def on_validation_epoch_end(self, trainer, pl_module):
node_emb = th.cat(self.val_outputs, 0)
g = trainer.datamodule.g
labels = g.ndata['labels']
f1_micro, f1_macro = compute_acc(
node_emb, labels, trainer.datamodule.train_nid,
trainer.datamodule.val_nid, trainer.datamodule.test_nid)
pl_module.log('val_f1_micro', f1_micro)
def evaluate(model, g, nfeat, labels, train_nids, val_nids, test_nids, device):
"""
Evaluate the model on the validation set specified by ``val_mask``.
g : The entire graph.
inputs : The features of all the nodes.
labels : The labels of all the nodes.
val_mask : A 0-1 mask indicating which nodes do we actually compute the accuracy for.
device : The GPU device to evaluate on.
"""
model.eval()
with th.no_grad():
# single gpu
if isinstance(model, SAGE):
pred = model.inference(g, nfeat, device, args.batch_size,
args.num_workers)
# multi gpu
else:
pred = model.module.inference(g, nfeat, device, args.batch_size,
args.num_workers)
model.train()
return compute_acc(pred, labels, train_nids, val_nids, test_nids)