def main_link_pred(models, args, data_loader):
for model, param_grid in models.items():
for params in param_grid:
if args.h_dim is None:
params["h_dim"] = args.node_emb_size
else:
params["h_dim"] = args.h_dim
params["num_steps"] = args.n_layers
date_time = str(datetime.now())
print("\n\n")
print(date_time)
print(f"Model: {model.__name__}, Params: {params}")
model_attempt = get_name(model, date_time)
model_base = get_model_base(args, model_attempt)
dataset = TestLinkPredGraph(data_loader=data_loader)
args.objectives = "link_pred"
from SourceCodeTools.models.graph.train.sampling_multitask2 import training_procedure
trainer, scores = training_procedure(dataset,
model,
copy(params),
args,
model_base,
trainer=TestTrainer)
return scores
def main(models, args):
for model, param_grid in models.items():
for params in param_grid:
if args.h_dim is None:
params["h_dim"] = args.node_emb_size
else:
params["h_dim"] = args.h_dim
date_time = str(datetime.now())
print("\n\n")
print(date_time)
print(f"Model: {model.__name__}, Params: {params}")
model_attempt = get_name(model, date_time)
model_base = get_model_base(args, model_attempt)
dataset = read_or_create_gnn_dataset(args=args, model_base=model_base)
from SourceCodeTools.models.graph.train.sampling_multitask2 import evaluation_procedure
checkpoints = detect_checkpoint_files(model_base)
for epoch, ckpt_path in checkpoints:
shutil.copy(ckpt_path, os.path.join(model_base, "saved_state.pt"))
evaluation_procedure(dataset, model, params, args, model_base)
def main(models, args):
for model, param_grid in models.items():
for params in param_grid:
if args.h_dim is None:
params["h_dim"] = args.node_emb_size
else:
params["h_dim"] = args.h_dim
params["num_steps"] = args.n_layers
date_time = str(datetime.now())
print("\n\n")
print(date_time)
print(f"Model: {model.__name__}, Params: {params}")
model_attempt = get_name(model, date_time)
model_base = get_model_base(args, model_attempt)
dataset = read_or_create_gnn_dataset(args=args,
model_base=model_base)
def write_params(args, params):
args = copy(args.__dict__)
args.update(params)
args['activation'] = args['activation'].__name__
with open(join(model_base, "params.json"), "w") as mdata:
mdata.write(json.dumps(args, indent=4))
if not args.restore_state:
write_params(args, params)
from SourceCodeTools.models.graph.train.sampling_multitask2 import training_procedure
trainer, scores = \
training_procedure(dataset, model, copy(params), args, model_base)
trainer.save_checkpoint(model_base)
print("Saving...")
params['activation'] = params['activation'].__name__
metadata = {
"base": model_base,
"name": model_attempt,
"parameters": params,
"layers": "embeddings.pkl",
"mappings": "nodes.csv",
"state": "state_dict.pt",
"scores": scores,
"time": date_time,
}
metadata.update(args.__dict__)
# pickle.dump(dataset, open(join(model_base, "dataset.pkl"), "wb"))
import pickle
pickle.dump(trainer.get_embeddings(),
open(join(model_base, metadata['layers']), "wb"))
with open(join(model_base, "metadata.json"), "w") as mdata:
mdata.write(json.dumps(metadata, indent=4))
print("Done saving")