=False,
num_workers=6)
if opt.testrefined:
# load individual, refined experts
ensemble.load_experts(opt.session, True)
elif opt.testinit:
help='robust square root loss after this threshold')
opt = parser.parse_args()
if opt.clusters < 0:
# === pre-clustered environment according to environment file ===========
from dataset import RoomDataset
trainset = RoomDataset("training", scene=opt.expert)
else:
# === large, connected environment, perform clustering ==================
from cluster_dataset import ClusterDataset
trainset = ClusterDataset("training",
num_clusters=opt.clusters,
cluster=opt.expert)
trainset_loader = torch.utils.data.DataLoader(trainset,
shuffle=True,
num_workers=6)
model = Expert(torch.zeros((3, )))
model.load_state_dict(
torch.load('expert_e%d_%s.net' % (opt.expert, opt.session)))
print("Successfully loaded model.")
model.cuda()
model.train()
if opt.clusters < 0:
# === pre-clustered environment according to environment file ===========
from room_dataset import RoomDataset
trainset = RoomDataset("training", training=True)
ensemble = ExpertEnsemble(trainset.num_experts, lr=opt.learningrate)
else:
# === large, connected environment, perform clustering ==================
from cluster_dataset import ClusterDataset
trainset = ClusterDataset("training",
num_clusters=opt.clusters,
training=True)
ensemble = ExpertEnsemble(
trainset.num_experts, lr=opt.learningrate, gating_capacity=2
) # for clustering environments we use a gating network with higher capacity
trainset_loader = torch.utils.data.DataLoader(trainset,
shuffle=True,
num_workers=6)
ensemble.load_experts(opt.session, opt.refined)
if opt.expertselection:
model_file = 'es_%s.net' % (opt.session)
train_log = open('log_es_%s.txt' % (opt.session), 'w', 1)
else:
model_file = 'esac_%s.net' % (opt.session)