def prepare(train_data, train_params, arch_gcn):
"""
Prepare some data structure and initialize model / minibatch handler before
the actual iterative training taking place.
"""
adj_full, adj_train, feat_full, class_arr, role = train_data
adj_full = adj_full.astype(np.int32)
adj_train = adj_train.astype(np.int32)
adj_full_norm = adj_norm(adj_full)
num_classes = class_arr.shape[1]
minibatch = Minibatch(adj_full_norm, adj_train, role, train_params)
model = GraphSAINT(num_classes, arch_gcn, train_params, feat_full,
class_arr)
printf("TOTAL NUM OF PARAMS = {}".format(
sum(p.numel() for p in model.parameters())),
style="yellow")
minibatch_eval = Minibatch(adj_full_norm,
adj_train,
role,
train_params,
cpu_eval=True)
model_eval = GraphSAINT(num_classes,
arch_gcn,
train_params,
feat_full,
class_arr,
cpu_eval=True)
if args_global.gpu >= 0:
model = model.cuda()
return model, minibatch, minibatch_eval, model_eval
def prepare(train_data, train_params, arch_gcn):
adj_full, adj_train, feat_full, class_arr, role = train_data
adj_full = adj_full.astype(np.int32)
adj_train = adj_train.astype(np.int32)
adj_full_norm = adj_norm(adj_full)
num_classes = class_arr.shape[1]
minibatch = Minibatch(adj_full_norm, adj_train, role, train_params)
model = GraphSAINT(num_classes, arch_gcn, train_params, feat_full,
class_arr)
if args_global.gpu >= 0:
model = model.cuda()
return model, minibatch