first commit

exp/exp_anomaly_detection.py

@@ -0,0 +1,207 @@
+from data_provider.data_factory import data_provider
+from exp.exp_basic import Exp_Basic
+from utils.tools import EarlyStopping, adjust_learning_rate, adjustment
+from sklearn.metrics import precision_recall_fscore_support
+from sklearn.metrics import accuracy_score
+import torch.multiprocessing
+
+torch.multiprocessing.set_sharing_strategy('file_system')
+import torch
+import torch.nn as nn
+from torch import optim
+import os
+import time
+import warnings
+import numpy as np
+
+warnings.filterwarnings('ignore')
+
+
+class Exp_Anomaly_Detection(Exp_Basic):
+    def __init__(self, args):
+        super(Exp_Anomaly_Detection, self).__init__(args)
+
+    def _build_model(self):
+        model = self.model_dict[self.args.model].Model(self.args).float()
+
+        if self.args.use_multi_gpu and self.args.use_gpu:
+            model = nn.DataParallel(model, device_ids=self.args.device_ids)
+        return model
+
+    def _get_data(self, flag):
+        data_set, data_loader = data_provider(self.args, flag)
+        return data_set, data_loader
+
+    def _select_optimizer(self):
+        model_optim = optim.Adam(self.model.parameters(), lr=self.args.learning_rate)
+        return model_optim
+
+    def _select_criterion(self):
+        criterion = nn.MSELoss()
+        return criterion
+
+    def vali(self, vali_data, vali_loader, criterion):
+        total_loss = []
+        self.model.eval()
+        with torch.no_grad():
+            for i, (batch_x, _) in enumerate(vali_loader):
+                batch_x = batch_x.float().to(self.device)
+
+                outputs = self.model(batch_x, None, None, None)
+
+                f_dim = -1 if self.args.features == 'MS' else 0
+                outputs = outputs[:, :, f_dim:]
+                pred = outputs.detach()
+                true = batch_x.detach()
+
+                loss = criterion(pred, true)
+                total_loss.append(loss.item())
+        total_loss = np.average(total_loss)
+        self.model.train()
+        return total_loss
+
+    def train(self, setting):
+        train_data, train_loader = self._get_data(flag='train')
+        vali_data, vali_loader = self._get_data(flag='val')
+        test_data, test_loader = self._get_data(flag='test')
+
+        path = os.path.join(self.args.checkpoints, setting)
+        if not os.path.exists(path):
+            os.makedirs(path)
+
+        time_now = time.time()
+
+        train_steps = len(train_loader)
+        early_stopping = EarlyStopping(patience=self.args.patience, verbose=True)
+
+        model_optim = self._select_optimizer()
+        criterion = self._select_criterion()
+
+        for epoch in range(self.args.train_epochs):
+            iter_count = 0
+            train_loss = []
+
+            self.model.train()
+            epoch_time = time.time()
+            for i, (batch_x, batch_y) in enumerate(train_loader):
+                iter_count += 1
+                model_optim.zero_grad()
+
+                batch_x = batch_x.float().to(self.device)
+
+                outputs = self.model(batch_x, None, None, None)
+
+                f_dim = -1 if self.args.features == 'MS' else 0
+                outputs = outputs[:, :, f_dim:]
+                loss = criterion(outputs, batch_x)
+                train_loss.append(loss.item())
+
+                if (i + 1) % 100 == 0:
+                    print("\titers: {0}, epoch: {1} | loss: {2:.7f}".format(i + 1, epoch + 1, loss.item()))
+                    speed = (time.time() - time_now) / iter_count
+                    left_time = speed * ((self.args.train_epochs - epoch) * train_steps - i)
+                    print('\tspeed: {:.4f}s/iter; left time: {:.4f}s'.format(speed, left_time))
+                    iter_count = 0
+                    time_now = time.time()
+
+                loss.backward()
+                model_optim.step()
+
+            print("Epoch: {} cost time: {}".format(epoch + 1, time.time() - epoch_time))
+            train_loss = np.average(train_loss)
+            vali_loss = self.vali(vali_data, vali_loader, criterion)
+            test_loss = self.vali(test_data, test_loader, criterion)
+
+            print("Epoch: {0}, Steps: {1} | Train Loss: {2:.7f} Vali Loss: {3:.7f} Test Loss: {4:.7f}".format(
+                epoch + 1, train_steps, train_loss, vali_loss, test_loss))
+            early_stopping(vali_loss, self.model, path)
+            if early_stopping.early_stop:
+                print("Early stopping")
+                break
+            adjust_learning_rate(model_optim, epoch + 1, self.args)
+
+        best_model_path = path + '/' + 'checkpoint.pth'
+        self.model.load_state_dict(torch.load(best_model_path))
+
+        return self.model
+
+    def test(self, setting, test=0):
+        test_data, test_loader = self._get_data(flag='test')
+        train_data, train_loader = self._get_data(flag='train')
+        if test:
+            print('loading model')
+            self.model.load_state_dict(torch.load(os.path.join('./checkpoints/' + setting, 'checkpoint.pth')))
+
+        attens_energy = []
+        folder_path = './test_results/' + setting + '/'
+        if not os.path.exists(folder_path):
+            os.makedirs(folder_path)
+
+        self.model.eval()
+        self.anomaly_criterion = nn.MSELoss(reduce=False)
+
+        # (1) stastic on the train set
+        with torch.no_grad():
+            for i, (batch_x, batch_y) in enumerate(train_loader):
+                batch_x = batch_x.float().to(self.device)
+                # reconstruction
+                outputs = self.model(batch_x, None, None, None)
+                # criterion
+                score = torch.mean(self.anomaly_criterion(batch_x, outputs), dim=-1)
+                score = score.detach().cpu().numpy()
+                attens_energy.append(score)
+
+        attens_energy = np.concatenate(attens_energy, axis=0).reshape(-1)
+        train_energy = np.array(attens_energy)
+
+        # (2) find the threshold
+        attens_energy = []
+        test_labels = []
+        for i, (batch_x, batch_y) in enumerate(test_loader):
+            batch_x = batch_x.float().to(self.device)
+            # reconstruction
+            outputs = self.model(batch_x, None, None, None)
+            # criterion
+            score = torch.mean(self.anomaly_criterion(batch_x, outputs), dim=-1)
+            score = score.detach().cpu().numpy()
+            attens_energy.append(score)
+            test_labels.append(batch_y)
+
+        attens_energy = np.concatenate(attens_energy, axis=0).reshape(-1)
+        test_energy = np.array(attens_energy)
+        combined_energy = np.concatenate([train_energy, test_energy], axis=0)
+        threshold = np.percentile(combined_energy, 100 - self.args.anomaly_ratio)
+        print("Threshold :", threshold)
+
+        # (3) evaluation on the test set
+        pred = (test_energy > threshold).astype(int)
+        test_labels = np.concatenate(test_labels, axis=0).reshape(-1)
+        test_labels = np.array(test_labels)
+        gt = test_labels.astype(int)
+
+        print("pred:   ", pred.shape)
+        print("gt:     ", gt.shape)
+
+        # (4) detection adjustment
+        gt, pred = adjustment(gt, pred)
+
+        pred = np.array(pred)
+        gt = np.array(gt)
+        print("pred: ", pred.shape)
+        print("gt:   ", gt.shape)
+
+        accuracy = accuracy_score(gt, pred)
+        precision, recall, f_score, support = precision_recall_fscore_support(gt, pred, average='binary')
+        print("Accuracy : {:0.4f}, Precision : {:0.4f}, Recall : {:0.4f}, F-score : {:0.4f} ".format(
+            accuracy, precision,
+            recall, f_score))
+
+        f = open("result_anomaly_detection.txt", 'a')
+        f.write(setting + "  \n")
+        f.write("Accuracy : {:0.4f}, Precision : {:0.4f}, Recall : {:0.4f}, F-score : {:0.4f} ".format(
+            accuracy, precision,
+            recall, f_score))
+        f.write('\n')
+        f.write('\n')
+        f.close()
+        return

exp/exp_basic.py

@@ -0,0 +1,80 @@
+import os
+import torch
+from models import Autoformer, Transformer, TimesNet, Nonstationary_Transformer, DLinear, FEDformer, \
+    Informer, LightTS, Reformer, ETSformer, Pyraformer, PatchTST, MICN, Crossformer, FiLM, iTransformer, \
+    Koopa, TiDE, FreTS, TimeMixer, TSMixer, SegRNN, MambaSimple, TemporalFusionTransformer, SCINet, PAttn, TimeXer, \
+    WPMixer, MultiPatchFormer, xPatch_SparseChannel
+
+
+class Exp_Basic(object):
+    def __init__(self, args):
+        self.args = args
+        self.model_dict = {
+            'TimesNet': TimesNet,
+            'Autoformer': Autoformer,
+            'Transformer': Transformer,
+            'Nonstationary_Transformer': Nonstationary_Transformer,
+            'DLinear': DLinear,
+            'FEDformer': FEDformer,
+            'Informer': Informer,
+            'LightTS': LightTS,
+            'Reformer': Reformer,
+            'ETSformer': ETSformer,
+            'PatchTST': PatchTST,
+            'Pyraformer': Pyraformer,
+            'MICN': MICN,
+            'Crossformer': Crossformer,
+            'FiLM': FiLM,
+            'iTransformer': iTransformer,
+            'Koopa': Koopa,
+            'TiDE': TiDE,
+            'FreTS': FreTS,
+            'MambaSimple': MambaSimple,
+            'TimeMixer': TimeMixer,
+            'TSMixer': TSMixer,
+            'SegRNN': SegRNN,
+            'TemporalFusionTransformer': TemporalFusionTransformer,
+            "SCINet": SCINet,
+            'PAttn': PAttn,
+            'TimeXer': TimeXer,
+            'WPMixer': WPMixer,
+            'MultiPatchFormer': MultiPatchFormer,
+            'xPatch_SparseChannel': xPatch_SparseChannel
+        }
+        if args.model == 'Mamba':
+            print('Please make sure you have successfully installed mamba_ssm')
+            from models import Mamba
+            self.model_dict['Mamba'] = Mamba
+
+        self.device = self._acquire_device()
+        self.model = self._build_model().to(self.device)
+
+    def _build_model(self):
+        raise NotImplementedError
+        return None
+
+    def _acquire_device(self):
+        if self.args.use_gpu and self.args.gpu_type == 'cuda':
+            os.environ["CUDA_VISIBLE_DEVICES"] = str(
+                self.args.gpu) if not self.args.use_multi_gpu else self.args.devices
+            device = torch.device('cuda:{}'.format(self.args.gpu))
+            print('Use GPU: cuda:{}'.format(self.args.gpu))
+        elif self.args.use_gpu and self.args.gpu_type == 'mps':
+            device = torch.device('mps')
+            print('Use GPU: mps')
+        else:
+            device = torch.device('cpu')
+            print('Use CPU')
+        return device
+
+    def _get_data(self):
+        pass
+
+    def vali(self):
+        pass
+
+    def train(self):
+        pass
+
+    def test(self):
+        pass

exp/exp_classification.py

@@ -0,0 +1,192 @@
+from data_provider.data_factory import data_provider
+from exp.exp_basic import Exp_Basic
+from utils.tools import EarlyStopping, adjust_learning_rate, cal_accuracy
+import torch
+import torch.nn as nn
+from torch import optim
+import os
+import time
+import warnings
+import numpy as np
+import pdb
+
+warnings.filterwarnings('ignore')
+
+
+class Exp_Classification(Exp_Basic):
+    def __init__(self, args):
+        super(Exp_Classification, self).__init__(args)
+
+    def _build_model(self):
+        # model input depends on data
+        train_data, train_loader = self._get_data(flag='TRAIN')
+        test_data, test_loader = self._get_data(flag='TEST')
+        self.args.seq_len = max(train_data.max_seq_len, test_data.max_seq_len)
+        self.args.pred_len = 96
+        self.args.enc_in = train_data.feature_df.shape[1]
+        self.args.num_class = len(train_data.class_names)
+        # model init
+        model = self.model_dict[self.args.model].Model(self.args).float()
+        if self.args.use_multi_gpu and self.args.use_gpu:
+            model = nn.DataParallel(model, device_ids=self.args.device_ids)
+        return model
+
+    def _get_data(self, flag):
+        data_set, data_loader = data_provider(self.args, flag)
+        return data_set, data_loader
+
+    def _select_optimizer(self):
+        # model_optim = optim.Adam(self.model.parameters(), lr=self.args.learning_rate)
+        model_optim = optim.RAdam(self.model.parameters(), lr=self.args.learning_rate)
+        return model_optim
+
+    def _select_criterion(self):
+        criterion = nn.CrossEntropyLoss()
+        return criterion
+
+    def vali(self, vali_data, vali_loader, criterion):
+        total_loss = []
+        preds = []
+        trues = []
+        self.model.eval()
+        with torch.no_grad():
+            for i, (batch_x, label, padding_mask) in enumerate(vali_loader):
+                batch_x = batch_x.float().to(self.device)
+                padding_mask = padding_mask.float().to(self.device)
+                label = label.to(self.device)
+
+                outputs = self.model(batch_x, padding_mask, None, None)
+
+                pred = outputs.detach()
+                loss = criterion(pred, label.long().squeeze())
+                total_loss.append(loss.item())
+
+                preds.append(outputs.detach())
+                trues.append(label)
+
+        total_loss = np.average(total_loss)
+
+        preds = torch.cat(preds, 0)
+        trues = torch.cat(trues, 0)
+        probs = torch.nn.functional.softmax(preds)  # (total_samples, num_classes) est. prob. for each class and sample
+        predictions = torch.argmax(probs, dim=1).cpu().numpy()  # (total_samples,) int class index for each sample
+        trues = trues.flatten().cpu().numpy()
+        accuracy = cal_accuracy(predictions, trues)
+
+        self.model.train()
+        return total_loss, accuracy
+
+    def train(self, setting):
+        train_data, train_loader = self._get_data(flag='TRAIN')
+        vali_data, vali_loader = self._get_data(flag='TEST')
+        test_data, test_loader = self._get_data(flag='TEST')
+
+        path = os.path.join(self.args.checkpoints, setting)
+        if not os.path.exists(path):
+            os.makedirs(path)
+
+        time_now = time.time()
+
+        train_steps = len(train_loader)
+        early_stopping = EarlyStopping(patience=self.args.patience, verbose=True)
+
+        model_optim = self._select_optimizer()
+        criterion = self._select_criterion()
+
+        for epoch in range(self.args.train_epochs):
+            iter_count = 0
+            train_loss = []
+
+            self.model.train()
+            epoch_time = time.time()
+
+            for i, (batch_x, label, padding_mask) in enumerate(train_loader):
+                iter_count += 1
+                model_optim.zero_grad()
+
+                batch_x = batch_x.float().to(self.device)
+                padding_mask = padding_mask.float().to(self.device)
+                label = label.to(self.device)
+
+                outputs = self.model(batch_x, padding_mask, None, None)
+                loss = criterion(outputs, label.long().squeeze(-1))
+                train_loss.append(loss.item())
+
+                if (i + 1) % 100 == 0:
+                    print("\titers: {0}, epoch: {1} | loss: {2:.7f}".format(i + 1, epoch + 1, loss.item()))
+                    speed = (time.time() - time_now) / iter_count
+                    left_time = speed * ((self.args.train_epochs - epoch) * train_steps - i)
+                    print('\tspeed: {:.4f}s/iter; left time: {:.4f}s'.format(speed, left_time))
+                    iter_count = 0
+                    time_now = time.time()
+
+                loss.backward()
+                nn.utils.clip_grad_norm_(self.model.parameters(), max_norm=4.0)
+                model_optim.step()
+
+            print("Epoch: {} cost time: {}".format(epoch + 1, time.time() - epoch_time))
+            train_loss = np.average(train_loss)
+            vali_loss, val_accuracy = self.vali(vali_data, vali_loader, criterion)
+            # test_loss, test_accuracy = self.vali(test_data, test_loader, criterion)
+
+            print(
+                "Epoch: {0}, Steps: {1} | Train Loss: {2:.3f} Vali Loss: {3:.3f} Vali Acc: {4:.3f}" # Test Loss: {5:.3f} Test Acc: {6:.3f}"
+                .format(epoch + 1, train_steps, train_loss, vali_loss, val_accuracy))
+            # test_loss, test_accuracy))
+            early_stopping(-val_accuracy, self.model, path)
+            if early_stopping.early_stop:
+                print("Early stopping")
+                break
+
+        best_model_path = path + '/' + 'checkpoint.pth'
+        self.model.load_state_dict(torch.load(best_model_path))
+
+        return self.model
+
+    def test(self, setting, test=0):
+        test_data, test_loader = self._get_data(flag='TEST')
+        if test:
+            print('loading model')
+            self.model.load_state_dict(torch.load(os.path.join('./checkpoints/' + setting, 'checkpoint.pth')))
+
+        preds = []
+        trues = []
+        folder_path = './test_results/' + setting + '/'
+        if not os.path.exists(folder_path):
+            os.makedirs(folder_path)
+
+        self.model.eval()
+        with torch.no_grad():
+            for i, (batch_x, label, padding_mask) in enumerate(test_loader):
+                batch_x = batch_x.float().to(self.device)
+                padding_mask = padding_mask.float().to(self.device)
+                label = label.to(self.device)
+
+                outputs = self.model(batch_x, padding_mask, None, None)
+
+                preds.append(outputs.detach())
+                trues.append(label)
+
+        preds = torch.cat(preds, 0)
+        trues = torch.cat(trues, 0)
+        print('test shape:', preds.shape, trues.shape)
+
+        probs = torch.nn.functional.softmax(preds)  # (total_samples, num_classes) est. prob. for each class and sample
+        predictions = torch.argmax(probs, dim=1).cpu().numpy()  # (total_samples,) int class index for each sample
+        trues = trues.flatten().cpu().numpy()
+        accuracy = cal_accuracy(predictions, trues)
+
+        # result save
+        folder_path = './results/' + setting + '/'
+        if not os.path.exists(folder_path):
+            os.makedirs(folder_path)
+
+        print('accuracy:{}'.format(accuracy))
+        file_name='result_classification.txt'
+        f = open(os.path.join(folder_path,file_name), 'a')
+        f.write(setting + "  \n")
+        f.write('accuracy:{}'.format(accuracy))
+        f.write('\n')
+        f.write('\n')
+        f.close()
+        return

exp/exp_imputation.py

@@ -0,0 +1,228 @@
+from data_provider.data_factory import data_provider
+from exp.exp_basic import Exp_Basic
+from utils.tools import EarlyStopping, adjust_learning_rate, visual
+from utils.metrics import metric
+import torch
+import torch.nn as nn
+from torch import optim
+import os
+import time
+import warnings
+import numpy as np
+
+warnings.filterwarnings('ignore')
+
+
+class Exp_Imputation(Exp_Basic):
+    def __init__(self, args):
+        super(Exp_Imputation, self).__init__(args)
+
+    def _build_model(self):
+        model = self.model_dict[self.args.model].Model(self.args).float()
+
+        if self.args.use_multi_gpu and self.args.use_gpu:
+            model = nn.DataParallel(model, device_ids=self.args.device_ids)
+        return model
+
+    def _get_data(self, flag):
+        data_set, data_loader = data_provider(self.args, flag)
+        return data_set, data_loader
+
+    def _select_optimizer(self):
+        model_optim = optim.Adam(self.model.parameters(), lr=self.args.learning_rate)
+        return model_optim
+
+    def _select_criterion(self):
+        criterion = nn.MSELoss()
+        return criterion
+
+    def vali(self, vali_data, vali_loader, criterion):
+        total_loss = []
+        self.model.eval()
+        with torch.no_grad():
+            for i, (batch_x, batch_y, batch_x_mark, batch_y_mark) in enumerate(vali_loader):
+                batch_x = batch_x.float().to(self.device)
+                batch_x_mark = batch_x_mark.float().to(self.device)
+
+                # random mask
+                B, T, N = batch_x.shape
+                """
+                B = batch size
+                T = seq len
+                N = number of features
+                """
+                mask = torch.rand((B, T, N)).to(self.device)
+                mask[mask <= self.args.mask_rate] = 0  # masked
+                mask[mask > self.args.mask_rate] = 1  # remained
+                inp = batch_x.masked_fill(mask == 0, 0)
+
+                outputs = self.model(inp, batch_x_mark, None, None, mask)
+
+                f_dim = -1 if self.args.features == 'MS' else 0
+                outputs = outputs[:, :, f_dim:]
+
+                # add support for MS
+                batch_x = batch_x[:, :, f_dim:]
+                mask = mask[:, :, f_dim:]
+
+                pred = outputs.detach()
+                true = batch_x.detach()
+                mask = mask.detach()
+
+                loss = criterion(pred[mask == 0], true[mask == 0])
+                total_loss.append(loss.item())
+        total_loss = np.average(total_loss)
+        self.model.train()
+        return total_loss
+
+    def train(self, setting):
+        train_data, train_loader = self._get_data(flag='train')
+        vali_data, vali_loader = self._get_data(flag='val')
+        test_data, test_loader = self._get_data(flag='test')
+
+        path = os.path.join(self.args.checkpoints, setting)
+        if not os.path.exists(path):
+            os.makedirs(path)
+
+        time_now = time.time()
+
+        train_steps = len(train_loader)
+        early_stopping = EarlyStopping(patience=self.args.patience, verbose=True)
+
+        model_optim = self._select_optimizer()
+        criterion = self._select_criterion()
+
+        for epoch in range(self.args.train_epochs):
+            iter_count = 0
+            train_loss = []
+
+            self.model.train()
+            epoch_time = time.time()
+            for i, (batch_x, batch_y, batch_x_mark, batch_y_mark) in enumerate(train_loader):
+                iter_count += 1
+                model_optim.zero_grad()
+
+                batch_x = batch_x.float().to(self.device)
+                batch_x_mark = batch_x_mark.float().to(self.device)
+
+                # random mask
+                B, T, N = batch_x.shape
+                mask = torch.rand((B, T, N)).to(self.device)
+                mask[mask <= self.args.mask_rate] = 0  # masked
+                mask[mask > self.args.mask_rate] = 1  # remained
+                inp = batch_x.masked_fill(mask == 0, 0)
+
+                outputs = self.model(inp, batch_x_mark, None, None, mask)
+
+                f_dim = -1 if self.args.features == 'MS' else 0
+                outputs = outputs[:, :, f_dim:]
+
+                # add support for MS
+                batch_x = batch_x[:, :, f_dim:]
+                mask = mask[:, :, f_dim:]
+
+                loss = criterion(outputs[mask == 0], batch_x[mask == 0])
+                train_loss.append(loss.item())
+
+                if (i + 1) % 100 == 0:
+                    print("\titers: {0}, epoch: {1} | loss: {2:.7f}".format(i + 1, epoch + 1, loss.item()))
+                    speed = (time.time() - time_now) / iter_count
+                    left_time = speed * ((self.args.train_epochs - epoch) * train_steps - i)
+                    print('\tspeed: {:.4f}s/iter; left time: {:.4f}s'.format(speed, left_time))
+                    iter_count = 0
+                    time_now = time.time()
+
+                loss.backward()
+                model_optim.step()
+
+            print("Epoch: {} cost time: {}".format(epoch + 1, time.time() - epoch_time))
+            train_loss = np.average(train_loss)
+            vali_loss = self.vali(vali_data, vali_loader, criterion)
+            test_loss = self.vali(test_data, test_loader, criterion)
+
+            print("Epoch: {0}, Steps: {1} | Train Loss: {2:.7f} Vali Loss: {3:.7f} Test Loss: {4:.7f}".format(
+                epoch + 1, train_steps, train_loss, vali_loss, test_loss))
+            early_stopping(vali_loss, self.model, path)
+            if early_stopping.early_stop:
+                print("Early stopping")
+                break
+            adjust_learning_rate(model_optim, epoch + 1, self.args)
+
+        best_model_path = path + '/' + 'checkpoint.pth'
+        self.model.load_state_dict(torch.load(best_model_path))
+
+        return self.model
+
+    def test(self, setting, test=0):
+        test_data, test_loader = self._get_data(flag='test')
+        if test:
+            print('loading model')
+            self.model.load_state_dict(torch.load(os.path.join('./checkpoints/' + setting, 'checkpoint.pth')))
+
+        preds = []
+        trues = []
+        masks = []
+        folder_path = './test_results/' + setting + '/'
+        if not os.path.exists(folder_path):
+            os.makedirs(folder_path)
+
+        self.model.eval()
+        with torch.no_grad():
+            for i, (batch_x, batch_y, batch_x_mark, batch_y_mark) in enumerate(test_loader):
+                batch_x = batch_x.float().to(self.device)
+                batch_x_mark = batch_x_mark.float().to(self.device)
+
+                # random mask
+                B, T, N = batch_x.shape
+                mask = torch.rand((B, T, N)).to(self.device)
+                mask[mask <= self.args.mask_rate] = 0  # masked
+                mask[mask > self.args.mask_rate] = 1  # remained
+                inp = batch_x.masked_fill(mask == 0, 0)
+
+                # imputation
+                outputs = self.model(inp, batch_x_mark, None, None, mask)
+
+                # eval
+                f_dim = -1 if self.args.features == 'MS' else 0
+                outputs = outputs[:, :, f_dim:]
+
+                # add support for MS 
+                batch_x = batch_x[:, :, f_dim:]
+                mask = mask[:, :, f_dim:]
+
+                outputs = outputs.detach().cpu().numpy()
+                pred = outputs
+                true = batch_x.detach().cpu().numpy()
+                preds.append(pred)
+                trues.append(true)
+                masks.append(mask.detach().cpu())
+
+                if i % 20 == 0:
+                    filled = true[0, :, -1].copy()
+                    filled = filled * mask[0, :, -1].detach().cpu().numpy() + \
+                             pred[0, :, -1] * (1 - mask[0, :, -1].detach().cpu().numpy())
+                    visual(true[0, :, -1], filled, os.path.join(folder_path, str(i) + '.pdf'))
+
+        preds = np.concatenate(preds, 0)
+        trues = np.concatenate(trues, 0)
+        masks = np.concatenate(masks, 0)
+        print('test shape:', preds.shape, trues.shape)
+
+        # result save
+        folder_path = './results/' + setting + '/'
+        if not os.path.exists(folder_path):
+            os.makedirs(folder_path)
+
+        mae, mse, rmse, mape, mspe = metric(preds[masks == 0], trues[masks == 0])
+        print('mse:{}, mae:{}'.format(mse, mae))
+        f = open("result_imputation.txt", 'a')
+        f.write(setting + "  \n")
+        f.write('mse:{}, mae:{}'.format(mse, mae))
+        f.write('\n')
+        f.write('\n')
+        f.close()
+
+        np.save(folder_path + 'metrics.npy', np.array([mae, mse, rmse, mape, mspe]))
+        np.save(folder_path + 'pred.npy', preds)
+        np.save(folder_path + 'true.npy', trues)
+        return

exp/exp_long_term_forecasting.py

@@ -0,0 +1,268 @@
+from data_provider.data_factory import data_provider
+from exp.exp_basic import Exp_Basic
+from utils.tools import EarlyStopping, adjust_learning_rate, visual
+from utils.metrics import metric
+import torch
+import torch.nn as nn
+from torch import optim
+import os
+import time
+import warnings
+import numpy as np
+from utils.dtw_metric import dtw, accelerated_dtw
+from utils.augmentation import run_augmentation, run_augmentation_single
+
+warnings.filterwarnings('ignore')
+
+
+class Exp_Long_Term_Forecast(Exp_Basic):
+    def __init__(self, args):
+        super(Exp_Long_Term_Forecast, self).__init__(args)
+
+    def _build_model(self):
+        model = self.model_dict[self.args.model].Model(self.args).float()
+
+        if self.args.use_multi_gpu and self.args.use_gpu:
+            model = nn.DataParallel(model, device_ids=self.args.device_ids)
+        return model
+
+    def _get_data(self, flag):
+        data_set, data_loader = data_provider(self.args, flag)
+        return data_set, data_loader
+
+    def _select_optimizer(self):
+        model_optim = optim.Adam(self.model.parameters(), lr=self.args.learning_rate)
+        return model_optim
+
+    def _select_criterion(self):
+        criterion = nn.MSELoss()
+        return criterion
+ 
+
+    def vali(self, vali_data, vali_loader, criterion):
+        total_loss = []
+        self.model.eval()
+        with torch.no_grad():
+            for i, (batch_x, batch_y, batch_x_mark, batch_y_mark) in enumerate(vali_loader):
+                batch_x = batch_x.float().to(self.device)
+                batch_y = batch_y.float()
+
+                batch_x_mark = batch_x_mark.float().to(self.device)
+                batch_y_mark = batch_y_mark.float().to(self.device)
+
+                # decoder input
+                dec_inp = torch.zeros_like(batch_y[:, -self.args.pred_len:, :]).float()
+                dec_inp = torch.cat([batch_y[:, :self.args.label_len, :], dec_inp], dim=1).float().to(self.device)
+                # encoder - decoder
+                if self.args.use_amp:
+                    with torch.cuda.amp.autocast():
+                        outputs = self.model(batch_x, batch_x_mark, dec_inp, batch_y_mark)
+                else:
+                    outputs = self.model(batch_x, batch_x_mark, dec_inp, batch_y_mark)
+                f_dim = -1 if self.args.features == 'MS' else 0
+                outputs = outputs[:, -self.args.pred_len:, f_dim:]
+                batch_y = batch_y[:, -self.args.pred_len:, f_dim:].to(self.device)
+
+                pred = outputs.detach()
+                true = batch_y.detach()
+
+                loss = criterion(pred, true)
+
+                total_loss.append(loss.item())
+        total_loss = np.average(total_loss)
+        self.model.train()
+        return total_loss
+
+    def train(self, setting):
+        train_data, train_loader = self._get_data(flag='train')
+        vali_data, vali_loader = self._get_data(flag='val')
+        test_data, test_loader = self._get_data(flag='test')
+
+        path = os.path.join(self.args.checkpoints, setting)
+        if not os.path.exists(path):
+            os.makedirs(path)
+
+        time_now = time.time()
+
+        train_steps = len(train_loader)
+        early_stopping = EarlyStopping(patience=self.args.patience, verbose=True)
+
+        model_optim = self._select_optimizer()
+        criterion = self._select_criterion()
+
+        if self.args.use_amp:
+            scaler = torch.cuda.amp.GradScaler()
+
+        for epoch in range(self.args.train_epochs):
+            iter_count = 0
+            train_loss = []
+
+            self.model.train()
+            epoch_time = time.time()
+            for i, (batch_x, batch_y, batch_x_mark, batch_y_mark) in enumerate(train_loader):
+                iter_count += 1
+                model_optim.zero_grad()
+                batch_x = batch_x.float().to(self.device)
+                batch_y = batch_y.float().to(self.device)
+                batch_x_mark = batch_x_mark.float().to(self.device)
+                batch_y_mark = batch_y_mark.float().to(self.device)
+
+                # decoder input
+                dec_inp = torch.zeros_like(batch_y[:, -self.args.pred_len:, :]).float()
+                dec_inp = torch.cat([batch_y[:, :self.args.label_len, :], dec_inp], dim=1).float().to(self.device)
+
+                # encoder - decoder
+                if self.args.use_amp:
+                    with torch.cuda.amp.autocast():
+                        outputs = self.model(batch_x, batch_x_mark, dec_inp, batch_y_mark)
+
+                        f_dim = -1 if self.args.features == 'MS' else 0
+                        outputs = outputs[:, -self.args.pred_len:, f_dim:]
+                        batch_y = batch_y[:, -self.args.pred_len:, f_dim:].to(self.device)
+                        loss = criterion(outputs, batch_y)
+                        train_loss.append(loss.item())
+                else:
+                    outputs = self.model(batch_x, batch_x_mark, dec_inp, batch_y_mark)
+
+                    f_dim = -1 if self.args.features == 'MS' else 0
+                    outputs = outputs[:, -self.args.pred_len:, f_dim:]
+                    batch_y = batch_y[:, -self.args.pred_len:, f_dim:].to(self.device)
+                    loss = criterion(outputs, batch_y)
+                    train_loss.append(loss.item())
+
+                if (i + 1) % 100 == 0:
+                    print("\titers: {0}, epoch: {1} | loss: {2:.7f}".format(i + 1, epoch + 1, loss.item()))
+                    speed = (time.time() - time_now) / iter_count
+                    left_time = speed * ((self.args.train_epochs - epoch) * train_steps - i)
+                    print('\tspeed: {:.4f}s/iter; left time: {:.4f}s'.format(speed, left_time))
+                    iter_count = 0
+                    time_now = time.time()
+
+                if self.args.use_amp:
+                    scaler.scale(loss).backward()
+                    scaler.step(model_optim)
+                    scaler.update()
+                else:
+                    loss.backward()
+                    model_optim.step()
+
+            print("Epoch: {} cost time: {}".format(epoch + 1, time.time() - epoch_time))
+            train_loss = np.average(train_loss)
+            vali_loss = self.vali(vali_data, vali_loader, criterion)
+            test_loss = self.vali(test_data, test_loader, criterion)
+
+            print("Epoch: {0}, Steps: {1} | Train Loss: {2:.7f} Vali Loss: {3:.7f} Test Loss: {4:.7f}".format(
+                epoch + 1, train_steps, train_loss, vali_loss, test_loss))
+            early_stopping(vali_loss, self.model, path)
+            if early_stopping.early_stop:
+                print("Early stopping")
+                break
+
+            adjust_learning_rate(model_optim, epoch + 1, self.args)
+
+        best_model_path = path + '/' + 'checkpoint.pth'
+        self.model.load_state_dict(torch.load(best_model_path))
+
+        return self.model
+
+    def test(self, setting, test=0):
+        test_data, test_loader = self._get_data(flag='test')
+        if test:
+            print('loading model')
+            self.model.load_state_dict(torch.load(os.path.join('./checkpoints/' + setting, 'checkpoint.pth')))
+
+        preds = []
+        trues = []
+        folder_path = './test_results/' + setting + '/'
+        if not os.path.exists(folder_path):
+            os.makedirs(folder_path)
+
+        self.model.eval()
+        with torch.no_grad():
+            for i, (batch_x, batch_y, batch_x_mark, batch_y_mark) in enumerate(test_loader):
+                batch_x = batch_x.float().to(self.device)
+                batch_y = batch_y.float().to(self.device)
+
+                batch_x_mark = batch_x_mark.float().to(self.device)
+                batch_y_mark = batch_y_mark.float().to(self.device)
+
+                # decoder input
+                dec_inp = torch.zeros_like(batch_y[:, -self.args.pred_len:, :]).float()
+                dec_inp = torch.cat([batch_y[:, :self.args.label_len, :], dec_inp], dim=1).float().to(self.device)
+                # encoder - decoder
+                if self.args.use_amp:
+                    with torch.cuda.amp.autocast():
+                        outputs = self.model(batch_x, batch_x_mark, dec_inp, batch_y_mark)
+                else:
+                    outputs = self.model(batch_x, batch_x_mark, dec_inp, batch_y_mark)
+
+                f_dim = -1 if self.args.features == 'MS' else 0
+                outputs = outputs[:, -self.args.pred_len:, :]
+                batch_y = batch_y[:, -self.args.pred_len:, :].to(self.device)
+                outputs = outputs.detach().cpu().numpy()
+                batch_y = batch_y.detach().cpu().numpy()
+                if test_data.scale and self.args.inverse:
+                    shape = batch_y.shape
+                    if outputs.shape[-1] != batch_y.shape[-1]:
+                        outputs = np.tile(outputs, [1, 1, int(batch_y.shape[-1] / outputs.shape[-1])])
+                    outputs = test_data.inverse_transform(outputs.reshape(shape[0] * shape[1], -1)).reshape(shape)
+                    batch_y = test_data.inverse_transform(batch_y.reshape(shape[0] * shape[1], -1)).reshape(shape)
+
+                outputs = outputs[:, :, f_dim:]
+                batch_y = batch_y[:, :, f_dim:]
+
+                pred = outputs
+                true = batch_y
+
+                preds.append(pred)
+                trues.append(true)
+                if i % 20 == 0:
+                    input = batch_x.detach().cpu().numpy()
+                    if test_data.scale and self.args.inverse:
+                        shape = input.shape
+                        input = test_data.inverse_transform(input.reshape(shape[0] * shape[1], -1)).reshape(shape)
+                    gt = np.concatenate((input[0, :, -1], true[0, :, -1]), axis=0)
+                    pd = np.concatenate((input[0, :, -1], pred[0, :, -1]), axis=0)
+                    visual(gt, pd, os.path.join(folder_path, str(i) + '.pdf'))
+
+        preds = np.concatenate(preds, axis=0)
+        trues = np.concatenate(trues, axis=0)
+        print('test shape:', preds.shape, trues.shape)
+        preds = preds.reshape(-1, preds.shape[-2], preds.shape[-1])
+        trues = trues.reshape(-1, trues.shape[-2], trues.shape[-1])
+        print('test shape:', preds.shape, trues.shape)
+
+        # result save
+        folder_path = './results/' + setting + '/'
+        if not os.path.exists(folder_path):
+            os.makedirs(folder_path)
+
+        # dtw calculation
+        if self.args.use_dtw:
+            dtw_list = []
+            manhattan_distance = lambda x, y: np.abs(x - y)
+            for i in range(preds.shape[0]):
+                x = preds[i].reshape(-1, 1)
+                y = trues[i].reshape(-1, 1)
+                if i % 100 == 0:
+                    print("calculating dtw iter:", i)
+                d, _, _, _ = accelerated_dtw(x, y, dist=manhattan_distance)
+                dtw_list.append(d)
+            dtw = np.array(dtw_list).mean()
+        else:
+            dtw = 'Not calculated'
+
+        mae, mse, rmse, mape, mspe = metric(preds, trues)
+        print('mse:{}, mae:{}, dtw:{}'.format(mse, mae, dtw))
+        f = open("result_long_term_forecast.txt", 'a')
+        f.write(setting + "  \n")
+        f.write('mse:{}, mae:{}, dtw:{}'.format(mse, mae, dtw))
+        f.write('\n')
+        f.write('\n')
+        f.close()
+
+        np.save(folder_path + 'metrics.npy', np.array([mae, mse, rmse, mape, mspe]))
+        np.save(folder_path + 'pred.npy', preds)
+        np.save(folder_path + 'true.npy', trues)
+
+        return

exp/exp_short_term_forecasting.py

@@ -0,0 +1,235 @@
+from data_provider.data_factory import data_provider
+from data_provider.m4 import M4Meta
+from exp.exp_basic import Exp_Basic
+from utils.tools import EarlyStopping, adjust_learning_rate, visual
+from utils.losses import mape_loss, mase_loss, smape_loss
+from utils.m4_summary import M4Summary
+import torch
+import torch.nn as nn
+from torch import optim
+import os
+import time
+import warnings
+import numpy as np
+import pandas
+
+warnings.filterwarnings('ignore')
+
+
+class Exp_Short_Term_Forecast(Exp_Basic):
+    def __init__(self, args):
+        super(Exp_Short_Term_Forecast, self).__init__(args)
+
+    def _build_model(self):
+        if self.args.data == 'm4':
+            self.args.pred_len = M4Meta.horizons_map[self.args.seasonal_patterns]  # Up to M4 config
+            self.args.seq_len = 2 * self.args.pred_len  # input_len = 2*pred_len
+            self.args.label_len = self.args.pred_len
+            self.args.frequency_map = M4Meta.frequency_map[self.args.seasonal_patterns]
+        model = self.model_dict[self.args.model].Model(self.args).float()
+
+        if self.args.use_multi_gpu and self.args.use_gpu:
+            model = nn.DataParallel(model, device_ids=self.args.device_ids)
+        return model
+
+    def _get_data(self, flag):
+        data_set, data_loader = data_provider(self.args, flag)
+        return data_set, data_loader
+
+    def _select_optimizer(self):
+        model_optim = optim.Adam(self.model.parameters(), lr=self.args.learning_rate)
+        return model_optim
+
+    def _select_criterion(self, loss_name='MSE'):
+        if loss_name == 'MSE':
+            return nn.MSELoss()
+        elif loss_name == 'MAPE':
+            return mape_loss()
+        elif loss_name == 'MASE':
+            return mase_loss()
+        elif loss_name == 'SMAPE':
+            return smape_loss()
+
+    def train(self, setting):
+        train_data, train_loader = self._get_data(flag='train')
+        vali_data, vali_loader = self._get_data(flag='val')
+
+        path = os.path.join(self.args.checkpoints, setting)
+        if not os.path.exists(path):
+            os.makedirs(path)
+
+        time_now = time.time()
+
+        train_steps = len(train_loader)
+        early_stopping = EarlyStopping(patience=self.args.patience, verbose=True)
+
+        model_optim = self._select_optimizer()
+        criterion = self._select_criterion(self.args.loss)
+        mse = nn.MSELoss()
+
+        for epoch in range(self.args.train_epochs):
+            iter_count = 0
+            train_loss = []
+
+            self.model.train()
+            epoch_time = time.time()
+            for i, (batch_x, batch_y, batch_x_mark, batch_y_mark) in enumerate(train_loader):
+                iter_count += 1
+                model_optim.zero_grad()
+                batch_x = batch_x.float().to(self.device)
+
+                batch_y = batch_y.float().to(self.device)
+                batch_y_mark = batch_y_mark.float().to(self.device)
+
+                # decoder input
+                dec_inp = torch.zeros_like(batch_y[:, -self.args.pred_len:, :]).float()
+                dec_inp = torch.cat([batch_y[:, :self.args.label_len, :], dec_inp], dim=1).float().to(self.device)
+
+                outputs = self.model(batch_x, None, dec_inp, None)
+
+                f_dim = -1 if self.args.features == 'MS' else 0
+                outputs = outputs[:, -self.args.pred_len:, f_dim:]
+                batch_y = batch_y[:, -self.args.pred_len:, f_dim:].to(self.device)
+
+                batch_y_mark = batch_y_mark[:, -self.args.pred_len:, f_dim:].to(self.device)
+                loss_value = criterion(batch_x, self.args.frequency_map, outputs, batch_y, batch_y_mark)
+                loss_sharpness = mse((outputs[:, 1:, :] - outputs[:, :-1, :]), (batch_y[:, 1:, :] - batch_y[:, :-1, :]))
+                loss = loss_value  # + loss_sharpness * 1e-5
+                train_loss.append(loss.item())
+
+                if (i + 1) % 100 == 0:
+                    print("\titers: {0}, epoch: {1} | loss: {2:.7f}".format(i + 1, epoch + 1, loss.item()))
+                    speed = (time.time() - time_now) / iter_count
+                    left_time = speed * ((self.args.train_epochs - epoch) * train_steps - i)
+                    print('\tspeed: {:.4f}s/iter; left time: {:.4f}s'.format(speed, left_time))
+                    iter_count = 0
+                    time_now = time.time()
+
+                loss.backward()
+                model_optim.step()
+
+            print("Epoch: {} cost time: {}".format(epoch + 1, time.time() - epoch_time))
+            train_loss = np.average(train_loss)
+            vali_loss = self.vali(train_loader, vali_loader, criterion)
+            test_loss = vali_loss
+            print("Epoch: {0}, Steps: {1} | Train Loss: {2:.7f} Vali Loss: {3:.7f} Test Loss: {4:.7f}".format(
+                epoch + 1, train_steps, train_loss, vali_loss, test_loss))
+            early_stopping(vali_loss, self.model, path)
+            if early_stopping.early_stop:
+                print("Early stopping")
+                break
+
+            adjust_learning_rate(model_optim, epoch + 1, self.args)
+
+        best_model_path = path + '/' + 'checkpoint.pth'
+        self.model.load_state_dict(torch.load(best_model_path))
+
+        return self.model
+
+    def vali(self, train_loader, vali_loader, criterion):
+        x, _ = train_loader.dataset.last_insample_window()
+        y = vali_loader.dataset.timeseries
+        x = torch.tensor(x, dtype=torch.float32).to(self.device)
+        x = x.unsqueeze(-1)
+
+        self.model.eval()
+        with torch.no_grad():
+            # decoder input
+            B, _, C = x.shape
+            dec_inp = torch.zeros((B, self.args.pred_len, C)).float().to(self.device)
+            dec_inp = torch.cat([x[:, -self.args.label_len:, :], dec_inp], dim=1).float()
+            # encoder - decoder
+            outputs = torch.zeros((B, self.args.pred_len, C)).float()  # .to(self.device)
+            id_list = np.arange(0, B, 500)  # validation set size
+            id_list = np.append(id_list, B)
+            for i in range(len(id_list) - 1):
+                outputs[id_list[i]:id_list[i + 1], :, :] = self.model(x[id_list[i]:id_list[i + 1]], None,
+                                                                      dec_inp[id_list[i]:id_list[i + 1]],
+                                                                      None).detach().cpu()
+            f_dim = -1 if self.args.features == 'MS' else 0
+            outputs = outputs[:, -self.args.pred_len:, f_dim:]
+            pred = outputs
+            true = torch.from_numpy(np.array(y))
+            batch_y_mark = torch.ones(true.shape)
+
+            loss = criterion(x.detach().cpu()[:, :, 0], self.args.frequency_map, pred[:, :, 0], true, batch_y_mark)
+
+        self.model.train()
+        return loss
+
+    def test(self, setting, test=0):
+        _, train_loader = self._get_data(flag='train')
+        _, test_loader = self._get_data(flag='test')
+        x, _ = train_loader.dataset.last_insample_window()
+        y = test_loader.dataset.timeseries
+        x = torch.tensor(x, dtype=torch.float32).to(self.device)
+        x = x.unsqueeze(-1)
+
+        if test:
+            print('loading model')
+            self.model.load_state_dict(torch.load(os.path.join('./checkpoints/' + setting, 'checkpoint.pth')))
+
+        folder_path = './test_results/' + setting + '/'
+        if not os.path.exists(folder_path):
+            os.makedirs(folder_path)
+
+        self.model.eval()
+        with torch.no_grad():
+            B, _, C = x.shape
+            dec_inp = torch.zeros((B, self.args.pred_len, C)).float().to(self.device)
+            dec_inp = torch.cat([x[:, -self.args.label_len:, :], dec_inp], dim=1).float()
+            # encoder - decoder
+            outputs = torch.zeros((B, self.args.pred_len, C)).float().to(self.device)
+            id_list = np.arange(0, B, 1)
+            id_list = np.append(id_list, B)
+            for i in range(len(id_list) - 1):
+                outputs[id_list[i]:id_list[i + 1], :, :] = self.model(x[id_list[i]:id_list[i + 1]], None,
+                                                                      dec_inp[id_list[i]:id_list[i + 1]], None)
+
+                if id_list[i] % 1000 == 0:
+                    print(id_list[i])
+
+            f_dim = -1 if self.args.features == 'MS' else 0
+            outputs = outputs[:, -self.args.pred_len:, f_dim:]
+            outputs = outputs.detach().cpu().numpy()
+
+            preds = outputs
+            trues = y
+            x = x.detach().cpu().numpy()
+
+            for i in range(0, preds.shape[0], preds.shape[0] // 10):
+                gt = np.concatenate((x[i, :, 0], trues[i]), axis=0)
+                pd = np.concatenate((x[i, :, 0], preds[i, :, 0]), axis=0)
+                visual(gt, pd, os.path.join(folder_path, str(i) + '.pdf'))
+
+        print('test shape:', preds.shape)
+
+        # result save
+        folder_path = './m4_results/' + self.args.model + '/'
+        if not os.path.exists(folder_path):
+            os.makedirs(folder_path)
+
+        forecasts_df = pandas.DataFrame(preds[:, :, 0], columns=[f'V{i + 1}' for i in range(self.args.pred_len)])
+        forecasts_df.index = test_loader.dataset.ids[:preds.shape[0]]
+        forecasts_df.index.name = 'id'
+        forecasts_df.set_index(forecasts_df.columns[0], inplace=True)
+        forecasts_df.to_csv(folder_path + self.args.seasonal_patterns + '_forecast.csv')
+
+        print(self.args.model)
+        file_path = './m4_results/' + self.args.model + '/'
+        if 'Weekly_forecast.csv' in os.listdir(file_path) \
+                and 'Monthly_forecast.csv' in os.listdir(file_path) \
+                and 'Yearly_forecast.csv' in os.listdir(file_path) \
+                and 'Daily_forecast.csv' in os.listdir(file_path) \
+                and 'Hourly_forecast.csv' in os.listdir(file_path) \
+                and 'Quarterly_forecast.csv' in os.listdir(file_path):
+            m4_summary = M4Summary(file_path, self.args.root_path)
+            # m4_forecast.set_index(m4_winner_forecast.columns[0], inplace=True)
+            smape_results, owa_results, mape, mase = m4_summary.evaluate()
+            print('smape:', smape_results)
+            print('mape:', mape)
+            print('mase:', mase)
+            print('owa:', owa_results)
+        else:
+            print('After all 6 tasks are finished, you can calculate the averaged index')
+        return