[ { "path": ".gitignore", "content": "data/\nTimeMachine_supervised/exp/__pycache__/\nTimeMachine_supervised/data_provider/__pycache__/\nTimeMachine_supervised/models/__pycache__/\nTimeMachine_supervised/RevIN/__pycache__/\nTimeMachine_supervised/utils/__pycache__/\nTimeMachine_supervised/checkpoints/\nTimeMachine_supervised/logs/LongForecasting/\nTimeMachine_supervised/results/\nTimeMachine_supervised/test_results/\nTimeMachine_supervised/result.txt\nTimeMachine_supervised/csv_results/\n.vscode/settings.json\n" }, { "path": "LICENSE", "content": " Apache License\n Version 2.0, January 2004\n http://www.apache.org/licenses/\n\n TERMS AND CONDITIONS FOR USE, REPRODUCTION, AND DISTRIBUTION\n\n 1. Definitions.\n\n \"License\" shall mean the terms and conditions for use, reproduction,\n and distribution as defined by Sections 1 through 9 of this document.\n\n \"Licensor\" shall mean the copyright owner or entity authorized by\n the copyright owner that is granting the License.\n\n \"Legal Entity\" shall mean the union of the acting entity and all\n other entities that control, are controlled by, or are under common\n control with that entity. For the purposes of this definition,\n \"control\" means (i) the power, direct or indirect, to cause the\n direction or management of such entity, whether by contract or\n otherwise, or (ii) ownership of fifty percent (50%) or more of the\n outstanding shares, or (iii) beneficial ownership of such entity.\n\n \"You\" (or \"Your\") shall mean an individual or Legal Entity\n exercising permissions granted by this License.\n\n \"Source\" form shall mean the preferred form for making modifications,\n including but not limited to software source code, documentation\n source, and configuration files.\n\n \"Object\" form shall mean any form resulting from mechanical\n transformation or translation of a Source form, including but\n not limited to compiled object code, generated documentation,\n and conversions to other media types.\n\n \"Work\" shall mean the work of authorship, whether in Source or\n Object form, made available under the License, as indicated by a\n copyright notice that is included in or attached to the work\n (an example is provided in the Appendix below).\n\n \"Derivative Works\" shall mean any work, whether in Source or Object\n form, that is based on (or derived from) the Work and for which the\n editorial revisions, annotations, elaborations, or other modifications\n represent, as a whole, an original work of authorship. For the purposes\n of this License, Derivative Works shall not include works that remain\n separable from, or merely link (or bind by name) to the interfaces of,\n the Work and Derivative Works thereof.\n\n \"Contribution\" shall mean any work of authorship, including\n the original version of the Work and any modifications or additions\n to that Work or Derivative Works thereof, that is intentionally\n submitted to Licensor for inclusion in the Work by the copyright owner\n or by an individual or Legal Entity authorized to submit on behalf of\n the copyright owner. For the purposes of this definition, \"submitted\"\n means any form of electronic, verbal, or written communication sent\n to the Licensor or its representatives, including but not limited to\n communication on electronic mailing lists, source code control systems,\n and issue tracking systems that are managed by, or on behalf of, the\n Licensor for the purpose of discussing and improving the Work, but\n excluding communication that is conspicuously marked or otherwise\n designated in writing by the copyright owner as \"Not a Contribution.\"\n\n \"Contributor\" shall mean Licensor and any individual or Legal Entity\n on behalf of whom a Contribution has been received by Licensor and\n subsequently incorporated within the Work.\n\n 2. Grant of Copyright License. Subject to the terms and conditions of\n this License, each Contributor hereby grants to You a perpetual,\n worldwide, non-exclusive, no-charge, royalty-free, irrevocable\n copyright license to reproduce, prepare Derivative Works of,\n publicly display, publicly perform, sublicense, and distribute the\n Work and such Derivative Works in Source or Object form.\n\n 3. Grant of Patent License. Subject to the terms and conditions of\n this License, each Contributor hereby grants to You a perpetual,\n worldwide, non-exclusive, no-charge, royalty-free, irrevocable\n (except as stated in this section) patent license to make, have made,\n use, offer to sell, sell, import, and otherwise transfer the Work,\n where such license applies only to those patent claims licensable\n by such Contributor that are necessarily infringed by their\n Contribution(s) alone or by combination of their Contribution(s)\n with the Work to which such Contribution(s) was submitted. If You\n institute patent litigation against any entity (including a\n cross-claim or counterclaim in a lawsuit) alleging that the Work\n or a Contribution incorporated within the Work constitutes direct\n or contributory patent infringement, then any patent licenses\n granted to You under this License for that Work shall terminate\n as of the date such litigation is filed.\n\n 4. Redistribution. You may reproduce and distribute copies of the\n Work or Derivative Works thereof in any medium, with or without\n modifications, and in Source or Object form, provided that You\n meet the following conditions:\n\n (a) You must give any other recipients of the Work or\n Derivative Works a copy of this License; and\n\n (b) You must cause any modified files to carry prominent notices\n stating that You changed the files; and\n\n (c) You must retain, in the Source form of any Derivative Works\n that You distribute, all copyright, patent, trademark, and\n attribution notices from the Source form of the Work,\n excluding those notices that do not pertain to any part of\n the Derivative Works; and\n\n (d) If the Work includes a \"NOTICE\" text file as part of its\n distribution, then any Derivative Works that You distribute must\n include a readable copy of the attribution notices contained\n within such NOTICE file, excluding those notices that do not\n pertain to any part of the Derivative Works, in at least one\n of the following places: within a NOTICE text file distributed\n as part of the Derivative Works; within the Source form or\n documentation, if provided along with the Derivative Works; or,\n within a display generated by the Derivative Works, if and\n wherever such third-party notices normally appear. The contents\n of the NOTICE file are for informational purposes only and\n do not modify the License. You may add Your own attribution\n notices within Derivative Works that You distribute, alongside\n or as an addendum to the NOTICE text from the Work, provided\n that such additional attribution notices cannot be construed\n as modifying the License.\n\n You may add Your own copyright statement to Your modifications and\n may provide additional or different license terms and conditions\n for use, reproduction, or distribution of Your modifications, or\n for any such Derivative Works as a whole, provided Your use,\n reproduction, and distribution of the Work otherwise complies with\n the conditions stated in this License.\n\n 5. Submission of Contributions. Unless You explicitly state otherwise,\n any Contribution intentionally submitted for inclusion in the Work\n by You to the Licensor shall be under the terms and conditions of\n this License, without any additional terms or conditions.\n Notwithstanding the above, nothing herein shall supersede or modify\n the terms of any separate license agreement you may have executed\n with Licensor regarding such Contributions.\n\n 6. Trademarks. This License does not grant permission to use the trade\n names, trademarks, service marks, or product names of the Licensor,\n except as required for reasonable and customary use in describing the\n origin of the Work and reproducing the content of the NOTICE file.\n\n 7. Disclaimer of Warranty. Unless required by applicable law or\n agreed to in writing, Licensor provides the Work (and each\n Contributor provides its Contributions) on an \"AS IS\" BASIS,\n WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or\n implied, including, without limitation, any warranties or conditions\n of TITLE, NON-INFRINGEMENT, MERCHANTABILITY, or FITNESS FOR A\n PARTICULAR PURPOSE. You are solely responsible for determining the\n appropriateness of using or redistributing the Work and assume any\n risks associated with Your exercise of permissions under this License.\n\n 8. Limitation of Liability. In no event and under no legal theory,\n whether in tort (including negligence), contract, or otherwise,\n unless required by applicable law (such as deliberate and grossly\n negligent acts) or agreed to in writing, shall any Contributor be\n liable to You for damages, including any direct, indirect, special,\n incidental, or consequential damages of any character arising as a\n result of this License or out of the use or inability to use the\n Work (including but not limited to damages for loss of goodwill,\n work stoppage, computer failure or malfunction, or any and all\n other commercial damages or losses), even if such Contributor\n has been advised of the possibility of such damages.\n\n 9. Accepting Warranty or Additional Liability. While redistributing\n the Work or Derivative Works thereof, You may choose to offer,\n and charge a fee for, acceptance of support, warranty, indemnity,\n or other liability obligations and/or rights consistent with this\n License. However, in accepting such obligations, You may act only\n on Your own behalf and on Your sole responsibility, not on behalf\n of any other Contributor, and only if You agree to indemnify,\n defend, and hold each Contributor harmless for any liability\n incurred by, or claims asserted against, such Contributor by reason\n of your accepting any such warranty or additional liability.\n\n END OF TERMS AND CONDITIONS\n\n APPENDIX: How to apply the Apache License to your work.\n\n To apply the Apache License to your work, attach the following\n boilerplate notice, with the fields enclosed by brackets \"[]\"\n replaced with your own identifying information. (Don't include\n the brackets!) The text should be enclosed in the appropriate\n comment syntax for the file format. We also recommend that a\n file or class name and description of purpose be included on the\n same \"printed page\" as the copyright notice for easier\n identification within third-party archives.\n\n Copyright [yyyy] [name of copyright owner]\n\n Licensed under the Apache License, Version 2.0 (the \"License\");\n you may not use this file except in compliance with the License.\n You may obtain a copy of the License at\n\n http://www.apache.org/licenses/LICENSE-2.0\n\n Unless required by applicable law or agreed to in writing, software\n distributed under the License is distributed on an \"AS IS\" BASIS,\n WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n See the License for the specific language governing permissions and\n limitations under the License.\n" }, { "path": "README.md", "content": "#
TimeMachine
\n\n![Alt text](./pics/TimeMachine.PNG)\n### Welcome to the official repository of: [TimeMachine: A Time Series is Worth 4 Mambas for Long-term Forecasting](https://arxiv.org/pdf/2403.09898.pdf). \n## :triangular_flag_on_post: TimeMachine is accepted to [**ECAI**](https://www.ecai2024.eu/) \n## Usage\n\n1. Install requirements. ```pip install -r requirements.txt```\n\n2. Navigate through our example scripts located at ```./scripts/TimeMachine```. You'll find the core of TimeMachine in ```models/TimeMachine.py```. For example, to get the multivariate forecasting results for weather dataset, just run the following command, and you can open ```./result.txt``` to see the results once the training is completed. Moreover, the results will also be available at ```csv_results```, which can be utilized to make queries in the dataframe:\n```\nsh ./scripts/TimeMachine/weather.sh\n```\n\nHyper-paramters can be tuned based upon needs (e.g. different look-back windows and prediction lengths). TimeMachine is built on the popular [PatchTST](https://github.com/yuqinie98/PatchTST) framework.\n\n\n## Acknowledgement\n\nWe are deeply grateful for the valuable code and efforts contributed by the following GitHub repositories. Their contributions have been immensely beneficial to our work.\n- Mamba (https://github.com/state-spaces/mamba)\n- PatchTST (https://github.com/yuqinie98/PatchTST)\n- iTransformer (https://github.com/thuml/iTransformer)\n- RevIN (https://github.com/ts-kim/RevIN)\n- Reformer (https://github.com/lucidrains/reformer-pytorch)\n- Informer (https://github.com/zhouhaoyi/Informer2020)\n- FlashAttention (https://github.com/shreyansh26/FlashAttention-PyTorch)\n- Autoformer (https://github.com/thuml/Autoformer)\n- Stationary (https://github.com/thuml/Nonstationary_Transformers)\n- Time-Series-Library (https://github.com/thuml/Time-Series-Library)\n\n\n\n## Citation\n\nIf you find this repo useful in your research, please consider citing our paper as follows:\n\n```\n@article{timemachine,\n title = {TimeMachine: A Time Series is Worth 4 Mambas for Long-term Forecasting},\n author = {Ahamed, Md Atik and Cheng, Qiang},\n journal = {arXiv preprint arXiv:2403.09898},\n year = {2024}\n}\n```\n\n" }, { "path": "TimeMachine_supervised/RevIN/RevIN.py", "content": "import torch\nimport torch.nn as nn\n\nclass RevIN(nn.Module):\n def __init__(self, num_features: int, eps=1e-5, affine=True):\n \"\"\"\n :param num_features: the number of features or channels\n :param eps: a value added for numerical stability\n :param affine: if True, RevIN has learnable affine parameters\n \"\"\"\n super(RevIN, self).__init__()\n self.num_features = num_features\n self.eps = eps\n self.affine = affine\n if self.affine:\n self._init_params()\n\n def forward(self, x, mode:str):\n if mode == 'norm':\n self._get_statistics(x)\n x = self._normalize(x)\n elif mode == 'denorm':\n x = self._denormalize(x)\n else: raise NotImplementedError\n return x\n\n def _init_params(self):\n # initialize RevIN params: (C,)\n self.affine_weight = nn.Parameter(torch.ones(self.num_features))\n self.affine_bias = nn.Parameter(torch.zeros(self.num_features))\n\n def _get_statistics(self, x):\n dim2reduce = tuple(range(1, x.ndim-1))\n self.mean = torch.mean(x, dim=dim2reduce, keepdim=True).detach()\n self.stdev = torch.sqrt(torch.var(x, dim=dim2reduce, keepdim=True, unbiased=False) + self.eps).detach()\n\n def _normalize(self, x):\n x = x - self.mean\n x = x / self.stdev\n if self.affine:\n x = x * self.affine_weight\n x = x + self.affine_bias\n return x\n\n def _denormalize(self, x):\n if self.affine:\n x = x - self.affine_bias\n x = x / (self.affine_weight + self.eps*self.eps)\n x = x * self.stdev\n x = x + self.mean\n return x\n" }, { "path": "TimeMachine_supervised/data_provider/data_factory.py", "content": "from data_provider.data_loader import Dataset_ETT_hour, Dataset_ETT_minute, Dataset_Custom, Dataset_Pred\nfrom torch.utils.data import DataLoader\n\ndata_dict = {\n 'ETTh1': Dataset_ETT_hour,\n 'ETTh2': Dataset_ETT_hour,\n 'ETTm1': Dataset_ETT_minute,\n 'ETTm2': Dataset_ETT_minute,\n 'custom': Dataset_Custom,\n}\n\n\ndef data_provider(args, flag):\n Data = data_dict[args.data]\n timeenc = 0 if args.embed != 'timeF' else 1\n\n if flag == 'test':\n shuffle_flag = False\n drop_last = True\n batch_size = args.batch_size\n freq = args.freq\n elif flag == 'pred':\n shuffle_flag = False\n drop_last = False\n batch_size = 1\n freq = args.freq\n Data = Dataset_Pred\n else:\n shuffle_flag = True\n drop_last = True\n batch_size = args.batch_size\n freq = args.freq\n\n data_set = Data(\n root_path=args.root_path,\n data_path=args.data_path,\n flag=flag,\n size=[args.seq_len, args.label_len, args.pred_len],\n features=args.features,\n target=args.target,\n timeenc=timeenc,\n freq=freq\n )\n print(flag, len(data_set))\n data_loader = DataLoader(\n data_set,\n batch_size=batch_size,\n shuffle=shuffle_flag,\n num_workers=args.num_workers,\n drop_last=drop_last)\n return data_set, data_loader\n" }, { "path": "TimeMachine_supervised/data_provider/data_loader.py", "content": "import os\nimport numpy as np\nimport pandas as pd\nimport os\nimport torch\nfrom torch.utils.data import Dataset, DataLoader\nfrom sklearn.preprocessing import StandardScaler\nfrom utils.timefeatures import time_features\nimport warnings\n\nwarnings.filterwarnings('ignore')\n\n\nclass Dataset_ETT_hour(Dataset):\n def __init__(self, root_path, flag='train', size=None,\n features='S', data_path='ETTh1.csv',\n target='OT', scale=True, timeenc=0, freq='h'):\n # size [seq_len, label_len, pred_len]\n # info\n if size == None:\n self.seq_len = 24 * 4 * 4\n self.label_len = 24 * 4\n self.pred_len = 24 * 4\n else:\n self.seq_len = size[0]\n self.label_len = size[1]\n self.pred_len = size[2]\n # init\n assert flag in ['train', 'test', 'val']\n type_map = {'train': 0, 'val': 1, 'test': 2}\n self.set_type = type_map[flag]\n\n self.features = features\n self.target = target\n self.scale = scale\n self.timeenc = timeenc\n self.freq = freq\n\n self.root_path = root_path\n self.data_path = data_path\n self.__read_data__()\n\n def __read_data__(self):\n self.scaler = StandardScaler()\n df_raw = pd.read_csv(os.path.join(self.root_path,\n self.data_path))\n\n border1s = [0, 12 * 30 * 24 - self.seq_len, 12 * 30 * 24 + 4 * 30 * 24 - self.seq_len]\n border2s = [12 * 30 * 24, 12 * 30 * 24 + 4 * 30 * 24, 12 * 30 * 24 + 8 * 30 * 24]\n border1 = border1s[self.set_type]\n border2 = border2s[self.set_type]\n\n if self.features == 'M' or self.features == 'MS':\n cols_data = df_raw.columns[1:]\n df_data = df_raw[cols_data]\n elif self.features == 'S':\n df_data = df_raw[[self.target]]\n\n if self.scale:\n train_data = df_data[border1s[0]:border2s[0]]\n self.scaler.fit(train_data.values)\n data = self.scaler.transform(df_data.values)\n else:\n data = df_data.values\n\n df_stamp = df_raw[['date']][border1:border2]\n df_stamp['date'] = pd.to_datetime(df_stamp.date)\n if self.timeenc == 0:\n df_stamp['month'] = df_stamp.date.apply(lambda row: row.month, 1)\n df_stamp['day'] = df_stamp.date.apply(lambda row: row.day, 1)\n df_stamp['weekday'] = df_stamp.date.apply(lambda row: row.weekday(), 1)\n df_stamp['hour'] = df_stamp.date.apply(lambda row: row.hour, 1)\n data_stamp = df_stamp.drop(['date'], axis=1).values\n elif self.timeenc == 1:\n data_stamp = time_features(pd.to_datetime(df_stamp['date'].values), freq=self.freq)\n data_stamp = data_stamp.transpose(1, 0)\n\n self.data_x = data[border1:border2]\n self.data_y = data[border1:border2]\n self.data_stamp = data_stamp\n\n def __getitem__(self, index):\n s_begin = index\n s_end = s_begin + self.seq_len\n r_begin = s_end - self.label_len\n r_end = r_begin + self.label_len + self.pred_len\n\n seq_x = self.data_x[s_begin:s_end]\n seq_y = self.data_y[r_begin:r_end]\n seq_x_mark = self.data_stamp[s_begin:s_end]\n seq_y_mark = self.data_stamp[r_begin:r_end]\n\n return seq_x, seq_y, seq_x_mark, seq_y_mark\n\n def __len__(self):\n return len(self.data_x) - self.seq_len - self.pred_len + 1\n\n def inverse_transform(self, data):\n return self.scaler.inverse_transform(data)\n\n\nclass Dataset_ETT_minute(Dataset):\n def __init__(self, root_path, flag='train', size=None,\n features='S', data_path='ETTm1.csv',\n target='OT', scale=True, timeenc=0, freq='t'):\n # size [seq_len, label_len, pred_len]\n # info\n if size == None:\n self.seq_len = 24 * 4 * 4\n self.label_len = 24 * 4\n self.pred_len = 24 * 4\n else:\n self.seq_len = size[0]\n self.label_len = size[1]\n self.pred_len = size[2]\n # init\n assert flag in ['train', 'test', 'val']\n type_map = {'train': 0, 'val': 1, 'test': 2}\n self.set_type = type_map[flag]\n\n self.features = features\n self.target = target\n self.scale = scale\n self.timeenc = timeenc\n self.freq = freq\n\n self.root_path = root_path\n self.data_path = data_path\n self.__read_data__()\n\n def __read_data__(self):\n self.scaler = StandardScaler()\n df_raw = pd.read_csv(os.path.join(self.root_path,\n self.data_path))\n\n border1s = [0, 12 * 30 * 24 * 4 - self.seq_len, 12 * 30 * 24 * 4 + 4 * 30 * 24 * 4 - self.seq_len]\n border2s = [12 * 30 * 24 * 4, 12 * 30 * 24 * 4 + 4 * 30 * 24 * 4, 12 * 30 * 24 * 4 + 8 * 30 * 24 * 4]\n border1 = border1s[self.set_type]\n border2 = border2s[self.set_type]\n\n if self.features == 'M' or self.features == 'MS':\n cols_data = df_raw.columns[1:]\n df_data = df_raw[cols_data]\n elif self.features == 'S':\n df_data = df_raw[[self.target]]\n\n if self.scale:\n train_data = df_data[border1s[0]:border2s[0]]\n self.scaler.fit(train_data.values)\n data = self.scaler.transform(df_data.values)\n else:\n data = df_data.values\n\n df_stamp = df_raw[['date']][border1:border2]\n df_stamp['date'] = pd.to_datetime(df_stamp.date)\n if self.timeenc == 0:\n df_stamp['month'] = df_stamp.date.apply(lambda row: row.month, 1)\n df_stamp['day'] = df_stamp.date.apply(lambda row: row.day, 1)\n df_stamp['weekday'] = df_stamp.date.apply(lambda row: row.weekday(), 1)\n df_stamp['hour'] = df_stamp.date.apply(lambda row: row.hour, 1)\n df_stamp['minute'] = df_stamp.date.apply(lambda row: row.minute, 1)\n df_stamp['minute'] = df_stamp.minute.map(lambda x: x // 15)\n data_stamp = df_stamp.drop(['date'], axis=1).values\n elif self.timeenc == 1:\n data_stamp = time_features(pd.to_datetime(df_stamp['date'].values), freq=self.freq)\n data_stamp = data_stamp.transpose(1, 0)\n\n self.data_x = data[border1:border2]\n self.data_y = data[border1:border2]\n self.data_stamp = data_stamp\n\n def __getitem__(self, index):\n s_begin = index\n s_end = s_begin + self.seq_len\n r_begin = s_end - self.label_len\n r_end = r_begin + self.label_len + self.pred_len\n\n seq_x = self.data_x[s_begin:s_end]\n seq_y = self.data_y[r_begin:r_end]\n seq_x_mark = self.data_stamp[s_begin:s_end]\n seq_y_mark = self.data_stamp[r_begin:r_end]\n\n return seq_x, seq_y, seq_x_mark, seq_y_mark\n\n def __len__(self):\n return len(self.data_x) - self.seq_len - self.pred_len + 1\n\n def inverse_transform(self, data):\n return self.scaler.inverse_transform(data)\n\n\nclass Dataset_Custom(Dataset):\n def __init__(self, root_path, flag='train', size=None,\n features='S', data_path='ETTh1.csv',\n target='OT', scale=True, timeenc=0, freq='h'):\n # size [seq_len, label_len, pred_len]\n # info\n if size == None:\n self.seq_len = 24 * 4 * 4\n self.label_len = 24 * 4\n self.pred_len = 24 * 4\n else:\n self.seq_len = size[0]\n self.label_len = size[1]\n self.pred_len = size[2]\n # init\n assert flag in ['train', 'test', 'val']\n type_map = {'train': 0, 'val': 1, 'test': 2}\n self.set_type = type_map[flag]\n\n self.features = features\n self.target = target\n self.scale = scale\n self.timeenc = timeenc\n self.freq = freq\n\n self.root_path = root_path\n self.data_path = data_path\n self.__read_data__()\n\n def __read_data__(self):\n self.scaler = StandardScaler()\n df_raw = pd.read_csv(os.path.join(self.root_path,\n self.data_path))\n\n '''\n df_raw.columns: ['date', ...(other features), target feature]\n '''\n cols = list(df_raw.columns)\n cols.remove(self.target)\n cols.remove('date')\n df_raw = df_raw[['date'] + cols + [self.target]]\n # print(cols)\n num_train = int(len(df_raw) * 0.7)\n num_test = int(len(df_raw) * 0.2)\n num_vali = len(df_raw) - num_train - num_test\n border1s = [0, num_train - self.seq_len, len(df_raw) - num_test - self.seq_len]\n border2s = [num_train, num_train + num_vali, len(df_raw)]\n border1 = border1s[self.set_type]\n border2 = border2s[self.set_type]\n\n if self.features == 'M' or self.features == 'MS':\n cols_data = df_raw.columns[1:]\n df_data = df_raw[cols_data]\n elif self.features == 'S':\n df_data = df_raw[[self.target]]\n\n if self.scale:\n train_data = df_data[border1s[0]:border2s[0]]\n self.scaler.fit(train_data.values)\n # print(self.scaler.mean_)\n # exit()\n data = self.scaler.transform(df_data.values)\n else:\n data = df_data.values\n\n df_stamp = df_raw[['date']][border1:border2]\n df_stamp['date'] = pd.to_datetime(df_stamp.date)\n if self.timeenc == 0:\n df_stamp['month'] = df_stamp.date.apply(lambda row: row.month, 1)\n df_stamp['day'] = df_stamp.date.apply(lambda row: row.day, 1)\n df_stamp['weekday'] = df_stamp.date.apply(lambda row: row.weekday(), 1)\n df_stamp['hour'] = df_stamp.date.apply(lambda row: row.hour, 1)\n data_stamp = df_stamp.drop(['date'], axis=1).values\n elif self.timeenc == 1:\n data_stamp = time_features(pd.to_datetime(df_stamp['date'].values), freq=self.freq)\n data_stamp = data_stamp.transpose(1, 0)\n\n self.data_x = data[border1:border2]\n self.data_y = data[border1:border2]\n self.data_stamp = data_stamp\n\n def __getitem__(self, index):\n s_begin = index\n s_end = s_begin + self.seq_len\n r_begin = s_end - self.label_len\n r_end = r_begin + self.label_len + self.pred_len\n\n seq_x = self.data_x[s_begin:s_end]\n seq_y = self.data_y[r_begin:r_end]\n seq_x_mark = self.data_stamp[s_begin:s_end]\n seq_y_mark = self.data_stamp[r_begin:r_end]\n\n return seq_x, seq_y, seq_x_mark, seq_y_mark\n\n def __len__(self):\n return len(self.data_x) - self.seq_len - self.pred_len + 1\n\n def inverse_transform(self, data):\n return self.scaler.inverse_transform(data)\n \n\nclass Dataset_Pred(Dataset):\n def __init__(self, root_path, flag='pred', size=None,\n features='S', data_path='ETTh1.csv',\n target='OT', scale=True, inverse=False, timeenc=0, freq='15min', cols=None):\n # size [seq_len, label_len, pred_len]\n # info\n if size == None:\n self.seq_len = 24 * 4 * 4\n self.label_len = 24 * 4\n self.pred_len = 24 * 4\n else:\n self.seq_len = size[0]\n self.label_len = size[1]\n self.pred_len = size[2]\n # init\n assert flag in ['pred']\n\n self.features = features\n self.target = target\n self.scale = scale\n self.inverse = inverse\n self.timeenc = timeenc\n self.freq = freq\n self.cols = cols\n self.root_path = root_path\n self.data_path = data_path\n self.__read_data__()\n\n def __read_data__(self):\n self.scaler = StandardScaler()\n df_raw = pd.read_csv(os.path.join(self.root_path,\n self.data_path))\n '''\n df_raw.columns: ['date', ...(other features), target feature]\n '''\n if self.cols:\n cols = self.cols.copy()\n cols.remove(self.target)\n else:\n cols = list(df_raw.columns)\n cols.remove(self.target)\n cols.remove('date')\n df_raw = df_raw[['date'] + cols + [self.target]]\n border1 = len(df_raw) - self.seq_len\n border2 = len(df_raw)\n\n if self.features == 'M' or self.features == 'MS':\n cols_data = df_raw.columns[1:]\n df_data = df_raw[cols_data]\n elif self.features == 'S':\n df_data = df_raw[[self.target]]\n\n if self.scale:\n self.scaler.fit(df_data.values)\n data = self.scaler.transform(df_data.values)\n else:\n data = df_data.values\n\n tmp_stamp = df_raw[['date']][border1:border2]\n tmp_stamp['date'] = pd.to_datetime(tmp_stamp.date)\n pred_dates = pd.date_range(tmp_stamp.date.values[-1], periods=self.pred_len + 1, freq=self.freq)\n\n df_stamp = pd.DataFrame(columns=['date'])\n df_stamp.date = list(tmp_stamp.date.values) + list(pred_dates[1:])\n if self.timeenc == 0:\n df_stamp['month'] = df_stamp.date.apply(lambda row: row.month, 1)\n df_stamp['day'] = df_stamp.date.apply(lambda row: row.day, 1)\n df_stamp['weekday'] = df_stamp.date.apply(lambda row: row.weekday(), 1)\n df_stamp['hour'] = df_stamp.date.apply(lambda row: row.hour, 1)\n df_stamp['minute'] = df_stamp.date.apply(lambda row: row.minute, 1)\n df_stamp['minute'] = df_stamp.minute.map(lambda x: x // 15)\n data_stamp = df_stamp.drop(['date'], axis=1).values\n elif self.timeenc == 1:\n data_stamp = time_features(pd.to_datetime(df_stamp['date'].values), freq=self.freq)\n data_stamp = data_stamp.transpose(1, 0)\n\n self.data_x = data[border1:border2]\n if self.inverse:\n self.data_y = df_data.values[border1:border2]\n else:\n self.data_y = data[border1:border2]\n self.data_stamp = data_stamp\n\n def __getitem__(self, index):\n s_begin = index\n s_end = s_begin + self.seq_len\n r_begin = s_end - self.label_len\n r_end = r_begin + self.label_len + self.pred_len\n\n seq_x = self.data_x[s_begin:s_end]\n if self.inverse:\n seq_y = self.data_x[r_begin:r_begin + self.label_len]\n else:\n seq_y = self.data_y[r_begin:r_begin + self.label_len]\n seq_x_mark = self.data_stamp[s_begin:s_end]\n seq_y_mark = self.data_stamp[r_begin:r_end]\n\n return seq_x, seq_y, seq_x_mark, seq_y_mark\n\n def __len__(self):\n return len(self.data_x) - self.seq_len + 1\n\n def inverse_transform(self, data):\n return self.scaler.inverse_transform(data)\n" }, { "path": "TimeMachine_supervised/exp/exp_basic.py", "content": "import os\nimport torch\nimport numpy as np\n\n\nclass Exp_Basic(object):\n def __init__(self, args):\n self.args = args\n self.device = self._acquire_device()\n self.model = self._build_model().to(self.device)\n\n def _build_model(self):\n raise NotImplementedError\n return None\n\n def _acquire_device(self):\n if self.args.use_gpu:\n os.environ[\"CUDA_VISIBLE_DEVICES\"] = str(\n self.args.gpu) if not self.args.use_multi_gpu else self.args.devices\n device = torch.device('cuda:{}'.format(self.args.gpu))\n print('Use GPU: cuda:{}'.format(self.args.gpu))\n else:\n device = torch.device('cpu')\n print('Use CPU')\n return device\n\n def _get_data(self):\n pass\n\n def vali(self):\n pass\n\n def train(self):\n pass\n\n def test(self):\n pass\n" }, { "path": "TimeMachine_supervised/exp/exp_main.py", "content": "from data_provider.data_factory import data_provider\nfrom exp.exp_basic import Exp_Basic\nfrom models import TimeMachine\nfrom utils.tools import EarlyStopping, adjust_learning_rate, visual, test_params_flop\nfrom utils.metrics import metric\n\nimport numpy as np\nimport torch\nimport torch.nn as nn\nfrom torch import optim\nfrom torch.optim import lr_scheduler \nimport pandas as pd\nimport os\nimport time\n\nimport warnings\nimport matplotlib.pyplot as plt\nimport numpy as np\n\nwarnings.filterwarnings('ignore')\n\nclass Exp_Main(Exp_Basic):\n def __init__(self, args):\n super(Exp_Main, self).__init__(args)\n\n def _build_model(self):\n model_dict = {\n 'TimeMachine':TimeMachine\n }\n model = model_dict[self.args.model].Model(self.args).float()\n\n if self.args.use_multi_gpu and self.args.use_gpu:\n model = nn.DataParallel(model, device_ids=self.args.device_ids)\n return model\n\n def _get_data(self, flag):\n data_set, data_loader = data_provider(self.args, flag)\n return data_set, data_loader\n\n def _select_optimizer(self):\n model_optim = optim.Adam(self.model.parameters(), lr=self.args.learning_rate)\n return model_optim\n\n def _select_criterion(self):\n criterion = nn.MSELoss()\n return criterion\n\n def vali(self, vali_data, vali_loader, criterion):\n total_loss = []\n self.model.eval()\n with torch.no_grad():\n for i, (batch_x, batch_y, batch_x_mark, batch_y_mark) in enumerate(vali_loader):\n batch_x = batch_x.float().to(self.device) \n batch_y = batch_y.float()\n batch_x_mark = batch_x_mark.float().to(self.device)\n batch_y_mark = batch_y_mark.float().to(self.device)\n\n # decoder input\n dec_inp = torch.zeros_like(batch_y[:, -self.args.pred_len:, :]).float()\n dec_inp = torch.cat([batch_y[:, :self.args.label_len, :], dec_inp], dim=1).float().to(self.device)\n # encoder - decoder\n if self.args.use_amp:\n with torch.cuda.amp.autocast():\n if 'Machine' in self.args.model:\n outputs = self.model(batch_x)\n \n else:\n if 'Machine' in self.args.model:\n outputs = self.model(batch_x)\n \n f_dim = -1 if self.args.features == 'MS' else 0\n outputs = outputs[:, -self.args.pred_len:, f_dim:]\n batch_y = batch_y[:, -self.args.pred_len:, f_dim:].to(self.device)\n\n pred = outputs.detach().cpu()\n true = batch_y.detach().cpu()\n\n loss = criterion(pred, true)\n\n total_loss.append(loss)\n total_loss = np.average(total_loss)\n self.model.train()\n return total_loss\n\n def train(self, setting):\n train_data, train_loader = self._get_data(flag='train')\n vali_data, vali_loader = self._get_data(flag='val')\n test_data, test_loader = self._get_data(flag='test')\n\n path = os.path.join(self.args.checkpoints, setting)\n if not os.path.exists(path):\n os.makedirs(path)\n\n time_now = time.time()\n\n train_steps = len(train_loader)\n early_stopping = EarlyStopping(patience=self.args.patience, verbose=True)\n\n model_optim = self._select_optimizer()\n criterion = self._select_criterion()\n\n if self.args.use_amp:\n scaler = torch.cuda.amp.GradScaler()\n \n scheduler = lr_scheduler.OneCycleLR(optimizer = model_optim,\n steps_per_epoch = train_steps,\n pct_start = self.args.pct_start,\n epochs = self.args.train_epochs,\n max_lr = self.args.learning_rate)\n \n for epoch in range(self.args.train_epochs):\n iter_count = 0\n train_loss = []\n\n self.model.train()\n epoch_time = time.time()\n for i, (batch_x, batch_y, batch_x_mark, batch_y_mark) in enumerate(train_loader):\n iter_count += 1\n model_optim.zero_grad()\n batch_x = batch_x.float().to(self.device)\n\n batch_y = batch_y.float().to(self.device)\n batch_x_mark = batch_x_mark.float().to(self.device)\n batch_y_mark = batch_y_mark.float().to(self.device)\n\n # decoder input\n dec_inp = torch.zeros_like(batch_y[:, -self.args.pred_len:, :]).float()\n dec_inp = torch.cat([batch_y[:, :self.args.label_len, :], dec_inp], dim=1).float().to(self.device)\n\n # encoder - decoder\n if self.args.use_amp:\n with torch.cuda.amp.autocast():\n if 'Machine' in self.args.model:\n outputs = self.model(batch_x)\n \n\n f_dim = -1 if self.args.features == 'MS' else 0\n outputs = outputs[:, -self.args.pred_len:, f_dim:]\n batch_y = batch_y[:, -self.args.pred_len:, f_dim:].to(self.device)\n loss = criterion(outputs, batch_y)\n train_loss.append(loss.item())\n else:\n if 'Machine' in self.args.model:\n outputs = self.model(batch_x)\n \n \n # print(outputs.shape,batch_y.shape)\n f_dim = -1 if self.args.features == 'MS' else 0\n outputs = outputs[:, -self.args.pred_len:, f_dim:]\n batch_y = batch_y[:, -self.args.pred_len:, f_dim:].to(self.device)\n loss = criterion(outputs, batch_y)\n train_loss.append(loss.item())\n\n if (i + 1) % 100 == 0:\n print(\"\\titers: {0}, epoch: {1} | loss: {2:.7f}\".format(i + 1, epoch + 1, loss.item()))\n speed = (time.time() - time_now) / iter_count\n left_time = speed * ((self.args.train_epochs - epoch) * train_steps - i)\n print('\\tspeed: {:.4f}s/iter; left time: {:.4f}s'.format(speed, left_time))\n iter_count = 0\n time_now = time.time()\n\n if self.args.use_amp:\n scaler.scale(loss).backward()\n scaler.step(model_optim)\n scaler.update()\n else:\n loss.backward()\n model_optim.step()\n \n if self.args.lradj == 'TST':\n adjust_learning_rate(model_optim, scheduler, epoch + 1, self.args, printout=False)\n scheduler.step()\n\n print(\"Epoch: {} cost time: {}\".format(epoch + 1, time.time() - epoch_time))\n train_loss = np.average(train_loss)\n vali_loss = self.vali(vali_data, vali_loader, criterion)\n test_loss = self.vali(test_data, test_loader, criterion)\n\n print(\"Epoch: {0}, Steps: {1} | Train Loss: {2:.7f} Vali Loss: {3:.7f} Test Loss: {4:.7f}\".format(\n epoch + 1, train_steps, train_loss, vali_loss, test_loss))\n early_stopping(vali_loss, self.model, path)\n if early_stopping.early_stop:\n print(\"Early stopping\")\n break\n\n if self.args.lradj != 'TST':\n adjust_learning_rate(model_optim, scheduler, epoch + 1, self.args)\n else:\n print('Updating learning rate to {}'.format(scheduler.get_last_lr()[0]))\n\n best_model_path = path + '/' + 'checkpoint.pth'\n self.model.load_state_dict(torch.load(best_model_path))\n\n return self.model\n\n def test(self, setting, test=0):\n test_data, test_loader = self._get_data(flag='test')\n \n if test:\n print('loading model')\n self.model.load_state_dict(torch.load(os.path.join('./checkpoints/' + setting, 'checkpoint.pth')))\n\n preds = []\n trues = []\n inputx = []\n folder_path = './test_results/' + setting + '/'\n if not os.path.exists(folder_path):\n os.makedirs(folder_path)\n\n self.model.eval()\n with torch.no_grad():\n for i, (batch_x, batch_y, batch_x_mark, batch_y_mark) in enumerate(test_loader):\n batch_x = batch_x.float().to(self.device)\n batch_y = batch_y.float().to(self.device)\n\n batch_x_mark = batch_x_mark.float().to(self.device)\n batch_y_mark = batch_y_mark.float().to(self.device)\n\n # decoder input\n dec_inp = torch.zeros_like(batch_y[:, -self.args.pred_len:, :]).float()\n dec_inp = torch.cat([batch_y[:, :self.args.label_len, :], dec_inp], dim=1).float().to(self.device)\n # encoder - decoder\n if self.args.use_amp:\n with torch.cuda.amp.autocast():\n if 'Machine' in self.args.model:\n outputs = self.model(batch_x)\n \n else:\n if 'Machine' in self.args.model:\n outputs = self.model(batch_x)\n\n f_dim = -1 if self.args.features == 'MS' else 0\n # print(outputs.shape,batch_y.shape)\n outputs = outputs[:, -self.args.pred_len:, f_dim:]\n batch_y = batch_y[:, -self.args.pred_len:, f_dim:].to(self.device)\n outputs = outputs.detach().cpu().numpy()\n batch_y = batch_y.detach().cpu().numpy()\n\n pred = outputs # outputs.detach().cpu().numpy() # .squeeze()\n true = batch_y # batch_y.detach().cpu().numpy() # .squeeze()\n\n preds.append(pred)\n trues.append(true)\n inputx.append(batch_x.detach().cpu().numpy())\n if i % 20 == 0:\n input = batch_x.detach().cpu().numpy()\n gt = np.concatenate((input[0, :, -1], true[0, :, -1]), axis=0)\n prd = np.concatenate((input[0, :, -1], pred[0, :, -1]), axis=0)\n visual(gt, prd, os.path.join(folder_path, str(i) + '.pdf'))\n\n if self.args.test_flop:\n test_params_flop((batch_x.shape[1],batch_x.shape[2]))\n exit()\n preds = np.array(preds)\n trues = np.array(trues)\n inputx = np.array(inputx)\n\n preds = preds.reshape(-1, preds.shape[-2], preds.shape[-1])\n trues = trues.reshape(-1, trues.shape[-2], trues.shape[-1])\n inputx = inputx.reshape(-1, inputx.shape[-2], inputx.shape[-1])\n\n # result save\n folder_path = './results/' + setting + '/'\n if not os.path.exists(folder_path):\n os.makedirs(folder_path)\n\n mae, mse, rmse, mape, mspe, rse, corr = metric(preds, trues)\n \n print('mse:{}, mae:{}, rse:{}'.format(mse, mae, rse))\n f = open(\"result.txt\", 'a')\n f.write(setting + \" \\n\")\n f.write('mse:{}, mae:{}, rse:{}'.format(mse, mae, rse))\n f.write('\\n')\n f.write('\\n')\n f.close()\n temp_df = pd.DataFrame()\n temp_df['Seed']=[self.args.random_seed]\n temp_df['Model']=[self.args.model]\n temp_df['seq_len']=[self.args.seq_len]\n temp_df['label_len']=[self.args.label_len]\n temp_df['pred_len']=[self.args.pred_len]\n temp_df['n1']=[self.args.n1]\n temp_df['n2']=[self.args.n2]\n temp_df['dropout']=[self.args.dropout]\n temp_df['train_epochs']=[self.args.train_epochs]\n temp_df['batch']=[self.args.batch_size]\n temp_df['patience']=[self.args.patience]\n temp_df['LR']=[self.args.learning_rate]\n temp_df['dropout']=[self.args.dropout]\n temp_df['ch_ind']=[self.args.ch_ind]\n temp_df['revin']=[self.args.revin]\n temp_df['e_fact']=[self.args.e_fact]\n temp_df['dconv']=[self.args.dconv]\n\n temp_df['MSE']=[mse]\n temp_df['MAE']=[mae]\n temp_df['residual']=[self.args.residual]\n temp_df['d_state']=[self.args.d_state]\n\n temp_df['checkpoint_path']=[setting]\n\n if not os.path.exists('./csv_results/'+'result_'+self.args.data_path):\n temp_df.to_csv('./csv_results/'+'result_'+self.args.data_path, index=False)\n else:\n result_df=pd.read_csv('./csv_results/'+'result_'+self.args.data_path)\n result_df = pd.concat([result_df,temp_df],ignore_index=True)\n result_df.to_csv('./csv_results/'+'result_'+self.args.data_path, index=False)\n\n # np.save(folder_path + 'metrics.npy', np.array([mae, mse, rmse, mape, mspe,rse, corr]))\n np.save(folder_path + 'pred.npy', preds)\n np.save(folder_path + 'true.npy', trues)\n np.save(folder_path + 'x.npy', inputx)\n return\n\n def predict(self, setting, load=False):\n pred_data, pred_loader = self._get_data(flag='pred')\n\n if load:\n path = os.path.join(self.args.checkpoints, setting)\n best_model_path = path + '/' + 'checkpoint.pth'\n self.model.load_state_dict(torch.load(best_model_path))\n\n preds = []\n\n self.model.eval()\n with torch.no_grad():\n for i, (batch_x, batch_y, batch_x_mark, batch_y_mark) in enumerate(pred_loader):\n batch_x = batch_x.float().to(self.device)\n batch_y = batch_y.float()\n batch_x_mark = batch_x_mark.float().to(self.device)\n batch_y_mark = batch_y_mark.float().to(self.device)\n\n # decoder input\n dec_inp = torch.zeros([batch_y.shape[0], self.args.pred_len, batch_y.shape[2]]).float().to(batch_y.device)\n dec_inp = torch.cat([batch_y[:, :self.args.label_len, :], dec_inp], dim=1).float().to(self.device)\n # encoder - decoder\n if self.args.use_amp:\n with torch.cuda.amp.autocast():\n if 'Machine' in self.args.model:\n outputs = self.model(batch_x)\n \n else:\n if 'Machine' in self.args.model:\n outputs = self.model(batch_x)\n \n pred = outputs.detach().cpu().numpy() # .squeeze()\n preds.append(pred)\n\n preds = np.array(preds)\n preds = preds.reshape(-1, preds.shape[-2], preds.shape[-1])\n\n # result save\n folder_path = './results/' + setting + '/'\n if not os.path.exists(folder_path):\n os.makedirs(folder_path)\n\n np.save(folder_path + 'real_prediction.npy', preds)\n\n return\n" }, { "path": "TimeMachine_supervised/models/TimeMachine.py", "content": "import torch\nfrom mamba_ssm import Mamba\nfrom RevIN.RevIN import RevIN\nclass Model(torch.nn.Module):\n def __init__(self,configs):\n super(Model, self).__init__()\n self.configs=configs\n if self.configs.revin==1:\n self.revin_layer = RevIN(self.configs.enc_in)\n\n self.lin1=torch.nn.Linear(self.configs.seq_len,self.configs.n1)\n self.dropout1=torch.nn.Dropout(self.configs.dropout)\n\n self.lin2=torch.nn.Linear(self.configs.n1,self.configs.n2)\n self.dropout2=torch.nn.Dropout(self.configs.dropout)\n if self.configs.ch_ind==1:\n self.d_model_param1=1\n self.d_model_param2=1\n\n else:\n self.d_model_param1=self.configs.n2\n self.d_model_param2=self.configs.n1\n\n self.mamba1=Mamba(d_model=self.d_model_param1,d_state=self.configs.d_state,d_conv=self.configs.dconv,expand=self.configs.e_fact) \n self.mamba2=Mamba(d_model=self.configs.n2,d_state=self.configs.d_state,d_conv=self.configs.dconv,expand=self.configs.e_fact) \n self.mamba3=Mamba(d_model=self.configs.n1,d_state=self.configs.d_state,d_conv=self.configs.dconv,expand=self.configs.e_fact)\n self.mamba4=Mamba(d_model=self.d_model_param2,d_state=self.configs.d_state,d_conv=self.configs.dconv,expand=self.configs.e_fact)\n\n self.lin3=torch.nn.Linear(self.configs.n2,self.configs.n1)\n self.lin4=torch.nn.Linear(2*self.configs.n1,self.configs.pred_len)\n\n\n\n\n\n def forward(self, x):\n if self.configs.revin==1:\n x=self.revin_layer(x,'norm')\n else:\n means = x.mean(1, keepdim=True).detach()\n x = x - means\n stdev = torch.sqrt(torch.var(x, dim=1, keepdim=True, unbiased=False) + 1e-5)\n x /= stdev\n \n x=torch.permute(x,(0,2,1))\n if self.configs.ch_ind==1:\n x=torch.reshape(x,(x.shape[0]*x.shape[1],1,x.shape[2]))\n\n x=self.lin1(x)\n x_res1=x\n x=self.dropout1(x)\n x3=self.mamba3(x)\n if self.configs.ch_ind==1:\n x4=torch.permute(x,(0,2,1))\n else:\n x4=x\n x4=self.mamba4(x4)\n if self.configs.ch_ind==1:\n x4=torch.permute(x4,(0,2,1))\n\n \n x4=x4+x3\n \n\n x=self.lin2(x)\n x_res2=x\n x=self.dropout2(x)\n \n if self.configs.ch_ind==1:\n x1=torch.permute(x,(0,2,1))\n else:\n x1=x \n x1=self.mamba1(x1)\n if self.configs.ch_ind==1:\n x1=torch.permute(x1,(0,2,1))\n \n x2=self.mamba2(x)\n\n if self.configs.residual==1:\n x=x1+x_res2+x2\n else:\n x=x1+x2\n \n x=self.lin3(x)\n if self.configs.residual==1:\n x=x+x_res1\n \n x=torch.cat([x,x4],dim=2)\n x=self.lin4(x) \n if self.configs.ch_ind==1:\n x=torch.reshape(x,(-1,self.configs.enc_in,self.configs.pred_len))\n \n x=torch.permute(x,(0,2,1))\n if self.configs.revin==1:\n x=self.revin_layer(x,'denorm')\n else:\n x = x * (stdev[:, 0, :].unsqueeze(1).repeat(1, self.configs.pred_len, 1))\n x = x + (means[:, 0, :].unsqueeze(1).repeat(1, self.configs.pred_len, 1))\n \n\n return x" }, { "path": "TimeMachine_supervised/requirements.txt", "content": "numpy\nmatplotlib\npandas\nscikit-learn\ntorch\nmamba-ssm\ncausal-conv1d>=1.2.0" }, { "path": "TimeMachine_supervised/run_longExp.py", "content": "import argparse\nimport os\nimport torch\nfrom exp.exp_main import Exp_Main\nimport random\nimport numpy as np\n\nif __name__ == '__main__':\n parser = argparse.ArgumentParser(description='Time Series Forecasting')\n\n # RANDOM SEED\n parser.add_argument('--random_seed', type=int, default=2021, help='random seed')\n\n # BASIC CONFIG\n parser.add_argument('--is_training', type=int, required=True, default=1, help='status')\n parser.add_argument('--model_id', type=str, required=True, default='test', help='model id')\n parser.add_argument('--model', type=str, required=True, default='Autoformer',\n help='model name, options: [TimeMachine]')\n parser.add_argument('--model_id_name', type=str, required=False, default='custom', help='model id name')\n\n # DATALOADER\n parser.add_argument('--data', type=str, required=True, default='ETTm1', help='dataset type')\n parser.add_argument('--root_path', type=str, default='./data/ETT/', help='root path of the data file')\n parser.add_argument('--data_path', type=str, default='ETTh1.csv', help='data file')\n parser.add_argument('--features', type=str, default='M',\n help='forecasting task, options:[M, S, MS]; M:multivariate predict multivariate, S:univariate predict univariate, MS:multivariate predict univariate')\n parser.add_argument('--target', type=str, default='OT', help='target feature in S or MS task')\n parser.add_argument('--freq', type=str, default='h',\n help='freq for time features encoding, options:[s:secondly, t:minutely, h:hourly, d:daily, b:business days, w:weekly, m:monthly], you can also use more detailed freq like 15min or 3h')\n parser.add_argument('--checkpoints', type=str, default='./checkpoints/', help='location of model checkpoints')\n\n # FORECASTING TASK\n parser.add_argument('--seq_len', type=int, default=96, help='input sequence length')\n parser.add_argument('--label_len', type=int, default=48, help='start token length')\n parser.add_argument('--pred_len', type=int, default=96, help='prediction sequence length')\n parser.add_argument('--n1',type=int,default=256,help='First Embedded representation')\n parser.add_argument('--n2',type=int,default=128,help='Second Embedded representation')\n\n\n # METHOD\n parser.add_argument('--revin', type=int, default=1, help='RevIN; True 1 False 0')\n parser.add_argument('--ch_ind', type=int, default=1, help='Channel Independence; True 1 False 0')\n parser.add_argument('--residual', type=int, default=1, help='Residual Connection; True 1 False 0')\n parser.add_argument('--d_state', type=int, default=256, help='d_state parameter of Mamba')\n parser.add_argument('--dconv', type=int, default=2, help='d_conv parameter of Mamba')\n parser.add_argument('--e_fact', type=int, default=1, help='expand factor parameter of Mamba')\n parser.add_argument('--enc_in', type=int, default=7, help='encoder input size') #Use this hyperparameter as the number of channels\n parser.add_argument('--dropout', type=float, default=0.05, help='dropout')\n parser.add_argument('--embed', type=str, default='timeF',\n help='time features encoding, options:[timeF, fixed, learned]')\n parser.add_argument('--do_predict', action='store_true', help='whether to predict unseen future data')\n \n # OPTIMIZATION\n parser.add_argument('--num_workers', type=int, default=10, help='data loader num workers')\n parser.add_argument('--itr', type=int, default=2, help='experiments times')\n parser.add_argument('--train_epochs', type=int, default=100, help='train epochs')\n parser.add_argument('--batch_size', type=int, default=16, help='batch size of train input data')\n parser.add_argument('--patience', type=int, default=100, help='early stopping patience')\n parser.add_argument('--learning_rate', type=float, default=0.0001, help='optimizer learning rate')\n parser.add_argument('--des', type=str, default='test', help='exp description')\n parser.add_argument('--loss', type=str, default='mse', help='loss function')\n parser.add_argument('--lradj', type=str, default='type3', help='adjust learning rate')\n parser.add_argument('--pct_start', type=float, default=0.3, help='pct_start')\n parser.add_argument('--use_amp', action='store_true', help='use automatic mixed precision training', default=False)\n\n # GPU\n parser.add_argument('--use_gpu', type=bool, default=True, help='use gpu')\n parser.add_argument('--gpu', type=int, default=0, help='gpu')\n parser.add_argument('--use_multi_gpu', action='store_true', help='use multiple gpus', default=False)\n parser.add_argument('--devices', type=str, default='0,1,2,3', help='device ids of multile gpus')\n parser.add_argument('--test_flop', action='store_true', default=False, help='See utils/tools for usage')\n\n args = parser.parse_args()\n\n # random seed\n fix_seed = args.random_seed\n random.seed(fix_seed)\n torch.manual_seed(fix_seed)\n np.random.seed(fix_seed)\n args.model_id_name=args.data_path[:-4]\n\n\n args.use_gpu = True if torch.cuda.is_available() and args.use_gpu else False\n\n if args.use_gpu and args.use_multi_gpu:\n args.dvices = args.devices.replace(' ', '')\n device_ids = args.devices.split(',')\n args.device_ids = [int(id_) for id_ in device_ids]\n args.gpu = args.device_ids[0]\n\n print('Args in experiment:')\n print(args)\n\n Exp = Exp_Main\n\n if args.is_training:\n for ii in range(args.itr):\n # setting record of experiments\n setting = '{}_{}_{}_ft{}_sl{}_ll{}_pl{}_n1{}_n2{}_dr{}_cin{}_rin{}_res{}_dst{}_dconv{}_efact{}'.format(\n args.model_id,\n args.model,\n args.model_id_name,\n args.features,\n args.seq_len,\n args.label_len,\n args.pred_len,\n args.n1,\n args.n2,\n args.dropout,\n args.ch_ind,\n args.revin,\n args.residual,\n args.d_state,\n args.dconv,\n args.e_fact)\n\n exp = Exp(args) # set experiments\n print('>>>>>>>start training : {}>>>>>>>>>>>>>>>>>>>>>>>>>>'.format(setting))\n exp.train(setting)\n\n print('>>>>>>>testing : {}<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<'.format(setting))\n exp.test(setting)\n\n if args.do_predict:\n print('>>>>>>>predicting : {}<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<'.format(setting))\n exp.predict(setting, True)\n\n torch.cuda.empty_cache()\n else:\n ii = 0\n setting = '{}_{}_{}_ft{}_sl{}_ll{}_pl{}_n1{}_n2{}_dr{}_cin{}_rin{}_res{}_dst{}_dconv{}_efact'.format(\n args.model_id,\n args.model,\n args.model_id_name,\n args.features,\n args.seq_len,\n args.label_len,\n args.pred_len,\n args.n1,\n args.n2,\n args.dropout,\n args.ch_ind,\n args.revin,\n args.residual,\n args.d_state,\n args.dconv,\n args.e_fact)\n\n exp = Exp(args) # set experiments\n print('>>>>>>>testing : {}<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<'.format(setting))\n exp.test(setting, test=1)\n torch.cuda.empty_cache()\n \n" }, { "path": "TimeMachine_supervised/scripts/TimeMachine/electricity.sh", "content": "if [ ! -d \"./logs\" ]; then\n mkdir ./logs\nfi\n\nif [ ! -d \"./logs/LongForecasting\" ]; then\n mkdir ./logs/LongForecasting\nfi\nif [ ! -d \"./csv_results\" ]; then\n mkdir ./csv_results\nfi\nif [ ! -d \"./results\" ]; then\n mkdir ./results\nfi\nif [ ! -d \"./test_results\" ]; then\n mkdir ./test_results\nfi\n\nmodel_name=TimeMachine\nroot_path_name=../data/electricity\ndata_path_name=electricity.csv\nmodel_id_name=electricity\ndata_name=custom\n\nrin=1\nrandom_seed=2024\none=96\ntwo=192\nthree=336\nfour=720\nresidual=1\nfc_drop=0.0\ndstate=256\ndconv=2\nfor seq_len in 96\ndo\n for pred_len in 96 192 336 720\n do \n for e_fact in 1\n do\n\n if [ $pred_len -eq $one ]\n then\n n1=512\n n2=128\n fc_drop=0.0\n fi\n if [ $pred_len -eq $two ]\n then\n n1=512\n n2=256\n fc_drop=0.0\n fi\n if [ $pred_len -eq $three ]\n then\n n1=512\n n2=256\n fc_drop=0.4\n fi\n if [ $pred_len -eq $four ]\n then\n n1=512\n n2=8\n fc_drop=0.0\n fi\n python -u run_longExp.py \\\n --random_seed $random_seed \\\n --is_training 1 \\\n --root_path $root_path_name \\\n --data_path $data_path_name \\\n --model_id $model_id_name_$seq_len'_'$pred_len \\\n --model $model_name \\\n --data $data_name \\\n --features M \\\n --seq_len $seq_len \\\n --pred_len $pred_len \\\n --enc_in 321 \\\n --n1 $n1 \\\n --n2 $n2 \\\n --dropout $fc_drop\\\n --revin 1\\\n --ch_ind 0\\\n --residual $residual\\\n --dconv $dconv \\\n --d_state $dstate\\\n --e_fact $e_fact\\\n --des 'Exp' \\\n --lradj '5' \\\n --pct_start 0.2 \\\n --train_epochs 100\\\n --itr 1 --batch_size 16 --learning_rate 0.001 >logs/LongForecasting/$model_name'_'$model_id_name'_'$seq_len'_'$pred_len'_'$n1'_'$n2'_'$fc_drop'_'$rin'_'$residual'_'$dstate'_'$dconv'_'$e_fact.log \n \n done \n done\ndone\n" }, { "path": "TimeMachine_supervised/scripts/TimeMachine/etth1.sh", "content": "if [ ! -d \"./logs\" ]; then\n mkdir ./logs\nfi\n\nif [ ! -d \"./logs/LongForecasting\" ]; then\n mkdir ./logs/LongForecasting\nfi\nif [ ! -d \"./csv_results\" ]; then\n mkdir ./csv_results\nfi\nif [ ! -d \"./results\" ]; then\n mkdir ./results\nfi\nif [ ! -d \"./test_results\" ]; then\n mkdir ./test_results\nfi\nmodel_name=TimeMachine\n\nroot_path_name=../data/ETT-small\ndata_path_name=ETTh1.csv\nmodel_id_name=ETTh1\ndata_name=ETTh1\n\nrin=1\nrandom_seed=2024\none=96\ntwo=192\nthree=336\nfour=720\nresidual=1\nfc_drop=0.7\ndstate=256\ndconv=2\nfor seq_len in 96\ndo\n for pred_len in 96 192 336 720\n do \n for e_fact in 1\n do\n\n if [ $pred_len -eq $one ]\n then\n n1=512\n n2=32\n fi\n if [ $pred_len -eq $two ]\n then\n n1=512\n n2=64\n fi\n if [ $pred_len -eq $three ]\n then\n n1=512\n n2=128\n fi\n if [ $pred_len -eq $four ]\n then\n n1=128\n n2=16\n fi\n python -u run_longExp.py \\\n --random_seed $random_seed \\\n --is_training 1 \\\n --root_path $root_path_name \\\n --data_path $data_path_name \\\n --model_id $model_id_name_$seq_len'_'$pred_len \\\n --model $model_name \\\n --data $data_name \\\n --features M \\\n --seq_len $seq_len \\\n --pred_len $pred_len \\\n --enc_in 7 \\\n --n1 $n1 \\\n --n2 $n2 \\\n --dropout $fc_drop\\\n --revin 1\\\n --ch_ind 1\\\n --residual $residual\\\n --dconv $dconv \\\n --d_state $dstate\\\n --e_fact $e_fact\\\n --des 'Exp' \\\n --train_epochs 100\\\n --itr 1 --batch_size 512 --learning_rate 0.001 >logs/LongForecasting/$model_name'_'$model_id_name'_'$seq_len'_'$pred_len'_'$n1'_'$n2'_'$fc_drop'_'$rin'_'$residual'_'$dstate'_'$dconv'_'$e_fact.log \n \n done \n done\ndone\n" }, { "path": "TimeMachine_supervised/scripts/TimeMachine/etth2.sh", "content": "if [ ! -d \"./logs\" ]; then\n mkdir ./logs\nfi\n\nif [ ! -d \"./logs/LongForecasting\" ]; then\n mkdir ./logs/LongForecasting\nfi\nif [ ! -d \"./csv_results\" ]; then\n mkdir ./csv_results\nfi\nif [ ! -d \"./results\" ]; then\n mkdir ./results\nfi\nif [ ! -d \"./test_results\" ]; then\n mkdir ./test_results\nfi\nmodel_name=TimeMachine\n\nroot_path_name=../data/ETT-small\ndata_path_name=ETTh2.csv\nmodel_id_name=ETTh2\ndata_name=ETTh2\none=96\ntwo=192\nthree=336\nfour=720\nresidual=1\nrin=1\nfc_drop=0.7\ndstate=256\nrandom_seed=2024\ndconv=2\nfor seq_len in 96\ndo\n for pred_len in 96 192 336 720\n do \n for e_fact in 1\n do\n\n if [ $pred_len -eq $one ]\n then\n n1=128\n n2=32\n fi\n if [ $pred_len -eq $two ]\n then\n n1=256\n n2=32\n fi\n if [ $pred_len -eq $three ]\n then\n n1=512\n n2=64\n fi\n if [ $pred_len -eq $four ]\n then\n n1=256\n n2=128\n fi\n python -u run_longExp.py \\\n --random_seed $random_seed \\\n --is_training 1 \\\n --root_path $root_path_name \\\n --data_path $data_path_name \\\n --model_id $model_id_name_$seq_len'_'$pred_len \\\n --model $model_name \\\n --data $data_name \\\n --features M \\\n --seq_len $seq_len \\\n --pred_len $pred_len \\\n --enc_in 7 \\\n --n1 $n1 \\\n --n2 $n2 \\\n --dropout $fc_drop\\\n --revin 1\\\n --ch_ind 1\\\n --d_state $dstate\\\n --dconv $dconv \\\n --residual $residual\\\n --e_fact $e_fact\\\n --des 'Exp' \\\n --train_epochs 100\\\n --itr 1 --batch_size 512 --learning_rate 0.001 >logs/LongForecasting/$model_name'_'$model_id_name'_'$seq_len'_'$pred_len'_'$n1'_'$n2'_'$fc_drop'_'$rin'_'$residual'_'$dstate'_'$dconv'_'$e_fact.log \n \n done \n done\ndone\n" }, { "path": "TimeMachine_supervised/scripts/TimeMachine/ettm1.sh", "content": "if [ ! -d \"./logs\" ]; then\n mkdir ./logs\nfi\n\nif [ ! -d \"./logs/LongForecasting\" ]; then\n mkdir ./logs/LongForecasting\nfi\nif [ ! -d \"./csv_results\" ]; then\n mkdir ./csv_results\nfi\nif [ ! -d \"./results\" ]; then\n mkdir ./results\nfi\nif [ ! -d \"./test_results\" ]; then\n mkdir ./test_results\nfi\nmodel_name=TimeMachine\n\nroot_path_name=../data/ETT-small\ndata_path_name=ETTm1.csv\nmodel_id_name=ETTm1\ndata_name=ETTm1\n\n\nrin=1\nrandom_seed=2024\none=96\ntwo=192\nthree=336\nfour=720\nresidual=1\nfc_drop=0.6\ndstate=256\ndconv=2\nfor seq_len in 96\ndo\n for pred_len in 96 192 336 720\n do \n for e_fact in 1\n do\n\n if [ $pred_len -eq $one ]\n then\n n1=256\n n2=64\n \n fi\n if [ $pred_len -eq $two ]\n then\n n1=512\n n2=32\n \n fi\n if [ $pred_len -eq $three ]\n then\n n1=128\n n2=16\n \n fi\n if [ $pred_len -eq $four ]\n then\n n1=512\n n2=128\n fi\n\n python -u run_longExp.py \\\n --random_seed $random_seed \\\n --is_training 1 \\\n --root_path $root_path_name \\\n --data_path $data_path_name \\\n --model_id $model_id_name_$seq_len'_'$pred_len \\\n --model $model_name \\\n --data $data_name \\\n --features M \\\n --seq_len $seq_len \\\n --pred_len $pred_len \\\n --enc_in 7 \\\n --n1 $n1 \\\n --n2 $n2 \\\n --dropout $fc_drop\\\n --des 'Exp' \\\n --train_epochs 100\\\n --itr 1 --batch_size 1024 --learning_rate 0.001 >logs/LongForecasting/$model_name'_'$model_id_name'_'$seq_len'_'$pred_len'_'$n1'_'$n2'_'$fc_drop'_'$rin'_'$residual'_'$dstate'_'$dconv'_'$e_fact.log\n done \n done\ndone" }, { "path": "TimeMachine_supervised/scripts/TimeMachine/ettm2.sh", "content": "if [ ! -d \"./logs\" ]; then\n mkdir ./logs\nfi\n\nif [ ! -d \"./logs/LongForecasting\" ]; then\n mkdir ./logs/LongForecasting\nfi\nif [ ! -d \"./csv_results\" ]; then\n mkdir ./csv_results\nfi\nif [ ! -d \"./results\" ]; then\n mkdir ./results\nfi\nif [ ! -d \"./test_results\" ]; then\n mkdir ./test_results\nfi\nmodel_name=TimeMachine\n\nroot_path_name=../data/ETT-small\ndata_path_name=ETTm2.csv\nmodel_id_name=ETTm2\ndata_name=ETTm2\n\nrin=1\nrandom_seed=2024\none=96\ntwo=192\nthree=336\nfour=720\nresidual=1\nfc_drop=0.6\ndstate=256\ndconv=2\nfor seq_len in 96\ndo\n for pred_len in 96 192 336 720\n do \n for e_fact in 1\n do\n\n if [ $pred_len -eq $one ]\n then\n n1=256\n n2=32\n fi\n if [ $pred_len -eq $two ]\n then\n n1=128\n n2=64\n fi\n if [ $pred_len -eq $three ]\n then\n n1=256\n n2=128\n fi\n if [ $pred_len -eq $four ]\n then\n n1=256\n n2=128\n fi\n\n python -u run_longExp.py \\\n --random_seed $random_seed \\\n --is_training 1 \\\n --root_path $root_path_name \\\n --data_path $data_path_name \\\n --model_id $model_id_name_$seq_len'_'$pred_len \\\n --model $model_name \\\n --data $data_name \\\n --features M \\\n --seq_len $seq_len \\\n --pred_len $pred_len \\\n --enc_in 7 \\\n --n1 $n1 \\\n --n2 $n2 \\\n --dropout $fc_drop\\\n --des 'Exp' \\\n --train_epochs 100\\\n --itr 1 --batch_size 1024 --learning_rate 0.001 >logs/LongForecasting/$model_name'_'$model_id_name'_'$seq_len'_'$pred_len'_'$n1'_'$n2'_'$fc_drop'_'$rin'_'$residual'_'$dstate'_'$dconv'_'$e_fact.log\n done \n done\ndone" }, { "path": "TimeMachine_supervised/scripts/TimeMachine/traffic.sh", "content": "if [ ! -d \"./logs\" ]; then\n mkdir ./logs\nfi\n\nif [ ! -d \"./logs/LongForecasting\" ]; then\n mkdir ./logs/LongForecasting\nfi\nif [ ! -d \"./csv_results\" ]; then\n mkdir ./csv_results\nfi\nif [ ! -d \"./results\" ]; then\n mkdir ./results\nfi\nif [ ! -d \"./test_results\" ]; then\n mkdir ./test_results\nfi\n\n\nmodel_name=TimeMachine\nroot_path_name=../data/traffic\ndata_path_name=traffic.csv\nmodel_id_name=traffic\ndata_name=custom\n\nrin=0\nrandom_seed=2024\none=96\ntwo=192\nthree=336\nfour=720\nresidual=1\nfc_drop=0.3\ndstate=256\ndconv=2\nfor seq_len in 96\ndo\n for pred_len in 96 192 336 720\n do \n for e_fact in 1\n do\n\n if [ $pred_len -eq $one ]\n then\n n1=512\n n2=16\n fc_drop=0.3\n fi\n if [ $pred_len -eq $two ]\n then\n n1=512\n n2=256\n fc_drop=0.1\n fi\n if [ $pred_len -eq $three ]\n then\n n1=512\n n2=256\n fc_drop=0.1\n fi\n if [ $pred_len -eq $four ]\n then\n n1=512\n n2=128\n fc_drop=0.1\n fi\n python -u run_longExp.py \\\n --random_seed $random_seed \\\n --is_training 1 \\\n --root_path $root_path_name \\\n --data_path $data_path_name \\\n --model_id $model_id_name_$seq_len'_'$pred_len \\\n --model $model_name \\\n --data $data_name \\\n --features M \\\n --seq_len $seq_len \\\n --pred_len $pred_len \\\n --enc_in 862 \\\n --n1 $n1 \\\n --n2 $n2 \\\n --dropout $fc_drop \\\n --revin 0 \\\n --ch_ind 0 \\\n --residual $residual \\\n --dconv $dconv \\\n --d_state $dstate \\\n --e_fact $e_fact \\\n --des 'Exp' \\\n --lradj '5' \\\n --pct_start 0.2 \\\n --train_epochs 100 \\\n --itr 1 --batch_size 16 --learning_rate 0.001 >logs/LongForecasting/$model_name'_'$model_id_name'_'$seq_len'_'$pred_len'_'$n1'_'$n2'_'$fc_drop'_'$rin'_'$residual'_'$dstate'_'$dconv'_'$e_fact.log \n \n done \n done\ndone\n" }, { "path": "TimeMachine_supervised/scripts/TimeMachine/weather.sh", "content": "if [ ! -d \"./logs\" ]; then\n mkdir ./logs\nfi\n\nif [ ! -d \"./logs/LongForecasting\" ]; then\n mkdir ./logs/LongForecasting\nfi\nif [ ! -d \"./csv_results\" ]; then\n mkdir ./csv_results\nfi\nif [ ! -d \"./results\" ]; then\n mkdir ./results\nfi\nif [ ! -d \"./test_results\" ]; then\n mkdir ./test_results\nfi\nmodel_name=TimeMachine\n\nroot_path_name=../data/weather\ndata_path_name=weather.csv\nmodel_id_name=weather\ndata_name=custom\n\nrin=1\nrandom_seed=2024\none=96\ntwo=192\nthree=336\nfour=720\nresidual=1\nfc_drop=0.1\ndstate=256\ndconv=2\nfor seq_len in 96\ndo\n for pred_len in 96 192 336 720\n do \n for e_fact in 1\n do\n\n if [ $pred_len -eq $one ]\n then\n n1=128\n n2=16\n fc_drop=0.1\n fi\n if [ $pred_len -eq $two ]\n then\n n1=512\n n2=128\n fc_drop=0.5\n fi\n if [ $pred_len -eq $three ]\n then\n n1=128\n n2=64\n fc_drop=0.0\n fi\n if [ $pred_len -eq $four ]\n then\n n1=512\n n2=256\n fc_drop=0.0\n fi\n python -u run_longExp.py \\\n --random_seed $random_seed \\\n --is_training 1 \\\n --root_path $root_path_name \\\n --data_path $data_path_name \\\n --model_id $model_id_name_$seq_len'_'$pred_len \\\n --model $model_name \\\n --data $data_name \\\n --features M \\\n --seq_len $seq_len \\\n --pred_len $pred_len \\\n --enc_in 21 \\\n --n1 $n1 \\\n --n2 $n2 \\\n --dropout $fc_drop\\\n --revin 1\\\n --ch_ind 1\\\n --residual $residual\\\n --dconv $dconv \\\n --d_state $dstate\\\n --e_fact $e_fact\\\n --des 'Exp' \\\n --lradj 'constant'\\\n --pct_start 0.2\\\n --itr 1 --batch_size 512 --learning_rate 0.001 >logs/LongForecasting/$model_name'_'$model_id_name'_'$seq_len'_'$pred_len'_'$n1'_'$n2'_'$fc_drop'_'$rin'_'$residual'_'$dstate'_'$dconv'_'$e_fact.log \n \n done \n done\ndone\n" }, { "path": "TimeMachine_supervised/utils/masking.py", "content": "import torch\n\n\nclass TriangularCausalMask():\n def __init__(self, B, L, device=\"cpu\"):\n mask_shape = [B, 1, L, L]\n with torch.no_grad():\n self._mask = torch.triu(torch.ones(mask_shape, dtype=torch.bool), diagonal=1).to(device)\n\n @property\n def mask(self):\n return self._mask\n\n\nclass ProbMask():\n def __init__(self, B, H, L, index, scores, device=\"cpu\"):\n _mask = torch.ones(L, scores.shape[-1], dtype=torch.bool).to(device).triu(1)\n _mask_ex = _mask[None, None, :].expand(B, H, L, scores.shape[-1])\n indicator = _mask_ex[torch.arange(B)[:, None, None],\n torch.arange(H)[None, :, None],\n index, :].to(device)\n self._mask = indicator.view(scores.shape).to(device)\n\n @property\n def mask(self):\n return self._mask\n" }, { "path": "TimeMachine_supervised/utils/metrics.py", "content": "import numpy as np\n\n\ndef RSE(pred, true):\n return np.sqrt(np.sum((true - pred) ** 2)) / np.sqrt(np.sum((true - true.mean()) ** 2))\n\n\ndef CORR(pred, true):\n u = ((true - true.mean(0)) * (pred - pred.mean(0))).sum(0)\n d = np.sqrt(((true - true.mean(0)) ** 2 * (pred - pred.mean(0)) ** 2).sum(0))\n d += 1e-12\n return 0.01*(u / d).mean(-1)\n\n\ndef MAE(pred, true):\n return np.mean(np.abs(pred - true))\n\n\ndef MSE(pred, true):\n return np.mean((pred - true) ** 2)\n\n\ndef RMSE(pred, true):\n return np.sqrt(MSE(pred, true))\n\n\ndef MAPE(pred, true):\n return np.mean(np.abs((pred - true) / true))\n\n\ndef MSPE(pred, true):\n return np.mean(np.square((pred - true) / true))\n\n\ndef metric(pred, true):\n mae = MAE(pred, true)\n mse = MSE(pred, true)\n rmse = RMSE(pred, true)\n mape = MAPE(pred, true)\n mspe = MSPE(pred, true)\n rse = RSE(pred, true)\n corr = CORR(pred, true)\n\n return mae, mse, rmse, mape, mspe, rse, corr\n" }, { "path": "TimeMachine_supervised/utils/timefeatures.py", "content": "from typing import List\n\nimport numpy as np\nimport pandas as pd\nfrom pandas.tseries import offsets\nfrom pandas.tseries.frequencies import to_offset\n\n\nclass TimeFeature:\n def __init__(self):\n pass\n\n def __call__(self, index: pd.DatetimeIndex) -> np.ndarray:\n pass\n\n def __repr__(self):\n return self.__class__.__name__ + \"()\"\n\n\nclass SecondOfMinute(TimeFeature):\n \"\"\"Minute of hour encoded as value between [-0.5, 0.5]\"\"\"\n\n def __call__(self, index: pd.DatetimeIndex) -> np.ndarray:\n return index.second / 59.0 - 0.5\n\n\nclass MinuteOfHour(TimeFeature):\n \"\"\"Minute of hour encoded as value between [-0.5, 0.5]\"\"\"\n\n def __call__(self, index: pd.DatetimeIndex) -> np.ndarray:\n return index.minute / 59.0 - 0.5\n\n\nclass HourOfDay(TimeFeature):\n \"\"\"Hour of day encoded as value between [-0.5, 0.5]\"\"\"\n\n def __call__(self, index: pd.DatetimeIndex) -> np.ndarray:\n return index.hour / 23.0 - 0.5\n\n\nclass DayOfWeek(TimeFeature):\n \"\"\"Hour of day encoded as value between [-0.5, 0.5]\"\"\"\n\n def __call__(self, index: pd.DatetimeIndex) -> np.ndarray:\n return index.dayofweek / 6.0 - 0.5\n\n\nclass DayOfMonth(TimeFeature):\n \"\"\"Day of month encoded as value between [-0.5, 0.5]\"\"\"\n\n def __call__(self, index: pd.DatetimeIndex) -> np.ndarray:\n return (index.day - 1) / 30.0 - 0.5\n\n\nclass DayOfYear(TimeFeature):\n \"\"\"Day of year encoded as value between [-0.5, 0.5]\"\"\"\n\n def __call__(self, index: pd.DatetimeIndex) -> np.ndarray:\n return (index.dayofyear - 1) / 365.0 - 0.5\n\n\nclass MonthOfYear(TimeFeature):\n \"\"\"Month of year encoded as value between [-0.5, 0.5]\"\"\"\n\n def __call__(self, index: pd.DatetimeIndex) -> np.ndarray:\n return (index.month - 1) / 11.0 - 0.5\n\n\nclass WeekOfYear(TimeFeature):\n \"\"\"Week of year encoded as value between [-0.5, 0.5]\"\"\"\n\n def __call__(self, index: pd.DatetimeIndex) -> np.ndarray:\n return (index.isocalendar().week - 1) / 52.0 - 0.5\n\n\ndef time_features_from_frequency_str(freq_str: str) -> List[TimeFeature]:\n \"\"\"\n Returns a list of time features that will be appropriate for the given frequency string.\n Parameters\n ----------\n freq_str\n Frequency string of the form [multiple][granularity] such as \"12H\", \"5min\", \"1D\" etc.\n \"\"\"\n\n features_by_offsets = {\n offsets.YearEnd: [],\n offsets.QuarterEnd: [MonthOfYear],\n offsets.MonthEnd: [MonthOfYear],\n offsets.Week: [DayOfMonth, WeekOfYear],\n offsets.Day: [DayOfWeek, DayOfMonth, DayOfYear],\n offsets.BusinessDay: [DayOfWeek, DayOfMonth, DayOfYear],\n offsets.Hour: [HourOfDay, DayOfWeek, DayOfMonth, DayOfYear],\n offsets.Minute: [\n MinuteOfHour,\n HourOfDay,\n DayOfWeek,\n DayOfMonth,\n DayOfYear,\n ],\n offsets.Second: [\n SecondOfMinute,\n MinuteOfHour,\n HourOfDay,\n DayOfWeek,\n DayOfMonth,\n DayOfYear,\n ],\n }\n\n offset = to_offset(freq_str)\n\n for offset_type, feature_classes in features_by_offsets.items():\n if isinstance(offset, offset_type):\n return [cls() for cls in feature_classes]\n\n supported_freq_msg = f\"\"\"\n Unsupported frequency {freq_str}\n The following frequencies are supported:\n Y - yearly\n alias: A\n M - monthly\n W - weekly\n D - daily\n B - business days\n H - hourly\n T - minutely\n alias: min\n S - secondly\n \"\"\"\n raise RuntimeError(supported_freq_msg)\n\n\ndef time_features(dates, freq='h'):\n return np.vstack([feat(dates) for feat in time_features_from_frequency_str(freq)])\n" }, { "path": "TimeMachine_supervised/utils/tools.py", "content": "import numpy as np\nimport torch\nimport matplotlib.pyplot as plt\nimport time\n\nplt.switch_backend('agg')\n\n\ndef adjust_learning_rate(optimizer, scheduler, epoch, args, printout=True):\n # lr = args.learning_rate * (0.2 ** (epoch // 2))\n if args.lradj == 'type1':\n lr_adjust = {epoch: args.learning_rate * (0.5 ** ((epoch - 1) // 1))}\n elif args.lradj == 'type2':\n lr_adjust = {\n 2: 5e-5, 4: 1e-5, 6: 5e-6, 8: 1e-6,\n 10: 5e-7, 15: 1e-7, 20: 5e-8\n }\n elif args.lradj == 'type3':\n lr_adjust = {epoch: args.learning_rate if epoch < 3 else args.learning_rate * (0.9 ** ((epoch - 3) // 1))}\n elif args.lradj == 'constant':\n lr_adjust = {epoch: args.learning_rate}\n elif args.lradj == '3':\n lr_adjust = {epoch: args.learning_rate if epoch < 10 else args.learning_rate*0.1}\n elif args.lradj == '4':\n lr_adjust = {epoch: args.learning_rate if epoch < 15 else args.learning_rate*0.1}\n elif args.lradj == '5':\n lr_adjust = {epoch: args.learning_rate if epoch < 25 else args.learning_rate*0.1}\n elif args.lradj == '6':\n lr_adjust = {epoch: args.learning_rate if epoch < 5 else args.learning_rate*0.1} \n elif args.lradj == 'TST':\n lr_adjust = {epoch: scheduler.get_last_lr()[0]}\n \n if epoch in lr_adjust.keys():\n lr = lr_adjust[epoch]\n for param_group in optimizer.param_groups:\n param_group['lr'] = lr\n if printout: print('Updating learning rate to {}'.format(lr))\n\n\nclass EarlyStopping:\n def __init__(self, patience=7, verbose=False, delta=0):\n self.patience = patience\n self.verbose = verbose\n self.counter = 0\n self.best_score = None\n self.early_stop = False\n self.val_loss_min = np.Inf\n self.delta = delta\n\n def __call__(self, val_loss, model, path):\n score = -val_loss\n if self.best_score is None:\n self.best_score = score\n self.save_checkpoint(val_loss, model, path)\n elif score < self.best_score + self.delta:\n self.counter += 1\n print(f'EarlyStopping counter: {self.counter} out of {self.patience}')\n if self.counter >= self.patience:\n self.early_stop = True\n else:\n self.best_score = score\n self.save_checkpoint(val_loss, model, path)\n self.counter = 0\n\n def save_checkpoint(self, val_loss, model, path):\n if self.verbose:\n print(f'Validation loss decreased ({self.val_loss_min:.6f} --> {val_loss:.6f}). Saving model ...')\n torch.save(model.state_dict(), path + '/' + 'checkpoint.pth')\n self.val_loss_min = val_loss\n\n\nclass dotdict(dict):\n \"\"\"dot.notation access to dictionary attributes\"\"\"\n __getattr__ = dict.get\n __setattr__ = dict.__setitem__\n __delattr__ = dict.__delitem__\n\n\nclass StandardScaler():\n def __init__(self, mean, std):\n self.mean = mean\n self.std = std\n\n def transform(self, data):\n return (data - self.mean) / self.std\n\n def inverse_transform(self, data):\n return (data * self.std) + self.mean\n\n\ndef visual(true, preds=None, name='./pic/test.pdf'):\n \"\"\"\n Results visualization\n \"\"\"\n plt.figure()\n plt.plot(true, label='GroundTruth', linewidth=2)\n if preds is not None:\n plt.plot(preds, label='Prediction', linewidth=2)\n plt.legend()\n plt.savefig(name, bbox_inches='tight')\n\ndef test_params_flop(model,x_shape):\n \"\"\"\n If you want to thest former's flop, you need to give default value to inputs in model.forward(), the following code can only pass one argument to forward()\n \"\"\"\n model_params = 0\n for parameter in model.parameters():\n model_params += parameter.numel()\n print('INFO: Trainable parameter count: {:.2f}M'.format(model_params / 1000000.0))\n from ptflops import get_model_complexity_info \n with torch.cuda.device(0):\n macs, params = get_model_complexity_info(model.cuda(), x_shape, as_strings=True, print_per_layer_stat=True)\n # print('Flops:' + flops)\n # print('Params:' + params)\n print('{:<30} {:<8}'.format('Computational complexity: ', macs))\n print('{:<30} {:<8}'.format('Number of parameters: ', params))" } ]