v1.1.0

.github/workflows/python-test.yml

@@ -5,9 +5,9 @@ name: Pytest
 
 on:
   push:
-    branches: [ "master" ]
+    branches: [ "master", "release/*" ]
   pull_request:
-    branches: [ "master" ]
+    branches: [ "master", "release/*" ]
 
 permissions:
   contents: read
@@ -29,6 +29,7 @@ jobs:
         pip install flake8 pytest
         pip install torch torchvision --index-url https://download.pytorch.org/whl/cpu
         if [ -f requirements.txt ]; then pip install -r requirements.txt; fi
+        if [ -f tests/requirements.txt ]; then pip install -r tests/requirements.txt; fi
     - name: Test with pytest
       run: |
         pytest

.isort.cfg

@@ -1,3 +1,3 @@
 [settings]
-src_paths=basicts,tests
-skip_glob=baselines/*,assets/*,examples/*
+src_paths=src/basicts,tests
+skip_glob=baselines/*,assets/*

.pylintrc

@@ -12,7 +12,7 @@ ignore=baselines,assets,checkpoints,examples,scripts
 
 # Files or directories matching the regex patterns are skipped. The regex
 # matches against base names, not paths.
-ignore-patterns=^\.|^_|^.*\.md|^.*\.txt|^.*\.csv|^.*\.CFF|^LICENSE
+ignore-patterns=^\.|^_|^.*\.md|^.*\.txt|^.*\.csv|^.*\.CFF|^LICENSE|^.*\.toml
 
 # Pickle collected data for later comparisons.
 persistent=no

examples/classification/classification_demo.py

@@ -1,6 +1,7 @@
 from basicts import BasicTSLauncher
 from basicts.configs import BasicTSClassificationConfig
-from basicts.models.iTransformer import iTransformerForClassification, iTransformerConfig
+from basicts.models.iTransformer import (iTransformerConfig,
+                                         iTransformerForClassification)
 
 
 def main():

examples/forecasting/forecasting_demo.py

@@ -1,9 +1,11 @@
+from torch.optim.lr_scheduler import MultiStepLR
+
 from basicts import BasicTSLauncher
 from basicts.configs import BasicTSForecastingConfig
-from basicts.models.iTransformer import iTransformerForForecasting, iTransformerConfig
-from basicts.runners.callback import EarlyStopping, GradientClipping
 from basicts.metrics import masked_mse
-from torch.optim.lr_scheduler import MultiStepLR
+from basicts.models.iTransformer import (iTransformerConfig,
+                                         iTransformerForForecasting)
+from basicts.runners.callback import EarlyStopping, GradientClipping
 
 
 def main():

examples/imputation/imputation_demo.py

@@ -1,6 +1,7 @@
 from basicts import BasicTSLauncher
 from basicts.configs import BasicTSImputationConfig
-from basicts.models.iTransformer import iTransformerForReconstruction, iTransformerConfig
+from basicts.models.iTransformer import (iTransformerConfig,
+                                         iTransformerForReconstruction)
 
 
 def main():

pyproject.toml

@@ -18,7 +18,7 @@ dependencies = [
     "sympy",
     "openpyxl",
     "setuptools==59.5.0",
-    "numpy==1.24.4",
+    "numpy",
     "tqdm==4.67.1",
     "tensorboard==2.18.0",
     "transformers==4.40.1"

src/basicts/__init__.py

@@ -1,5 +1,5 @@
 from .launcher import BasicTSLauncher
 
-__version__ = '1.0.2'
+__version__ = '1.1.0'
 
 __all__ = ['__version__', 'BasicTSLauncher']

src/basicts/configs/base_config.py

@@ -9,17 +9,22 @@
 from functools import partial
 from numbers import Number
 from types import FunctionType
-from typing import Callable, List, Literal, Optional, Tuple, Union
+from typing import (TYPE_CHECKING, Callable, List, Literal, Optional, Tuple,
+                    Union)
 
 import numpy as np
 import torch
-from basicts.runners.callback import BasicTSCallback
-from basicts.runners.taskflow import BasicTSTaskFlow
 from easydict import EasyDict
 from torch.optim.lr_scheduler import LRScheduler
 
 from .model_config import BasicTSModelConfig
 
+# avoid circular imports
+if TYPE_CHECKING:
+    from basicts.runners.callback import BasicTSCallback
+    from basicts.runners.taskflow import BasicTSTaskFlow
+
+
 
 @dataclass(init=False)
 class BasicTSConfig(EasyDict):
@@ -35,8 +40,8 @@ class BasicTSConfig(EasyDict):
     model_config: BasicTSModelConfig
 
     dataset_name: str
-    taskflow: BasicTSTaskFlow
-    callbacks: List[BasicTSCallback]
+    taskflow: "BasicTSTaskFlow"
+    callbacks: List["BasicTSCallback"]
 
     ############################## General Configuration ##############################
 
@@ -277,7 +282,7 @@ def _pack_params(self, obj: type, obj_params: Union[dict, None]) -> dict:
             elif issubclass(obj, LRScheduler) and k == "optimizer":
                 continue
             # short cut has higher priority than params in config
-            elif k in self:
+            elif k in self and self[k] is not None:
                 obj_params[k] = self[k]
         return obj_params
 
@@ -338,7 +343,7 @@ def _serialize_obj(self, obj: object) -> object:
                     if not isinstance(is_default, bool):
                         raise ValueError(f"Parameter {k} of {obj.__class__.__name__} is not serializable.")
                     if not is_default:
-                        params[k] = repr(v)
+                        params[k] = self._serialize_obj(v)
 
             return {
                 "name": obj.__class__.__name__,

src/basicts/configs/tsc_config.py

@@ -2,12 +2,13 @@
 from typing import Callable, List, Literal, Tuple, Union
 
 import numpy as np
+from torch.nn import CrossEntropyLoss
+from torch.optim import Adam
+
 from basicts.data import UEADataset
 from basicts.runners.callback import BasicTSCallback
 from basicts.runners.taskflow import (BasicTSClassificationTaskFlow,
                                       BasicTSTaskFlow)
-from torch.nn import CrossEntropyLoss
-from torch.optim import Adam
 
 from .base_config import BasicTSConfig
 from .model_config import BasicTSModelConfig
@@ -99,9 +100,11 @@ class BasicTSClassificationConfig(BasicTSConfig):
 
     # Dataset settings
     dataset_type: type = field(default=UEADataset, metadata={"help": "Dataset type."})
-    dataset_params: Union[dict, None] = field(default=None, metadata={"help": "Dataset parameters."})
+    dataset_params: Union[dict, None] = field(
+        default_factory=lambda: {"memmap": False},
+        metadata={"help": "Dataset parameters."})
     use_timestamps: bool = field(default=False, metadata={"help": "Whether to use timestamps as supplementary."})
-    memmap: bool = field(default=False, metadata={"help": "Whether to use memmap to load datasets."})
+    memmap: bool = field(default=None, metadata={"help": "Whether to use memmap to load datasets."})
     null_val: float = field(default=np.nan, metadata={"help": "Null value."})
     null_to_num: float = field(default=0.0, metadata={"help": "Null value to number."})
 
@@ -148,7 +151,7 @@ class BasicTSClassificationConfig(BasicTSConfig):
     optimizer_params: dict = field(
         default_factory=lambda: {"lr": 2e-4, "weight_decay": 5e-4},
         metadata={"help": "Optimizer parameters."})
-    lr: float = field(default=2e-4, metadata={"help": "Learning rate."})
+    lr: float = field(default=None, metadata={"help": "Learning rate."})
 
     # Learning rate scheduler
     lr_scheduler: Union[type, None] = field(default=None, metadata={"help": "Learning rate scheduler type."})

src/basicts/configs/tsf_config.py

@@ -2,12 +2,13 @@
 from typing import Callable, List, Literal, Tuple, Union
 
 import numpy as np
+from torch.optim import Adam
+
 from basicts.data import BasicTSForecastingDataset
 from basicts.runners.callback import BasicTSCallback
 from basicts.runners.taskflow import (BasicTSForecastingTaskFlow,
                                       BasicTSTaskFlow)
 from basicts.scaler import ZScoreScaler
-from torch.optim import Adam
 
 from .base_config import BasicTSConfig
 from .model_config import BasicTSModelConfig
@@ -99,17 +100,26 @@ class BasicTSForecastingConfig(BasicTSConfig):
 
     # Dataset settings
     dataset_type: type = field(default=BasicTSForecastingDataset, metadata={"help": "Dataset type."})
-    dataset_params: Union[dict, None] = field(default=None, metadata={"help": "Dataset parameters."})
-    input_len: int = field(default=336, metadata={"help": "Input length."})
-    output_len: int = field(default=336, metadata={"help": "Output length."})
-    use_timestamps: bool = field(default=True, metadata={"help": "Whether to use timestamps as supplementary."})
-    memmap: bool = field(default=False, metadata={"help": "Whether to use memmap to load datasets."})
-    null_val: float = field(default=np.nan, metadata={"help": "Null value."})
-    null_to_num: float = field(default=0.0, metadata={"help": "Null value to number."})
-
+    dataset_params: Union[dict, None] = field(
+        default_factory=lambda: {
+            "input_len": 336,
+            "output_len": 336,
+            "use_timestamps": True,
+            "memmap": False,
+        }, metadata={"help": "Dataset parameters."})
+
+    # shortcuts
+    input_len: int = field(default=None, metadata={"help": "Input length."})
+    output_len: int = field(default=None, metadata={"help": "Output length."})
+    use_timestamps: bool = field(default=None, metadata={"help": "Whether to use timestamps as supplementary."})
+    memmap: bool = field(default=None, metadata={"help": "Whether to use memmap to load datasets."})
     batch_size: Union[int, None] = field(
         default=None, metadata={"help": "Batch size. If setted, all dataloaders will be setted to the same batch size."})
 
+
+    null_val: float = field(default=np.nan, metadata={"help": "Null value."})
+    null_to_num: float = field(default=0.0, metadata={"help": "Null value to number."})
+
     # Scaler settings
     scaler: type = field(default=ZScoreScaler, metadata={"help": "Scaler type."})
     norm_each_channel: bool = field(default=True, metadata={"help": "Whether to normalize data for each channel independently."})
@@ -147,7 +157,7 @@ class BasicTSForecastingConfig(BasicTSConfig):
     optimizer_params: dict = field(
         default_factory=lambda: {"lr": 2e-4, "weight_decay": 5e-4},
         metadata={"help": "Optimizer parameters."})
-    lr: float = field(default=2e-4, metadata={"help": "Learning rate."})
+    lr: float = field(default=None, metadata={"help": "Learning rate."})
 
     # Learning rate scheduler
     lr_scheduler: Union[type, None] = field(default=None, metadata={"help": "Learning rate scheduler type."})

src/basicts/configs/tsfm_config.py

@@ -3,12 +3,13 @@
 
 import numpy as np
 import torch
+from torch.optim import AdamW
+
 from basicts.data import BasicTSForecastingDataset
 from basicts.runners.callback import BasicTSCallback
 from basicts.runners.optim.lr_schedulers import CosineWarmup
 from basicts.runners.taskflow import (BasicTSForecastingTaskFlow,
                                       BasicTSTaskFlow)
-from torch.optim import AdamW
 
 from .base_config import BasicTSConfig
 from .model_config import BasicTSModelConfig
@@ -100,10 +101,10 @@ class BasicTSFoundationModelConfig(BasicTSConfig):
     # Dataset settings
     dataset_type: type = field(default=BasicTSForecastingDataset, metadata={"help": "Dataset type."})
     dataset_params: dict = field(default_factory=dict)
-    input_len: int = field(default=336, metadata={"help": "Input length."})
-    output_len: int = field(default=336, metadata={"help": "Output length."})
-    use_timestamps: bool = field(default=False, metadata={"help": "Whether to use timestamps as supplementary."})
-    memmap: bool = field(default=False, metadata={"help": "Whether to use memmap to load datasets."})
+    input_len: int = field(default=None, metadata={"help": "Input length."})
+    output_len: int = field(default=None, metadata={"help": "Output length."})
+    use_timestamps: bool = field(default=None, metadata={"help": "Whether to use timestamps as supplementary."})
+    memmap: bool = field(default=None, metadata={"help": "Whether to use memmap to load datasets."})
     batch_size: Optional[int] = field(default=None, metadata={"help": "Batch size. If setted, all dataloaders will be setted to the same batch size."})
     null_val: float = field(default=np.nan, metadata={"help": "Null value."})
     null_to_num: float = field(default=0.0, metadata={"help": "Null value to number."})
@@ -142,7 +143,7 @@ class BasicTSFoundationModelConfig(BasicTSConfig):
     # Optimizer
     optimizer: type = field(default=AdamW)
     optimizer_params: dict = field(default_factory=lambda: {"lr": 1e-3, "fused": True})
-    lr: float = field(default=1e-3, metadata={"help": "Learning rate."})
+    lr: float = field(default=None, metadata={"help": "Learning rate."})
 
     # Learning rate scheduler
     lr_scheduler: type = field(default=CosineWarmup)

src/basicts/configs/tsi_config.py

@@ -2,11 +2,12 @@
 from typing import Callable, List, Literal, Tuple, Union
 
 import numpy as np
+from torch.optim import Adam
+
 from basicts.data import BasicTSImputationDataset
 from basicts.runners.callback import BasicTSCallback
 from basicts.runners.taskflow import BasicTSImputationTaskFlow, BasicTSTaskFlow
 from basicts.scaler import ZScoreScaler
-from torch.optim import Adam
 
 from .base_config import BasicTSConfig
 from .model_config import BasicTSModelConfig
@@ -98,17 +99,24 @@ class BasicTSImputationConfig(BasicTSConfig):
 
     # Dataset settings
     dataset_type: type = field(default=BasicTSImputationDataset, metadata={"help": "Dataset type."})
-    dataset_params: Union[dict, None] = field(default=None, metadata={"help": "Dataset parameters."})
-    input_len: int = field(default=336, metadata={"help": "Input length."})
+    dataset_params: Union[dict, None] = field(
+        default_factory=lambda: {
+            "input_len": 336,
+            "use_timestamps": True,
+            "memmap": False,
+        }, metadata={"help": "Dataset parameters."})
+
+    # shortcuts
+    input_len: int = field(default=None, metadata={"help": "Input length."})
+    use_timestamps: bool = field(default=None, metadata={"help": "Whether to use timestamps as supplementary."})
+    memmap: bool = field(default=None, metadata={"help": "Whether to use memmap to load datasets."})
+    batch_size: Union[int, None] = field(
+        default=None, metadata={"help": "Batch size. If setted, all dataloaders will be setted to the same batch size."})
+
     mask_ratio: float = field(default=0.25, metadata={"help": "Mask ratio."})
-    use_timestamps: bool = field(default=True, metadata={"help": "Whether to use timestamps as supplementary."})
-    memmap: bool = field(default=False, metadata={"help": "Whether to use memmap to load datasets."})
     null_val: float = field(default=np.nan, metadata={"help": "Null value."})
     null_to_num: float = field(default=0.0, metadata={"help": "Null value to number."})
 
-    batch_size: Union[int, None] = field(
-        default=None, metadata={"help": "Batch size. If setted, all dataloaders will be setted to the same batch size."})
-
     # Scaler settings
     scaler: type = field(default=ZScoreScaler, metadata={"help": "Scaler type."})
     norm_each_channel: bool = field(default=True, metadata={"help": "Whether to normalize data for each channel independently."})
@@ -146,7 +154,7 @@ class BasicTSImputationConfig(BasicTSConfig):
     optimizer_params: dict = field(
         default_factory=lambda: {"lr": 2e-4, "weight_decay": 5e-4},
         metadata={"help": "Optimizer parameters."})
-    lr: float = field(default=2e-4, metadata={"help": "Learning rate."})
+    lr: float = field(default=None, metadata={"help": "Learning rate."})
 
     # Learning rate scheduler
     lr_scheduler: Union[type, None] = field(default=None, metadata={"help": "Learning rate scheduler type."})

src/basicts/data/base_dataset.py

@@ -1,9 +1,10 @@
 from typing import Union
 
 import numpy as np
-from basicts.utils.constants import BasicTSMode
 from torch.utils.data import Dataset
 
+from basicts.utils.constants import BasicTSMode
+
 
 class BasicTSDataset(Dataset):
     """

src/basicts/data/blast.py

@@ -3,6 +3,7 @@
 from typing import Optional, Union
 
 import numpy as np
+
 from basicts.utils.constants import BasicTSMode
 
 from .base_dataset import BasicTSDataset
@@ -56,7 +57,7 @@ class BLAST(BasicTSDataset):
 
     def __post_init__(self):
         # load data
-        self.data = self._load_data()
+        self._data = self._load_data()
         self.output_len = self.output_len or 0
 
         # minimum valid history sequence length
@@ -244,6 +245,15 @@ def __getitem__(self, idx: int) -> tuple:
     def __len__(self):
         return self.data.shape[0]
 
+    def __getstate__(self):
+        state = self.__dict__.copy()
+        del state["_data"]
+        return state
+
+    def __setstate__(self, state):
+        self.__dict__.update(state)
+        self._data = self._load_data()
+
     @property
     def data(self) -> np.ndarray:
         return self._data

src/basicts/data/tsf_dataset.py

@@ -2,6 +2,7 @@
 from typing import Union
 
 import numpy as np
+
 from basicts.utils.constants import BasicTSMode
 
 from .base_dataset import BasicTSDataset

src/basicts/data/tsi_dataset.py

@@ -2,6 +2,7 @@
 from typing import Union
 
 import numpy as np
+
 from basicts.utils.constants import BasicTSMode
 
 from .base_dataset import BasicTSDataset