oodeel 0.4.0__py3-none-any.whl

oodeel/datasets/torch_data_handler.py

@@ -0,0 +1,672 @@
+# -*- coding: utf-8 -*-
+# Copyright IRT Antoine de Saint Exupéry et Université Paul Sabatier Toulouse III - All
+# rights reserved. DEEL is a research program operated by IVADO, IRT Saint Exupéry,
+# CRIAQ and ANITI - https://www.deel.ai/
+#
+# Permission is hereby granted, free of charge, to any person obtaining a copy
+# of this software and associated documentation files (the "Software"), to deal
+# in the Software without restriction, including without limitation the rights
+# to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
+# copies of the Software, and to permit persons to whom the Software is
+# furnished to do so, subject to the following conditions:
+#
+# The above copyright notice and this permission notice shall be included in all
+# copies or substantial portions of the Software.
+#
+# THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+# IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+# FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+# AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+# LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+# OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+# SOFTWARE.
+import copy
+from typing import get_args
+
+import numpy as np
+import torch
+import torch.nn.functional as F
+import torchvision
+from datasets import load_dataset as hf_load_dataset
+from torch.utils.data import DataLoader
+from torch.utils.data import Dataset
+from torch.utils.data import Subset
+from torch.utils.data import TensorDataset
+from torch.utils.data.dataloader import default_collate
+
+from ..types import Any
+from ..types import Callable
+from ..types import ItemType
+from ..types import List
+from ..types import Optional
+from ..types import TensorType
+from ..types import Tuple
+from ..types import Union
+from .data_handler import DataHandler
+
+
+def dict_only_ds(ds_handling_method: Callable) -> Callable:
+    """Decorator to ensure that the dataset is a dict dataset and that the column_name
+    given as argument matches one of the column names.
+    matches one of the column names. The signature of decorated functions
+    must be function(dataset, *args, **kwargs) with column_name either in kwargs or
+    args[0] when relevant.
+
+
+    Args:
+        ds_handling_method: method to decorate
+
+    Returns:
+        decorated method
+    """
+
+    def wrapper(dataset: Dataset, *args, **kwargs):
+        assert isinstance(
+            dataset[0], dict
+        ), "Dataset must be an instance of DictDataset"
+
+        if "column_name" in kwargs:
+            column_name = kwargs["column_name"]
+        elif len(args) > 0:
+            column_name = args[0]
+
+        # If column_name is provided, check that it is in the dataset column names
+        if (len(args) > 0) or ("column_name" in kwargs):
+            if isinstance(column_name, str):
+                column_name = [column_name]
+            for name in column_name:
+                assert (
+                    name in dataset.column_names
+                ), f"The input dataset has no column named {name}"
+        return ds_handling_method(dataset, *args, **kwargs)
+
+    return wrapper
+
+
+def to_torch(array: TensorType) -> torch.Tensor:
+    """Convert an array into a torch Tensor
+
+    Args:
+        array (TensorType): array to convert
+
+    Returns:
+        torch.Tensor: converted array
+    """
+    if isinstance(array, np.ndarray):
+        return torch.Tensor(array)
+    elif isinstance(array, torch.Tensor):
+        return array
+    else:
+        raise TypeError("Input array must be of numpy or torch type")
+
+
+class DictDataset(Dataset):
+    r"""Dictionary pytorch dataset
+
+    Wrapper to output a dictionary of tensors at the __getitem__ call of a dataset.
+    Some mapping, filtering and concatenation methods are implemented to imitate
+    tensorflow datasets features.
+
+    Args:
+        dataset (Dataset): Dataset to wrap.
+        columns (columns[str]): Column names describing the output tensors.
+    """
+
+    def __init__(
+        self, dataset: Dataset, column_names: List[str] = ["input", "label"]
+    ) -> None:
+        self._dataset = dataset
+        self._raw_columns = column_names
+        self.map_fns = []
+        self._check_init_args()
+
+    @property
+    def column_names(self) -> list:
+        """Get the list of columns in a dict-based item from the dataset.
+
+        Returns:
+            list: column names of the dataset.
+        """
+        dummy_item = self[0]
+        return list(dummy_item.keys())
+
+    def _check_init_args(self) -> None:
+        """Check validity of dataset and column names provided at init"""
+        dummy_item = self._dataset[0]
+        assert isinstance(
+            dummy_item, (tuple, dict, list, torch.Tensor)
+        ), "Dataset to be wrapped needs to return tuple, list or dict of tensors"
+        if isinstance(dummy_item, torch.Tensor):
+            dummy_item = [dummy_item]
+        assert len(dummy_item) == len(
+            self._raw_columns
+        ), "Length mismatch between dataset item and provided column names"
+
+    def __getitem__(self, index: int) -> dict:
+        """Return a dictionary of tensors corresponding to a specfic index.
+
+        Args:
+            index (int): the index of the item to retrieve.
+
+        Returns:
+            dict: tensors for the item at the specific index.
+        """
+        item = self._dataset[index]
+
+        # convert item to a list / tuple of tensors
+        if isinstance(item, torch.Tensor):
+            tensors = [item]
+        elif isinstance(item, dict):
+            tensors = list(item.values())
+        else:
+            tensors = item
+
+        # build output dictionary
+        output_dict = {key: tensor for (key, tensor) in zip(self._raw_columns, tensors)}
+
+        # apply map functions
+        for map_fn in self.map_fns:
+            output_dict = map_fn(output_dict)
+        return output_dict
+
+    def map(self, map_fn: Callable, inplace: bool = False) -> "DictDataset":
+        """Map the dataset
+
+        Args:
+            map_fn (Callable): map function f: dict -> dict
+            inplace (bool): if False, applies the mapping on a copied version of\
+                the dataset. Defaults to False.
+
+        Return:
+            DictDataset: Mapped dataset
+        """
+        dataset = self if inplace else copy.deepcopy(self)
+        dataset.map_fns.append(map_fn)
+        return dataset
+
+    def filter(self, filter_fn: Callable, inplace: bool = False) -> "DictDataset":
+        """Filter the dataset
+
+        Args:
+            filter_fn (Callable): filter function f: dict -> bool
+            inplace (bool): if False, applies the filtering on a copied version of\
+                the dataset. Defaults to False.
+
+        Returns:
+            DictDataset: Filtered dataset
+        """
+        indices = [i for i in range(len(self)) if filter_fn(self[i])]
+        dataset = self if inplace else copy.deepcopy(self)
+        dataset._dataset = Subset(self._dataset, indices)
+        return dataset
+
+    def __len__(self) -> int:
+        """Return the length of the dataset, i.e. the number of items.
+
+        Returns:
+            int: length of the dataset.
+        """
+        return len(self._dataset)
+
+
+class TorchDataHandler(DataHandler):
+    """
+    Class to manage torch DictDataset. The aim is to provide a simple interface
+    for working with torch datasets and manage them without having to use
+    torch syntax.
+    """
+
+    def __init__(self) -> None:
+        """
+        Initializes the TorchDataHandler.
+        Attributes:
+            backend (str): The backend framework used, set to "torch".
+            channel_order (str): The channel order format, set to "channels_first".
+        """
+
+        super().__init__()
+        self.backend = "torch"
+        self.channel_order = "channels_first"
+
+    @staticmethod
+    def _default_target_transform(y: Any) -> torch.Tensor:
+        """Format int or float item target as a torch tensor
+
+        Args:
+            y (Any): dataset item target
+
+        Returns:
+            torch.Tensor: target as a torch.Tensor
+        """
+        return torch.tensor(y) if isinstance(y, (float, int)) else y
+
+    def load_dataset(
+        cls,
+        dataset_id: Union[Dataset, ItemType, str],
+        columns: Optional[list] = None,
+        hub: Optional[str] = "torchvision",
+        load_kwargs: dict = {},
+    ) -> DictDataset:
+        """Load dataset from different manners
+
+        Args:
+            dataset_id (Union[Dataset, ItemType, str]): dataset identification.
+            Can be the name of a dataset from torchvision, a torch Dataset,
+            or a tuple/dict of np.ndarrays/torch tensors.
+            columns (list, optional): Column names. If None, assigned as "input_i"
+                for i-th feature. Defaults to None.
+            load_kwargs (dict, optional): Additional loading kwargs. Defaults to {}.
+
+        Returns:
+            DictDataset: dataset
+        """
+
+        assert hub in {
+            "torchvision",
+            "huggingface",
+        }, "hub must be either 'torchvision' or 'huggingface'"
+
+        if isinstance(dataset_id, str):
+            if hub == "torchvision":
+                assert "root" in load_kwargs.keys()
+                dataset = cls.load_from_torchvision(dataset_id, load_kwargs)
+            elif hub == "huggingface":
+                dataset = cls.load_from_huggingface(dataset_id, load_kwargs)
+        elif isinstance(dataset_id, Dataset):
+            dataset = cls.load_custom_dataset(dataset_id, columns)
+        elif isinstance(dataset_id, get_args(ItemType)):
+            dataset = cls.load_dataset_from_arrays(dataset_id, columns)
+        return dataset
+
+    @staticmethod
+    def load_dataset_from_arrays(
+        dataset_id: ItemType,
+        columns: Optional[list] = None,
+    ) -> DictDataset:
+        """Load a torch.utils.data.Dataset from an array or a tuple/dict of arrays.
+
+        Args:
+            dataset_id (ItemType):
+                numpy / torch array(s) to load.
+            columns (list, optional): Column names to assign. If None,
+                assigned as "input_i" for i-th feature. Defaults to None.
+
+        Returns:
+            DictDataset: dataset
+        """
+        # If dataset_id is an array
+        if isinstance(dataset_id, get_args(TensorType)):
+            tensors = tuple(to_torch(dataset_id))
+            columns = columns or ["input"]
+
+        # If dataset_id is a tuple of arrays
+        elif isinstance(dataset_id, tuple):
+            len_elem = len(dataset_id)
+            if columns is None:
+                if len_elem == 2:
+                    columns = ["input", "label"]
+                else:
+                    columns = [f"input_{i}" for i in range(len_elem - 1)] + ["label"]
+                    print(
+                        "Loading torch.utils.data.Dataset with elems as dicts, "
+                        'assigning "input_i" key to the i-th tuple dimension and'
+                        ' "label" key to the last tuple dimension.'
+                    )
+            assert len(columns) == len(dataset_id)
+            tensors = tuple(to_torch(array) for array in dataset_id)
+
+        # If dataset_id is a dictionary of arrays
+        elif isinstance(dataset_id, dict):
+            columns = columns or list(dataset_id.keys())
+            assert len(columns) == len(dataset_id)
+            tensors = tuple(to_torch(array) for array in dataset_id.values())
+
+        # create torch dictionary dataset from tensors tuple and columns
+        dataset = DictDataset(TensorDataset(*tensors), columns)
+        return dataset
+
+    @staticmethod
+    def load_custom_dataset(
+        dataset_id: Dataset, columns: Optional[list] = None
+    ) -> DictDataset:
+        """Load a custom Dataset by ensuring it has the correct format (dict-based)
+
+        Args:
+            dataset_id (Dataset): Dataset
+            columns (list, optional): Column names to use for elements if dataset_id is
+                tuple based. If None, assigned as "input_i"
+                for i-th column. Defaults to None.
+
+        Returns:
+            DictDataset
+        """
+        # If dataset_id is a tuple based Dataset, convert it to a DictDataset
+        dummy_item = dataset_id[0]
+        if not isinstance(dummy_item, dict):
+            assert isinstance(
+                dummy_item, (Tuple, torch.Tensor)
+            ), "Custom dataset should be either dictionary based or tuple based"
+            if columns is None:
+                len_elem = len(dummy_item)
+                if len_elem == 2:
+                    columns = ["input", "label"]
+                else:
+                    columns = [f"input_{i}" for i in range(len_elem - 1)] + ["label"]
+                    print(
+                        "Feature name not found, assigning 'input_i' "
+                        "key to the i-th tensor and 'label' key to the last"
+                    )
+            dataset_id = DictDataset(dataset_id, columns)
+
+        dataset = dataset_id
+        return dataset
+
+    @classmethod
+    def load_from_huggingface(
+        cls,
+        dataset_id: str,
+        load_kwargs: dict = {},
+    ) -> DictDataset:
+        """Load a Dataset from the Hugging Face datasets catalog
+
+        Args:
+            dataset_id (str): Identifier of the dataset
+            load_kwargs (dict): Loading kwargs to add to the initialization
+            of the dataset.
+
+        Returns:
+            DictDataset: dataset
+        """
+        if "transform" in load_kwargs.keys():
+            transform = load_kwargs["transform"]
+            load_kwargs.pop("transform")
+        else:
+
+            def transform(x):
+                return x
+
+        dataset = hf_load_dataset(dataset_id, **load_kwargs)
+
+        def transform_full(examples):
+            examples = transform(examples)
+            examples["label"] = [
+                cls._default_target_transform(example) for example in examples["label"]
+            ]
+            return examples
+
+        dataset = dataset.with_transform(transform_full)
+        return dataset  # HF datasets are already dict-based
+
+    @classmethod
+    def load_from_torchvision(
+        cls,
+        dataset_id: str,
+        load_kwargs: dict = {},
+    ) -> DictDataset:
+        """Load a Dataset from the torchvision datasets catalog
+
+        Args:
+            dataset_id (str): Identifier of the dataset
+            root (str): Root directory of dataset
+            transform (Callable, optional): Transform function to apply to the input.
+                Defaults to DEFAULT_TRANSFORM.
+            target_transform (Callable, optional): Transform function to apply
+                to the target. Defaults to DEFAULT_TARGET_TRANSFORM.
+            download (bool):  If true, downloads the dataset from the internet and puts
+                it in root directory. If dataset is already downloaded, it is not
+                downloaded again. Defaults to False.
+            load_kwargs (dict): Loading kwargs to add to the initialization
+                of dataset.
+
+        Returns:
+            DictDataset: dataset
+        """
+        assert (
+            dataset_id in torchvision.datasets.__all__
+        ), "Dataset not available on torchvision datasets catalog"
+
+        if "transform" not in load_kwargs.keys():
+            load_kwargs["transform"] = torchvision.transforms.PILToTensor()
+        if "target_transform" not in load_kwargs.keys():
+            load_kwargs["target_transform"] = cls._default_target_transform
+
+        dataset = getattr(torchvision.datasets, dataset_id)(
+            **load_kwargs,
+        )
+        return cls.load_custom_dataset(dataset)
+
+    @staticmethod
+    @dict_only_ds
+    def get_ds_column_names(dataset: DictDataset) -> list:
+        """Get the column names of a DictDataset
+
+        Args:
+            dataset (DictDataset): Dataset to get the column names from
+
+        Returns:
+            list: List of column names
+        """
+        return dataset.column_names
+
+    @staticmethod
+    def map_ds(
+        dataset: DictDataset,
+        map_fn: Callable,
+    ) -> DictDataset:
+        """Map a function to a DictDataset
+
+        Args:
+            dataset (DictDataset): Dataset to map the function to
+            map_fn (Callable): Function to map
+
+        Returns:
+            DictDataset: Mapped dataset
+        """
+        return dataset.map(map_fn)
+
+    @staticmethod
+    @dict_only_ds
+    def filter_by_value(
+        dataset: DictDataset,
+        column_name: str,
+        values: list,
+        excluded: bool = False,
+    ) -> DictDataset:
+        """Filter the dataset by checking if the value of a column is in `values`
+
+        !!! note
+            This function can be a bit of time consuming since it needs to iterate
+            over the whole dataset.
+
+        Args:
+            dataset (DictDataset): Dataset to filter
+            column_name (str): Column to filter the dataset with
+            values (list): Column values to keep
+            excluded (bool, optional): To keep (False) or exclude (True) the samples
+                with column values included in Values. Defaults to False.
+
+        Returns:
+            DictDataset: Filtered dataset
+        """
+
+        if len(dataset[0][column_name].shape) > 0:
+            value_dim = dataset[0][column_name].shape[-1]
+            values = [
+                F.one_hot(torch.tensor(value).long(), value_dim) for value in values
+            ]
+
+        def filter_fn(x):
+            keep = any([torch.all(x[column_name] == v) for v in values])
+            return keep if not excluded else not keep
+
+        filtered_dataset = dataset.filter(filter_fn)
+        return filtered_dataset
+
+    @classmethod
+    def prepare(
+        cls,
+        dataset: DictDataset,
+        batch_size: int,
+        preprocess_fn: Optional[Callable] = None,
+        augment_fn: Optional[Callable] = None,
+        columns: Optional[list] = None,
+        shuffle: bool = False,
+        dict_based_fns: bool = True,
+        return_tuple: bool = True,
+        num_workers: int = 0,
+    ) -> DataLoader:
+        """Prepare a DataLoader for training
+
+        Args:
+            dataset (DictDataset): Dataset to prepare
+            batch_size (int): Batch size
+            preprocess_fn (Callable, optional): Preprocessing function to apply to
+                the dataset. Defaults to None.
+            augment_fn (Callable, optional): Augment function to be used (when the
+                returned dataset is to be used for training). Defaults to None.
+            columns (list, optional): List of column names corresponding to the columns
+                that will be returned. Keep all features if None. Defaults to None.
+            shuffle (bool, optional): To shuffle the returned dataset or not.
+                Defaults to False.
+            dict_based_fns (bool): Whether to use preprocess and DA functions as dict
+                based (if True) or as tuple based (if False). Defaults to True.
+            return_tuple (bool, optional): Whether to return each dataset item
+                as a tuple. Defaults to True.
+            num_workers (int, optional): Number of workers to use for the dataloader.
+
+        Returns:
+            DataLoader: dataloader
+        """
+        columns = columns or cls.get_ds_column_names(dataset)
+
+        def collate_fn(batch: List[dict]):
+            if dict_based_fns:
+                # preprocess + DA: List[dict] -> List[dict]
+                preprocess_func = preprocess_fn or (lambda x: x)
+                augment_func = augment_fn or (lambda x: x)
+                batch = [augment_func(preprocess_func(d)) for d in batch]
+                # to dict of batchs
+                if return_tuple:
+                    return tuple(
+                        default_collate([d[key] for d in batch]) for key in columns
+                    )
+                return {
+                    key: default_collate([d[key] for d in batch]) for key in columns
+                }
+            else:
+                # preprocess + DA: List[dict] -> List[tuple]
+                preprocess_func = preprocess_fn or (lambda *x: x)
+                augment_func = augment_fn or (lambda *x: x)
+                batch = [
+                    augment_func(*preprocess_func(*tuple(d[key] for key in columns)))
+                    for d in batch
+                ]
+                # to tuple of batchs
+                return default_collate(batch)
+
+        loader = DataLoader(
+            dataset,
+            batch_size=batch_size,
+            shuffle=shuffle,
+            collate_fn=collate_fn,
+            num_workers=num_workers,
+        )
+        return loader
+
+    @staticmethod
+    def get_item_length(dataset: Dataset) -> int:
+        """Number of elements in a dataset item
+
+        Args:
+            dataset (DictDataset): Dataset
+
+        Returns:
+            int: Item length
+        """
+        return len(dataset[0])
+
+    @staticmethod
+    def get_dataset_length(dataset: Dataset) -> int:
+        """Number of items in a dataset
+
+        Args:
+            dataset (DictDataset): Dataset
+
+        Returns:
+            int: Dataset length
+        """
+        return len(dataset)
+
+    @staticmethod
+    def get_column_elements_shape(
+        dataset: Dataset, column_name: Union[str, int]
+    ) -> tuple:
+        """Get the shape of the elements of a column of dataset identified by
+        column_name
+
+        Args:
+            dataset (Dataset): a Dataset
+            column_name (Union[str, int]): The column name to get
+                the element shape from.
+
+        Returns:
+            tuple: the shape of an element from column_name
+        """
+        return tuple(dataset[0][column_name].shape)
+
+    @staticmethod
+    def get_columns_shapes(dataset: Dataset) -> dict:
+        """Get the shapes of the elements of all columns of a dataset
+
+        Args:
+            dataset (Dataset): a Dataset
+
+        Returns:
+            dict: dictionary of column names and their corresponding shape
+        """
+        shapes = {}
+        for key in dataset.column_names:
+            try:
+                shapes[key] = tuple(dataset[0][key].shape)
+            except AttributeError:
+                pass
+        return shapes
+
+    @staticmethod
+    def get_input_from_dataset_item(elem: ItemType) -> Any:
+        """Get the tensor that is to be feed as input to a model from a dataset element.
+
+        Args:
+            elem (ItemType): dataset element to extract input from
+
+        Returns:
+            Any: Input tensor
+        """
+        if isinstance(elem, (tuple, list)):
+            tensor = elem[0]
+        elif isinstance(elem, dict):
+            tensor = elem[list(elem.keys())[0]]
+        else:
+            tensor = elem
+        return tensor
+
+    @staticmethod
+    def get_label_from_dataset_item(item: ItemType):
+        """Retrieve label tensor from item as a tuple/list. Label must be at index 1
+        in the item tuple. If one-hot encoded, labels are converted to single value.
+
+        Args:
+            elem (ItemType): dataset element to extract label from
+
+        Returns:
+            Any: Label tensor
+        """
+        label = item[1]  # labels must be at index 1 in the batch tuple
+        # If labels are one-hot encoded, take the argmax
+        if len(label.shape) > 1 and label.shape[1] > 1:
+            label = label.view(label.size(0), -1)
+            label = torch.argmax(label, dim=1)
+        # If labels are in two dimensions, squeeze them
+        if len(label.shape) > 1:
+            label = label.view([label.shape[0]])
+        return label

oodeel/eval/__init__.py

@@ -0,0 +1,22 @@
+# -*- coding: utf-8 -*-
+# Copyright IRT Antoine de Saint Exupéry et Université Paul Sabatier Toulouse III - All
+# rights reserved. DEEL is a research program operated by IVADO, IRT Saint Exupéry,
+# CRIAQ and ANITI - https://www.deel.ai/
+#
+# Permission is hereby granted, free of charge, to any person obtaining a copy
+# of this software and associated documentation files (the "Software"), to deal
+# in the Software without restriction, including without limitation the rights
+# to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
+# copies of the Software, and to permit persons to whom the Software is
+# furnished to do so, subject to the following conditions:
+#
+# The above copyright notice and this permission notice shall be included in all
+# copies or substantial portions of the Software.
+#
+# THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+# IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+# FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+# AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+# LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+# OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+# SOFTWARE.