def start_fold(sim_data_node,
data_dict,
flags,
hyper_params,
prot_desc_dict,
tasks,
trainer,
transformers_dict,
view,
prot_profile,
k=None):
data = trainer.data_provider(k, flags, data_dict)
model, optimizer, data_loaders, metrics, frozen_models = trainer.initialize(
hparams=hyper_params,
train_dataset=data["train"],
val_dataset=data["val"],
test_dataset=data["test"],
protein_profile=prot_profile)
if flags["eval"]:
trainer.evaluate_model(model,
flags["model_dir"],
flags["eval_model_name"],
data_loaders,
metrics,
transformers_dict,
prot_desc_dict,
tasks,
view_lbl=view,
sim_data_node=sim_data_node)
elif flags["explain"]:
trainer.explain_model(model, flags["model_dir"],
flags["eval_model_name"], data_loaders,
transformers_dict, prot_desc_dict, view,
sim_data_node)
else:
# Train the model
results = trainer.train(model,
optimizer,
data_loaders,
metrics,
frozen_models,
transformers_dict,
prot_desc_dict,
tasks,
n_iters=10000,
view_lbl=view,
sim_data_node=sim_data_node)
model, score, epoch = results['model'], results['score'], results[
'epoch']
# Save the model.
split_label = flags.split
save_model(
model, flags["model_dir"],
"{}_pcnna_{}_{}_{}_{}_{:.4f}".format(flags["dataset_name"], view,
flags["model_name"],
split_label, epoch, score))
def start_fold(sim_data_node,
data_dict,
flags,
hyper_params,
prot_desc_dict,
tasks,
trainer,
transformers_dict,
tb_writer,
view,
k=None):
data = trainer.data_provider(k, flags, data_dict)
model, optimizer, data_loaders, metrics, weighted_loss, n_dist = trainer.initialize(
hparams=hyper_params,
train_dataset=data["train"],
val_dataset=data["val"],
test_dataset=data["test"])
if flags["eval"]:
trainer.evaluate_model(model[0],
flags["model_dir"],
flags["eval_model_name"],
data_loaders,
metrics,
transformers_dict,
prot_desc_dict,
tasks,
sim_data_node=sim_data_node)
else:
# Train the model
results = trainer.train(model,
optimizer,
data_loaders,
metrics,
weighted_loss,
n_dist,
transformers_dict,
prot_desc_dict,
tasks,
n_iters=10000,
sim_data_node=sim_data_node,
tb_writer=tb_writer)
model, score, epoch = results['model'], results['score'], results[
'epoch']
# Save the model.
split_label = "warm" if flags["split_warm"] else "cold_target" if flags["cold_target"] else "cold_drug" if \
flags["cold_drug"] else "None"
save_model(
model, flags["model_dir"],
"{}_{}_{}_{}_{}_{:.4f}".format(flags["dataset_name"], view,
flags["model_name"], split_label,
epoch, score))
def start_fold(sim_data_node, data_dict, flags, hyper_params, prot_desc_dict, tasks, trainer,
transformers_dict, view, protein_profile, summary_writer_creator=None, k=None):
data = trainer.data_provider(k, flags, data_dict)
model, optimizer, data_loaders, metrics = trainer.initialize(hparams=hyper_params,
train_dataset=data["train"],
val_dataset=data["val"],
test_dataset=data["test"],
protein_profile=protein_profile)
if flags["eval"]:
trainer.evaluate_model(model, flags["model_dir"], flags["eval_model_name"],
data_loaders, metrics, transformers_dict,
prot_desc_dict, tasks, view=view, sim_data_node=sim_data_node)
else:
# Train the model
results = trainer.train(model, optimizer, data_loaders, metrics, transformers_dict, prot_desc_dict,
tasks, n_iters=10000, view=view, sim_data_node=sim_data_node,
tb_writer=summary_writer_creator)
model, score, epoch = results['model'], results['score'], results['epoch']
# Save the model.
save_model(model, flags["model_dir"],
"{}_{}_{}_{}_{}_{:.4f}".format(flags["dataset_name"], '_'.join(view), flags["model_name"],
flags.split,
epoch, score))