oodeel 0.4.0__py3-none-any.whl

oodeel/datasets/deprecated/DEPRECATED_ooddataset.py

@@ -0,0 +1,330 @@
+# -*- 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 numpy as np
+
+from ...types import Callable
+from ...types import DatasetType
+from ...types import Optional
+from ...types import Tuple
+from ...types import Union
+
+
+class OODDataset(object):
+    """Class for managing loading and processing of datasets that are to be used for
+    OOD detection. The class encapsulates a dataset like object augmented with OOD
+    related information, and then returns a dataset like object that is suited for
+    scoring or training with the .prepare method.
+
+    Args:
+        dataset_id (Union[DatasetType, tuple, dict, str]): The dataset to load.
+            Can be loaded from tensorflow or torch datasets catalog when the str matches
+            one of the datasets. Defaults to Union[DatasetType, tuple, dict, str].
+        backend (str, optional): Whether the dataset is to be used for tensorflow
+             or torch models. Defaults to "tensorflow". Alternative: "torch".
+        keys (list, optional): keys to use for dataset elems. Default to None
+        load_kwargs (dict, optional): Additional loading kwargs when loading from
+            tensorflow_datasets catalog. Defaults to {}.
+        load_from_tensorflow_datasets (bool, optional): In the case where if the backend
+            is torch but the user still wants to import from tensorflow_datasets
+            catalog. In that case, tf.Tensor will not be loaded in VRAM and converted as
+            torch.Tensors on the fly. Defaults to False.
+        input_key (str, optional): The key of the element/item to consider as the
+            model input tensor. If None, taken as the first key. Defaults to None.
+    """
+
+    def __init__(
+        self,
+        dataset_id: Union[DatasetType, tuple, dict, str],
+        backend: str = "tensorflow",
+        keys: Optional[list] = None,
+        load_kwargs: dict = {},
+        load_from_tensorflow_datasets: bool = False,
+        input_key: Optional[str] = None,
+    ):
+
+        self.backend = backend
+        self.load_from_tensorflow_datasets = load_from_tensorflow_datasets
+
+        # The length of the dataset is kept as attribute to avoid redundant
+        # iterations over self.data
+        self.length = None
+
+        # Set the load parameters for tfds / torchvision
+        if backend == "tensorflow":
+            load_kwargs["as_supervised"] = False
+        # Set the channel order depending on the backend
+        if self.backend == "torch":
+            if load_from_tensorflow_datasets:
+                from .DEPRECATED_tf_data_handler import TFDataHandler
+                import tensorflow as tf
+
+                tf.config.set_visible_devices([], "GPU")
+                self._data_handler = TFDataHandler()
+                load_kwargs["as_supervised"] = False
+            else:
+                from .DEPRECATED_torch_data_handler import TorchDataHandler
+
+                self._data_handler = TorchDataHandler()
+            self.channel_order = "channels_first"
+        else:
+            from .DEPRECATED_tf_data_handler import TFDataHandler
+
+            self._data_handler = TFDataHandler()
+            self.channel_order = "channels_last"
+
+        self.load_params = load_kwargs
+        # Load the dataset depending on the type of dataset_id
+        self.data = self._data_handler.load_dataset(dataset_id, keys, load_kwargs)
+
+        # Get the length of the elements/items in the dataset
+        self.len_item = self._data_handler.get_item_length(self.data)
+        if self.has_ood_label:
+            self.len_item -= 1
+
+        # Get the key of the tensor to feed the model with
+        if input_key is None:
+            self.input_key = self._data_handler.get_ds_feature_keys(self.data)[0]
+        else:
+            self.input_key = input_key
+
+    def __len__(self) -> int:
+        """get the length of the dataset.
+
+        Returns:
+            int: length of the dataset
+        """
+        if self.length is None:
+            self.length = self._data_handler.get_dataset_length(self.data)
+        return self.length
+
+    @property
+    def has_ood_label(self) -> bool:
+        """Check if the dataset has an out-of-distribution label.
+
+        Returns:
+            bool: True if data handler has a "ood_label" feature key.
+        """
+        return self._data_handler.has_feature_key(self.data, "ood_label")
+
+    def get_ood_labels(
+        self,
+    ) -> np.ndarray:
+        """Get ood_labels from self.data if any
+
+        Returns:
+            np.ndarray: array of labels
+        """
+        assert self._data_handler.has_feature_key(
+            self.data, "ood_label"
+        ), "The data has no ood_labels"
+        labels = self._data_handler.get_feature_from_ds(self.data, "ood_label")
+        return labels
+
+    def add_out_data(
+        self,
+        out_dataset: Union["OODDataset", DatasetType],
+        in_value: int = 0,
+        out_value: int = 1,
+        resize: Optional[bool] = False,
+        shape: Optional[Tuple[int]] = None,
+    ) -> "OODDataset":
+        """Concatenate two OODDatasets. Useful for scoring on multiple datasets, or
+        training with added out-of-distribution data.
+
+        Args:
+            out_dataset (Union[OODDataset, DatasetType]): dataset of
+                out-of-distribution data
+            in_value (int): ood label value for in-distribution data. Defaults to 0
+            out_value (int): ood label value for out-of-distribution data. Defaults to 1
+            resize (Optional[bool], optional):toggles if input tensors of the
+                datasets have to be resized to have the same shape. Defaults to False.
+            shape (Optional[Tuple[int]], optional):shape to use for resizing input
+                tensors. If None, the tensors are resized with the shape of the
+                in_dataset input tensors. Defaults to None.
+
+        Returns:
+            OODDataset: a Dataset object with the concatenated data
+        """
+
+        # Creating an OODDataset object from out_dataset if necessary and make sure
+        # the two OODDatasets have compatible parameters
+        if isinstance(out_dataset, type(self)):
+            out_dataset = out_dataset.data
+        else:
+            out_dataset = OODDataset(out_dataset, backend=self.backend).data
+
+        # Assign the correct ood_label to self.data, depending on out_as_in
+        self.data = self._data_handler.assign_feature_value(
+            self.data, "ood_label", in_value
+        )
+        out_dataset = self._data_handler.assign_feature_value(
+            out_dataset, "ood_label", out_value
+        )
+
+        # Merge the two underlying Datasets
+        merge_kwargs = (
+            {"channel_order": self.channel_order}
+            if self.backend == "tensorflow"
+            else {}
+        )
+        data = self._data_handler.merge(
+            self.data,
+            out_dataset,
+            resize=resize,
+            shape=shape,
+            **merge_kwargs,
+        )
+
+        # Create a new OODDataset from the merged Dataset
+        output_ds = OODDataset(
+            dataset_id=data,
+            backend=self.backend,
+        )
+
+        return output_ds
+
+    def split_by_class(
+        self,
+        in_labels: Optional[Union[np.ndarray, list]] = None,
+        out_labels: Optional[Union[np.ndarray, list]] = None,
+    ) -> Optional[Tuple["OODDataset"]]:
+        """Filter the dataset by assigning ood labels depending on labels
+        value (typically, class id).
+
+        Args:
+            in_labels (Optional[Union[np.ndarray, list]], optional): set of labels
+                to be considered as in-distribution. Defaults to None.
+            out_labels (Optional[Union[np.ndarray, list]], optional): set of labels
+                to be considered as out-of-distribution. Defaults to None.
+
+        Returns:
+            Optional[Tuple[OODDataset]]: Tuple of in-distribution and
+                out-of-distribution OODDatasets
+        """
+        # Make sure the dataset has labels
+        assert (in_labels is not None) or (
+            out_labels is not None
+        ), "specify labels to filter with"
+        assert self.len_item >= 2, "the dataset has no labels"
+
+        # Filter the dataset depending on in_labels and out_labels given
+        if (out_labels is not None) and (in_labels is not None):
+            in_data = self._data_handler.filter_by_feature_value(
+                self.data, "label", in_labels
+            )
+            out_data = self._data_handler.filter_by_feature_value(
+                self.data, "label", out_labels
+            )
+
+        if out_labels is None:
+            in_data = self._data_handler.filter_by_feature_value(
+                self.data, "label", in_labels
+            )
+            out_data = self._data_handler.filter_by_feature_value(
+                self.data, "label", in_labels, excluded=True
+            )
+
+        elif in_labels is None:
+            in_data = self._data_handler.filter_by_feature_value(
+                self.data, "label", out_labels, excluded=True
+            )
+            out_data = self._data_handler.filter_by_feature_value(
+                self.data, "label", out_labels
+            )
+
+        # Return the filtered OODDatasets
+        return (
+            OODDataset(in_data, backend=self.backend),
+            OODDataset(out_data, backend=self.backend),
+        )
+
+    def prepare(
+        self,
+        batch_size: int = 128,
+        preprocess_fn: Optional[Callable] = None,
+        augment_fn: Optional[Callable] = None,
+        with_ood_labels: bool = False,
+        with_labels: bool = True,
+        shuffle: bool = False,
+        **kwargs_prepare,
+    ) -> DatasetType:
+        """Prepare self.data for scoring or training
+
+        Args:
+            batch_size (int, optional): Batch_size of the returned dataset like object.
+                Defaults to 128.
+            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.
+            with_ood_labels (bool, optional): To return the dataset with ood_labels
+                or not. Defaults to True.
+            with_labels (bool, optional): To return the dataset with labels or not.
+                Defaults to True.
+            shuffle (bool, optional): To shuffle the returned dataset or not.
+                Defaults to False.
+            kwargs_prepare (dict): Additional parameters to be passed to the
+                data_handler.prepare_for_training method.
+
+
+        Returns:
+            DatasetType: prepared dataset
+        """
+        # Check if the dataset has at least one of label and ood_label
+        assert (
+            with_ood_labels or with_labels
+        ), "The dataset must have at least one of label and ood_label"
+
+        # Check if the dataset has ood_labels when asked to return with_ood_labels
+        if with_ood_labels:
+            assert (
+                self.has_ood_label
+            ), "Please assign ood labels before preparing with ood_labels"
+
+        dataset_to_prepare = self.data
+
+        # Making the dataset channel first if the backend is torch
+        if self.backend == "torch" and self.load_from_tensorflow_datasets:
+            dataset_to_prepare = self._data_handler.make_channel_first(
+                self.input_key, dataset_to_prepare
+            )
+
+        # # Select the keys to be returned
+        keys = [self.input_key, "label", "ood_label"]
+        if not with_labels:
+            keys.remove("label")
+        if not with_ood_labels:
+            keys.remove("ood_label")
+
+        # Prepare the dataset for training or scoring
+        dataset = self._data_handler.prepare_for_training(
+            dataset=dataset_to_prepare,
+            batch_size=batch_size,
+            shuffle=shuffle,
+            preprocess_fn=preprocess_fn,
+            augment_fn=augment_fn,
+            output_keys=keys,
+            **kwargs_prepare,
+        )
+
+        return dataset