heavyedge-dataset 0.2.0__py3-none-any.whl → 1.0.0.post0__py3-none-any.whl

heavyedge_dataset/__init__.py

@@ -1,97 +1,31 @@
-"""Custom dataset classes for edge profiles.
+"""Package to load edge profile data using PyTorch dataset.
 
-Refer to `PyTorch tutorial <tutorial>`_ for information about custom PyTorch dataset.
+Refer to `PyTorch tutorial <tutorial>`_ for information about custom dataset.
 
 .. _tutorial: https://docs.pytorch.org/tutorials/beginner/data_loading_tutorial.html
 """
 
-import abc
 import numbers
 from collections.abc import Sequence
 
 import numpy as np
-from heavyedge.api import landmarks_type3
 from torch.utils.data import Dataset
 
 __all__ = [
     "ProfileDataset",
-    "PseudoLandmarkDataset",
-    "MathematicalLandmarkDataset",
 ]
 
 
-class ProfileDatasetBase(abc.ABC):
-    """Abstract base class for profile dataset."""
+class ProfileDataset(Dataset):
+    """Edge profile dataset.
 
-    @property
-    @abc.abstractmethod
-    def file(self):
-        """Profile data file.
+    Loads data as a tuple of two numpy arrays:
 
-        Returns
-        -------
-        heavyedge.ProfileData
-        """
+    1. Profile data, shape: (N, m, L).
+    2. Length of each profile, shape: (N,).
 
-    @property
-    @abc.abstractmethod
-    def transform(self):
-        """Optional transformation to be applied on samples.
-
-        Returns
-        -------
-        Callable
-        """
-
-    def __len__(self):
-        return len(self.file)
-
-    def __getitem__(self, idx):
-        if isinstance(idx, numbers.Integral):
-            Y, L, _ = self.file[idx]
-            Ys, Ls = [Y], [L]
-        else:
-            # Support multi-indexing
-            idxs = idx
-            needs_sort = isinstance(idx, (Sequence, np.ndarray))
-            if needs_sort:
-                # idxs must be sorted for h5py
-                idxs = np.array(idxs)
-                sort_idx = np.argsort(idxs)
-                idxs = idxs[sort_idx]
-            Ys, Ls, _ = self.file[idxs]
-            if needs_sort:
-                reverse_idx = np.argsort(sort_idx)
-                Ys = Ys[reverse_idx]
-                Ls = Ls[reverse_idx]
-
-        ret = self.default_transform(Ys, Ls)
-        if self.transform:
-            ret = self.transform(ret)
-        return ret
-
-    def __getitems__(self, idxs):
-        # PyTorch API
-        return self.__getitem__(idxs)
-
-    @abc.abstractmethod
-    def default_transform(self, profiles, lengths):
-        """Default data transformation.
-
-        Subclass must implement this method to transform profile data into target data.
-
-        Parameters
-        ----------
-        profiles : (N, M) array
-            Profile data.
-        lengths : (N,) array
-            Length of each profile in *profiles*.
-        """
-        pass
-
-
-class ProfileDataset(ProfileDatasetBase, Dataset):
-    """Full profile dataset in 1-D or 2-D.
+    N is the number of loaded data, m is dimension of coordinates, and
+    L is the maximum length of profiles.
 
     Parameters
     ----------
@@ -108,165 +42,78 @@ class ProfileDataset(ProfileDatasetBase, Dataset):
     >>> from heavyedge import get_sample_path, ProfileData
     >>> from heavyedge_dataset import ProfileDataset
     >>> with ProfileData(get_sample_path("Prep-Type2.h5")) as file:
-    ...     data = ProfileDataset(file, 2)[:]
-    >>> import matplotlib.pyplot as plt  # doctest: +SKIP
-    ... for coords in data:
-    ...     plt.plot(*coords, color="gray")
-    """
-
-    def __init__(self, file, m, transform=None):
-        self._file = file
-        self.m = m
-        self._transform = transform
-
-        self.x = file.x()
-
-    @property
-    def file(self):
-        return self._file
-
-    @property
-    def transform(self):
-        return self._transform
-
-    def default_transform(self, profiles, lengths):
-        """Crop profiles by their contact points.
+    ...     profiles, lengths = ProfileDataset(file, m=2)[:]
+    >>> profiles.shape
+    (22, 2, 3200)
+    >>> lengths.shape
+    (22,)
 
-        Parameters
-        ----------
-        profiles : (N, M) array
-            Profile data.
-        lengths : (N,) array
-            Length of each profile in *profiles*.
-        """
-        if self.m == 1:
-            ret = [Y[:L].reshape(1, -1) for Y, L in zip(profiles, lengths)]
-        elif self.m == 2:
-            ret = [np.stack([self.x[:L], Y[:L]]) for Y, L in zip(profiles, lengths)]
-        else:
-            raise ValueError(f"Invalid dimension: {self.m}")
-        return ret
-
-
-class PseudoLandmarkDataset(ProfileDatasetBase, Dataset):
-    """Pseudo-landmark dataset in 1-D or 2-D.
-
-    Pseudo-landmarks are points that are equidistantly sampled.
-
-    Parameters
-    ----------
-    file : heavyedge.ProfileData
-        Open hdf5 file.
-    k : int
-        Number of landmarks to sample.
-    m : {1, 2}
-        Profile data dimension.
-        1 means only y coordinates, and 2 means both x and y coordinates.
-    transform : callable, optional
-        Optional transformation to be applied on samples.
+    Should this dataset be used for :class:`torch.utils.data.DataLoader`,
+    ``collate_fn`` argument should be passed to the data loader.
 
-    Examples
-    --------
-    >>> from heavyedge import get_sample_path, ProfileData
-    >>> from heavyedge_dataset import PseudoLandmarkDataset
+    >>> from torch.utils.data import DataLoader
     >>> with ProfileData(get_sample_path("Prep-Type2.h5")) as file:
-    ...     data = PseudoLandmarkDataset(file, 10, 2)[:]
-    >>> import matplotlib.pyplot as plt  # doctest: +SKIP
-    ... plt.plot(*data.transpose(1, 2, 0), color="gray")
+    ...     dataset = ProfileDataset(file, m=2)
+    ...     loader = DataLoader(dataset, collate_fn=lambda x: x)
+    ...     profiles, lengths = next(iter(loader))
+    >>> profiles.shape
+    (1, 2, 3200)
+    >>> lengths.shape
+    (1,)
+
+    If data should be loaded as :class:`torch.Tensor`, pass ``transform`` argument.
+
+    >>> import torch
+    >>> def to_tensor(sample):
+    ...     return (torch.from_numpy(sample[0]), torch.from_numpy(sample[1]))
+    >>> with ProfileData(get_sample_path("Prep-Type2.h5")) as file:
+    ...     dataset = ProfileDataset(file, m=2, transform=to_tensor)
+    ...     loader = DataLoader(dataset, collate_fn=lambda x: x)
+    ...     profiles, lengths = next(iter(loader))
+    >>> type(profiles)
+    <class 'torch.Tensor'>
     """
 
-    def __init__(self, file, k, m, transform=None):
-        self._file = file
-        self.k = k
+    def __init__(self, file, m=1, transform=None):
+        self.file = file
         self.m = m
-        self._transform = transform
-
+        self.transform = transform
         self.x = file.x()
 
-    @property
-    def file(self):
-        return self._file
-
-    @property
-    def transform(self):
-        return self._transform
-
-    def default_transform(self, profiles, lengths):
-        """Sample pseudo-landmarks from profiles.
+    def __len__(self):
+        return len(self.file)
 
-        Parameters
-        ----------
-        profiles : (N, M) array
-            Profile data.
-        lengths : (N,) array
-            Length of each profile in *profiles*.
-        """
-        ret = []
+    def __getitem__(self, idx):
+        if isinstance(idx, numbers.Integral):
+            Y, L, _ = self.file[idx]
+            Y = Y[np.newaxis, :]
+        else:
+            # Support multi-indexing
+            idxs = idx
+            needs_sort = isinstance(idx, (Sequence, np.ndarray))
+            if needs_sort:
+                # idxs must be sorted for h5py
+                idxs = np.array(idxs)
+                sort_idx = np.argsort(idxs)
+                idxs = idxs[sort_idx]
+            Y, L, _ = self.file[idxs]
+            if needs_sort:
+                reverse_idx = np.argsort(sort_idx)
+                Y = Y[reverse_idx]
+                L = L[reverse_idx]
+            Y = Y[:, np.newaxis, :]
         if self.m == 1:
-            for Y, L in zip(profiles, lengths):
-                idxs = np.linspace(0, L - 1, self.k, dtype=int)
-                ret.append(Y[idxs].reshape(1, -1))
+            pass
         elif self.m == 2:
-            for Y, L in zip(profiles, lengths):
-                idxs = np.linspace(0, L - 1, self.k, dtype=int)
-                ret.append(np.stack([self.x[idxs], Y[idxs]]))
+            x = np.tile(self.x, Y.shape[:-1] + (1,))
+            Y = np.concatenate([x, Y], axis=-2)
         else:
-            raise ValueError(f"Invalid dimension: {self.m}")
-        return np.array(ret)
-
-
-class MathematicalLandmarkDataset(ProfileDatasetBase, Dataset):
-    """Mathematical landmark dataset in 1-D.
-
-    Mathematical landmarks are points which are choosed by their
-    mathematical properties, i.e., slope or curvature.
-
-    Parameters
-    ----------
-    file : heavyedge.ProfileData
-        Open hdf5 file.
-    sigma : scalar
-        Standard deviation of Gaussian kernel for landmark detection.
-    transform : callable, optional
-        Optional transformation to be applied on samples.
-
-    Examples
-    --------
-    >>> from heavyedge import get_sample_path, ProfileData
-    >>> from heavyedge_dataset import MathematicalLandmarkDataset
-    >>> with ProfileData(get_sample_path("Prep-Type2.h5")) as file:
-    ...     data = MathematicalLandmarkDataset(file, 32)[:]
-    >>> import matplotlib.pyplot as plt  # doctest: +SKIP
-    ... plt.plot(*data.transpose(1, 2, 0), color="gray")
-    """
-
-    def __init__(self, file, sigma, transform=None):
-        self._file = file
-        self.sigma = sigma
-        self._transform = transform
-
-    @property
-    def file(self):
-        return self._file
-
-    @property
-    def transform(self):
-        return self._transform
-
-    def default_transform(self, profiles, lengths):
-        """Detect mathematical landmarks from profiles.
+            raise ValueError(f"Unsupported dimension: {self.m} (Must be 1 or 2).")
+        ret = (Y, L)
+        if self.transform is not None:
+            ret = self.transform(ret)
+        return ret
 
-        Parameters
-        ----------
-        profiles : (N, M) array
-            Profile data.
-        lengths : (N,) array
-            Length of each profile in *profiles*.
-        """
-        ret = []
-        for Y, L in zip(profiles, lengths):
-            Y = Y[:L]
-            indices = np.flip(landmarks_type3(Y, self.sigma))
-            y = np.concat([[np.mean(Y[: indices[0]])], Y[indices]])
-            ret.append(y.reshape(1, -1))
-        return np.array(ret)
+    def __getitems__(self, idxs):
+        # PyTorch API
+        return self.__getitem__(idxs)

{heavyedge_dataset-0.2.0.dist-info → heavyedge_dataset-1.0.0.post0.dist-info}/METADATA

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: heavyedge-dataset
-Version: 0.2.0
+Version: 1.0.0.post0
 Summary: PyTorch-compatible edge profile dataset API
 Author-email: Jisoo Song <jeesoo9595@snu.ac.kr>
 License-Expression: MIT
@@ -28,7 +28,6 @@ Provides-Extra: doc
 Requires-Dist: sphinx; extra == "doc"
 Requires-Dist: numpydoc; extra == "doc"
 Requires-Dist: pydata_sphinx_theme; extra == "doc"
-Requires-Dist: matplotlib; extra == "doc"
 Provides-Extra: dev
 Requires-Dist: flake8; extra == "dev"
 Requires-Dist: black; extra == "dev"
@@ -49,14 +48,14 @@ Package to load edge profile data as PyTorch dataset.
 
 ## Usage
 
-HeavyEdge-Dataset provides custom dataset classes which wraps profile data file.
+HeavyEdge-Dataset provides `ProfileDataset` which wraps profile data file.
 
-A simple use case to load a list of profiles as two-dimensional coordinates:
+A simple use case to load two-dimensional coordinates of profiles and their lengths:
 
 ```python
 from heavyedge import get_sample_path, ProfileData
 from heavyedge_dataset import ProfileDataset
-with ProfileData(get_sample_path("Prep-Type2.h5")) as file:  # Profile data file object
+with ProfileData(get_sample_path("Prep-Type2.h5")) as file:
     data = ProfileDataset(file, 2)[:]
 ```
 

heavyedge_dataset-1.0.0.post0.dist-info/RECORD

@@ -0,0 +1,6 @@
+heavyedge_dataset/__init__.py,sha256=CmquaxuwXJYTCGInmzI8Tp0Z9hJBVtEK9eH3NLyoREY,3737
+heavyedge_dataset-1.0.0.post0.dist-info/licenses/LICENSE,sha256=pBq2E7YJkUcEINdYeERL4RVFOQICd_MwJq6OusuAPGc,1066
+heavyedge_dataset-1.0.0.post0.dist-info/METADATA,sha256=34mMpniE5Nn4E6mwdVKhT4m6mnEUKRoCce9Tv6aGNHI,3434
+heavyedge_dataset-1.0.0.post0.dist-info/WHEEL,sha256=_zCd3N1l69ArxyTb8rzEoP9TpbYXkqRFSNOD5OuxnTs,91
+heavyedge_dataset-1.0.0.post0.dist-info/top_level.txt,sha256=wpRFI8TlkYFGetc17appkyybauBvzhGGvyueunsdJTc,18
+heavyedge_dataset-1.0.0.post0.dist-info/RECORD,,

heavyedge_dataset-0.2.0.dist-info/RECORD

@@ -1,6 +0,0 @@
-heavyedge_dataset/__init__.py,sha256=iBFvupZMaHpTGUxZOJvVKfD4bAOAdikK60a8bd4oxB0,7746
-heavyedge_dataset-0.2.0.dist-info/licenses/LICENSE,sha256=pBq2E7YJkUcEINdYeERL4RVFOQICd_MwJq6OusuAPGc,1066
-heavyedge_dataset-0.2.0.dist-info/METADATA,sha256=vZF2VwCYUmf7OzXfCO-pg9cHEeE7kCYIfl1JGmfTad0,3496
-heavyedge_dataset-0.2.0.dist-info/WHEEL,sha256=_zCd3N1l69ArxyTb8rzEoP9TpbYXkqRFSNOD5OuxnTs,91
-heavyedge_dataset-0.2.0.dist-info/top_level.txt,sha256=wpRFI8TlkYFGetc17appkyybauBvzhGGvyueunsdJTc,18
-heavyedge_dataset-0.2.0.dist-info/RECORD,,