def __init__(self, config, dataset, train_loader, subgraph_loader, eval_split='test', \
chkpt=None, get_emb=False, get_logits=False, get_attn=False):
super().__init__()
self.config = config
self.dataset = dataset
self.batch_loader = train_loader
self.subgraph_loader = subgraph_loader
self.learning_rate = config['lr']
self.get_emb = get_emb
self.get_logits = get_logits
self.get_attn = get_attn
self.net = NsLstmGNN(self.config)
if chkpt is not None:
self.net.load_state_dict(chkpt['state_dict'], strict=False)
print('Loaded states')
self.loss = get_loss_function(config['task'], config['class_weights'])
self.collect_outputs = lambda x: collect_outputs(
x, config['multi_gpu'])
self.compute_metrics = lambda truth, pred: get_metrics(
truth, pred, config['verbose'], config['classification'])
self.per_node_metrics = lambda truth, pred: get_per_node_result(
truth, pred, self.dataset.idx_test, config['classification'])
self.eval_split = eval_split
self.eval_mask = self.dataset.data.val_mask if eval_split == 'test' else self.dataset.data.test_mask
entire_set = LstmDataset(config)
collate = lambda x: collate_fn(x, config['task'])
self.ts_loader = DataLoader(entire_set, collate_fn=collate, \
batch_size=config['batch_size'], num_workers=config['num_workers'], shuffle=False)
self.lg_alpha = config['lg_alpha']
def test(model, ds, criterion, device, dataset_type="test"):
"""
Evaluates and prints the result to the console.
"""
test_dl = utils.get_dl(ds, bs=64, shuffle=False)
test_losses, test_acts, test_dog_human_targets, \
test_breed_targets = engine.eval_loop(test_dl, model,
criterion, device)
test_metrics = metrics.get_metrics(
test_losses, test_acts, test_dog_human_targets, test_breed_targets,
criterion.dog_idx, criterion.human_idx)
output = []
output.append([f'{dataset_type}_loss', f'{dataset_type}_acc_human',
f'{dataset_type}_acc_dog', f'{dataset_type}_acc_breed',
f'{dataset_type}_acc_f1score'])
output.append([
test_metrics['loss'],
test_metrics['accuracy_human'].item(),
test_metrics['accuracy_dog'].item(),
test_metrics['accuracy_breed'].item(),
test_metrics['f1score_breed']])
print(tabulate(output))
return {
'loss': test_losses,
'acts': test_acts,
'dog_human_targets': test_dog_human_targets,
'breed_targets': test_breed_targets
}
def __init__(self,
config,
dataset,
train_loader,
subgraph_loader,
eval_split='test'):
super().__init__()
self.config = config
self.dataset = dataset
self.batch_loader = train_loader
self.subgraph_loader = subgraph_loader
self.learning_rate = config['lr']
self.net = NsGNN(config)
self.loss = get_loss_function(config['task'], config['class_weights'])
self.collect_outputs = lambda x: collect_outputs(
x, config['multi_gpu'])
self.compute_metrics = lambda x: get_metrics(x['truth'], x[
'pred'], config['verbose'], config['classification'])
self.per_node_metrics = lambda x: get_per_node_result(
x['truth'], x['pred'], self.dataset.idx_test, config[
'classification'])
self.eval_split = eval_split
self.eval_mask = self.dataset.data.val_mask if eval_split == 'val' else self.dataset.data.test_mask
def back_test_async(i, parametros, df, data_inicio, data_final):
operacoes, param = back_test(parametros, df, data_inicio, data_final)
results = metrics.get_metrics(operacoes, parametros[0] * 25000)
output = {'iteracao': i}
output.update(param)
output.update(results)
return output
def __init__(self, config, collate, train_set=None, val_set=None, test_set=None):
super().__init__()
self.config = config
self.trainset = train_set
self.validset = val_set
self.testset = test_set
self.learning_rate = self.config['lr']
self.collate = collate
self.task = config['task']
self.is_cls = config['classification']
self.net = DynamicLstmGnn(config)
self.loss = get_loss_function(self.task, config['class_weights'])
self.collect_outputs = lambda x: collect_outputs(x, config['multi_gpu'])
self.compute_metrics = lambda truth, pred : get_metrics(truth, pred, config['verbose'], config['classification'])
self.per_node_metrics = lambda truth, pred : get_per_node_result(truth, pred, self.testset.idx_test, config['classification'])
self.lg_alpha = config['lg_alpha']
def __init__(self,
config,
dataset,
train_loader,
subgraph_loader,
eval_split='test',
get_lstm_out=False,
get_logits=False):
super().__init__()
self.config = config
self.dataset = dataset
self.batch_loader = train_loader
self.subgraph_loader = subgraph_loader
self.learning_rate = config['lr']
self.get_lstm_out = get_lstm_out
self.get_logits = get_logits
self.net = Net(config)
self.loss = get_loss_function(config['task'], config['class_weights'])
self.collect_outputs = lambda x: collect_outputs(
x, config['multi_gpu'])
self.compute_metrics = lambda x: get_metrics(x['truth'], x[
'pred'], config['verbose'], config['classification'])
self.per_node_metrics = lambda x: get_per_node_result(
x['truth'], x['pred'], self.dataset.idx_test, config[
'classification'])
self.eval_split = eval_split
self.eval_mask = self.dataset.data.val_mask if eval_split == 'val' else self.dataset.data.test_mask
entire_set = LstmDataset(config)
collate = lambda x: collate_fn(x, config['task'])
self.ts_loader = DataLoader(entire_set, collate_fn=collate, \
batch_size=config['batch_size'], num_workers=config['num_workers'], shuffle=False)