dev(narugo): use new metrics and loss (736333a1) · Commits · git-mirror / Imgutils

zoo/ccip/loss.py

+3 −2

Original line number	Diff line number	Diff line
		@@ -33,9 +33,10 @@ class NTXentLoss(nn.Module):
		Inspired from https://blog.csdn.net/cziun/article/details/119118768 .
		"""

		def __init__(self, tau: float = 1.0):
		def __init__(self, tau: float = 1.0, eps: float = 1e-8):
		nn.Module.__init__(self)
		self.register_buffer('tau', torch.as_tensor(tau, dtype=torch.float))
		self.register_buffer('eps', torch.as_tensor(eps, dtype=torch.float))

		def forward(self, sim_tensors, state_tensors):
		"""
		@@ -44,4 +45,4 @@ class NTXentLoss(nn.Module):
		"""
		log_items = -torch.log(torch.softmax(sim_tensors / self.tau, dim=-1))
		positive_items = log_items[state_tensors]
		return positive_items.mean()
		return (positive_items.sum() + self.eps) / (positive_items.shape[0] + self.eps)

zoo/ccip/train_.py

+26 −30

Original line number	Diff line number	Diff line
		@@ -63,18 +63,20 @@ def _sample_analysis(poss, negs, svm_samples: int = 10000):
		else:
		s_poss, s_negs = poss, negs

		features = torch.cat([s_poss, s_negs])
		labels = torch.cat([torch.ones_like(s_poss), torch.zeros_like(s_negs)])
		s_poss, s_negs = s_poss.cpu(), s_negs.cpu()
		features = torch.cat([s_poss, s_negs]).detach().numpy()
		labels = torch.cat([torch.ones_like(s_poss), torch.zeros_like(s_negs)]).detach().numpy()

		model = svm.SVC(kernel='linear') # 线性核
		model.fit(features.reshape(-1, 1), labels)
		predictions = model.predict(features.reshape(-1, 1))

		coef = model.coef_.reshape(-1)[0].item()
		inter = model.intercept_.reshape(-1)[0].item()
		coef = model.coef_.reshape(-1)[0].tolist()
		inter = model.intercept_.reshape(-1)[0].tolist()
		threshold = -coef / inter

		return poss.mean(), poss.std(), negs.mean(), negs.std(), threshold, accuracy_score(labels, predictions)
		return poss.mean().item(), poss.std().item(), negs.mean().item(), negs.std().item(), \
		threshold, accuracy_score(labels, predictions)


		def train(dataset_dir: str, session_name: Optional[str] = None, from_ckpt: Optional[str] = None,
		@@ -101,15 +103,9 @@ def train(dataset_dir: str, session_name: Optional[str] = None, from_ckpt: Optio
		os.makedirs(_CKPT_DIR, exist_ok=True)
		writer = SummaryWriter(_log_dir)
		writer.add_custom_scalars({
		"general": {
		"accuracy": ["Multiline", ["train/accuracy", "test/accuracy"]],
		"false": ["Multiline", ["test/fn", "test/fp", "train/fn", "train/fp"]],
		},
		"test": {
		"false": ["Multiline", ["test/fn", "test/fp"]],
		},
		"train": {
		"false": ["Multiline", ["train/fn", "train/fp"]],
		"contrastive": {
		"train": ["Multiline", ["train/pos/mean", "train/neg/mean"]],
		"test": ["Multiline", ["test/pos/mean", "test/neg/mean"]],
		},
		})
		else:
		@@ -173,20 +169,20 @@ def train(dataset_dir: str, session_name: Optional[str] = None, from_ckpt: Optio
		epoch_loss = running_loss / train_pos_total
		train_psims = torch.cat(positive_sims)
		train_nsims = torch.cat(negative_sims)
		train_psim_mean, train_psim_std, train_msim_mean, train_msim_std, train_threshold, train_acc_svm = \
		train_pos_mean, train_pos_std, train_neg_mean, train_neg_std, train_threshold, train_acc_svm = \
		_sample_analysis(train_psims, train_nsims)

		if accelerator.is_local_main_process:
		logging.info(f'Epoch [{epoch}/{max_epochs}], loss: {epoch_loss:.6f}, '
		f'acc_scm: {train_acc_svm:.6f}, threshold: {train_threshold:.6f}.')
		f'acc_svm: {train_acc_svm:.6f}, threshold: {train_threshold:.6f}.')
		if writer:
		writer.add_scalar('train/loss', epoch_loss, epoch)
		writer.add_scalar('train/psim/mean', train_psim_mean, epoch)
		writer.add_scalar('train/psim/std', train_psim_std, epoch)
		writer.add_scalar('train/msim/mean', train_msim_mean, epoch)
		writer.add_scalar('train/msim/std', train_msim_std, epoch)
		writer.add_scalar('train/threshold', train_threshold)
		writer.add_scalar('train/acc_svm', train_acc_svm)
		writer.add_scalar('train/pos/mean', train_pos_mean, epoch)
		writer.add_scalar('train/pos/std', train_pos_std, epoch)
		writer.add_scalar('train/neg/mean', train_neg_mean, epoch)
		writer.add_scalar('train/neg/std', train_neg_std, epoch)
		writer.add_scalar('train/threshold', train_threshold, epoch)
		writer.add_scalar('train/acc_svm', train_acc_svm, epoch)

		model.eval()
		if epoch % eval_epoch == 0:
		@@ -208,19 +204,19 @@ def train(dataset_dir: str, session_name: Optional[str] = None, from_ckpt: Optio

		test_psims = torch.cat(positive_sims)
		test_nsims = torch.cat(negative_sims)
		test_psim_mean, test_psim_std, test_msim_mean, test_msim_std, test_threshold, test_acc_svm = \
		test_pos_mean, test_pos_std, test_neg_mean, test_neg_std, test_threshold, test_acc_svm = \
		_sample_analysis(test_psims, test_nsims)

		if accelerator.is_local_main_process:
		logging.info(f'Epoch {epoch}, '
		f'acc_scm: {test_acc_svm:.6f}, threshold: {test_threshold:.6f}')
		f'acc_svm: {test_acc_svm:.6f}, threshold: {test_threshold:.6f}')
		if writer:
		writer.add_scalar('test/psim/mean', test_psim_mean, epoch)
		writer.add_scalar('test/psim/std', test_psim_std, epoch)
		writer.add_scalar('test/msim/mean', test_msim_mean, epoch)
		writer.add_scalar('test/msim/std', test_msim_std, epoch)
		writer.add_scalar('test/threshold', test_threshold)
		writer.add_scalar('test/acc_svm', test_acc_svm)
		writer.add_scalar('test/pos/mean', test_pos_mean, epoch)
		writer.add_scalar('test/pos/std', test_pos_std, epoch)
		writer.add_scalar('test/neg/mean', test_neg_mean, epoch)
		writer.add_scalar('test/neg/std', test_neg_std, epoch)
		writer.add_scalar('test/threshold', test_threshold, epoch)
		writer.add_scalar('test/acc_svm', test_acc_svm, epoch)

		if accelerator.is_local_main_process and epoch % save_per_epoch == 0:
		current_ckpt_file = os.path.join(_CKPT_DIR, f'ccip-{session_name}-{epoch}.ckpt')