feat: added MNIST, Fashion-MNIST and KMNIST datasets (#164)

Closes #161 
Closes #162 
Closes #163 

### Summary of Changes

feat: added MNIST, Fashion-MNIST and KMNIST datasets
build: bump safe-ds to ^0.24.0

---------

Co-authored-by: megalinter-bot <[email protected]>
Co-authored-by: Lars Reimann <[email protected]>

pyproject.toml

@@ -14,7 +14,7 @@ packages = [
 
 [tool.poetry.dependencies]
 python = "^3.11,<3.13"
-safe-ds = ">=0.17,<0.27"
+safe-ds = ">=0.24,<0.27"
 
 [tool.poetry.group.dev.dependencies]
 pytest = ">=7.2.1,<9.0.0"

src/safeds_datasets/image/__init__.py

@@ -0,0 +1,5 @@
+"""Image datasets."""
+
+from ._mnist import load_fashion_mnist, load_kmnist, load_mnist
+
+__all__ = ["load_fashion_mnist", "load_kmnist", "load_mnist"]

src/safeds_datasets/image/_mnist/__init__.py

@@ -0,0 +1,5 @@
+"""MNIST like Datasets."""
+
+from ._mnist import load_fashion_mnist, load_kmnist, load_mnist
+
+__all__ = ["load_fashion_mnist", "load_kmnist", "load_mnist"]

src/safeds_datasets/image/_mnist/_mnist.py

@@ -0,0 +1,256 @@
+import gzip
+import os
+import struct
+import sys
+import urllib.request
+from array import array
+from pathlib import Path
+from typing import TYPE_CHECKING
+from urllib.error import HTTPError
+
+import torch
+from safeds._config import _init_default_device
+from safeds.data.image.containers._single_size_image_list import _SingleSizeImageList
+from safeds.data.labeled.containers import ImageDataset
+from safeds.data.tabular.containers import Column
+
+if TYPE_CHECKING:
+    from safeds.data.image.containers import ImageList
+
+_mnist_links: list[str] = ["http://yann.lecun.com/exdb/mnist/", "https://ossci-datasets.s3.amazonaws.com/mnist/"]
+_mnist_files: dict[str, str] = {
+    "train-images-idx3": "train-images-idx3-ubyte.gz",
+    "train-labels-idx1": "train-labels-idx1-ubyte.gz",
+    "test-images-idx3": "t10k-images-idx3-ubyte.gz",
+    "test-labels-idx1": "t10k-labels-idx1-ubyte.gz",
+}
+_mnist_labels: dict[int, str] = {0: "0", 1: "1", 2: "2", 3: "3", 4: "4", 5: "5", 6: "6", 7: "7", 8: "8", 9: "9"}
+_mnist_folder: str = "mnist"
+
+_fashion_mnist_links: list[str] = ["http://fashion-mnist.s3-website.eu-central-1.amazonaws.com/"]
+_fashion_mnist_files: dict[str, str] = _mnist_files
+_fashion_mnist_labels: dict[int, str] = {
+    0: "T-shirt/top",
+    1: "Trouser",
+    2: "Pullover",
+    3: "Dress",
+    4: "Coat",
+    5: "Sandal",
+    6: "Shirt",
+    7: "Sneaker",
+    8: "Bag",
+    9: "Ankle boot",
+}
+_fashion_mnist_folder: str = "fashion-mnist"
+
+_kuzushiji_mnist_links: list[str] = ["http://codh.rois.ac.jp/kmnist/dataset/kmnist/"]
+_kuzushiji_mnist_files: dict[str, str] = _mnist_files
+_kuzushiji_mnist_labels: dict[int, str] = {
+    0: "\u304a",
+    1: "\u304d",
+    2: "\u3059",
+    3: "\u3064",
+    4: "\u306a",
+    5: "\u306f",
+    6: "\u307e",
+    7: "\u3084",
+    8: "\u308c",
+    9: "\u3092",
+}
+_kuzushiji_mnist_folder: str = "kmnist"
+
+
+def load_mnist(path: str | Path, download: bool = True) -> tuple[ImageDataset[Column], ImageDataset[Column]]:
+    """
+    Load the `MNIST <http://yann.lecun.com/exdb/mnist/>`_ datasets.
+
+    Parameters
+    ----------
+    path:
+        the path were the files are stored or will be downloaded to
+    download:
+        whether the files should be downloaded to the given path
+
+    Returns
+    -------
+    train_dataset, test_dataset:
+        The train and test datasets.
+
+    Raises
+    ------
+    FileNotFoundError
+        if a file of the dataset cannot be found
+    """
+    path = Path(path) / _mnist_folder
+    path.mkdir(parents=True, exist_ok=True)
+    path_files = os.listdir(path)
+    missing_files = []
+    for file_path in _mnist_files.values():
+        if file_path not in path_files:
+            missing_files.append(file_path)
+    if len(missing_files) > 0:
+        if download:
+            _download_mnist_like(
+                path,
+                {name: f_path for name, f_path in _mnist_files.items() if f_path in missing_files},
+                _mnist_links,
+            )
+        else:
+            raise FileNotFoundError(f"Could not find files {[str(path / file) for file in missing_files]}")
+    return _load_mnist_like(path, _mnist_files, _mnist_labels)
+
+
+def load_fashion_mnist(path: str | Path, download: bool = True) -> tuple[ImageDataset[Column], ImageDataset[Column]]:
+    """
+    Load the `Fashion-MNIST <https://github.com/zalandoresearch/fashion-mnist>`_ datasets.
+
+    Parameters
+    ----------
+    path:
+        the path were the files are stored or will be downloaded to
+    download:
+        whether the files should be downloaded to the given path
+
+    Returns
+    -------
+    train_dataset, test_dataset:
+        The train and test datasets.
+
+    Raises
+    ------
+    FileNotFoundError
+        if a file of the dataset cannot be found
+    """
+    path = Path(path) / _fashion_mnist_folder
+    path.mkdir(parents=True, exist_ok=True)
+    path_files = os.listdir(path)
+    missing_files = []
+    for file_path in _fashion_mnist_files.values():
+        if file_path not in path_files:
+            missing_files.append(file_path)
+    if len(missing_files) > 0:
+        if download:
+            _download_mnist_like(
+                path,
+                {name: f_path for name, f_path in _fashion_mnist_files.items() if f_path in missing_files},
+                _fashion_mnist_links,
+            )
+        else:
+            raise FileNotFoundError(f"Could not find files {[str(path / file) for file in missing_files]}")
+    return _load_mnist_like(path, _fashion_mnist_files, _fashion_mnist_labels)
+
+
+def load_kmnist(path: str | Path, download: bool = True) -> tuple[ImageDataset[Column], ImageDataset[Column]]:
+    """
+    Load the `Kuzushiji-MNIST <https://github.com/rois-codh/kmnist>`_ datasets.
+
+    Parameters
+    ----------
+    path:
+        the path were the files are stored or will be downloaded to
+    download:
+        whether the files should be downloaded to the given path
+
+    Returns
+    -------
+    train_dataset, test_dataset:
+        The train and test datasets.
+
+    Raises
+    ------
+    FileNotFoundError
+        if a file of the dataset cannot be found
+    """
+    path = Path(path) / _kuzushiji_mnist_folder
+    path.mkdir(parents=True, exist_ok=True)
+    path_files = os.listdir(path)
+    missing_files = []
+    for file_path in _kuzushiji_mnist_files.values():
+        if file_path not in path_files:
+            missing_files.append(file_path)
+    if len(missing_files) > 0:
+        if download:
+            _download_mnist_like(
+                path,
+                {name: f_path for name, f_path in _kuzushiji_mnist_files.items() if f_path in missing_files},
+                _kuzushiji_mnist_links,
+            )
+        else:
+            raise FileNotFoundError(f"Could not find files {[str(path / file) for file in missing_files]}")
+    return _load_mnist_like(path, _kuzushiji_mnist_files, _kuzushiji_mnist_labels)
+
+
+def _load_mnist_like(
+    path: str | Path,
+    files: dict[str, str],
+    labels: dict[int, str],
+) -> tuple[ImageDataset[Column], ImageDataset[Column]]:
+    _init_default_device()
+
+    path = Path(path)
+    test_labels: Column | None = None
+    train_labels: Column | None = None
+    test_image_list: ImageList | None = None
+    train_image_list: ImageList | None = None
+    for file_name, file_path in files.items():
+        if "idx1" in file_name:
+            with gzip.open(path / file_path, mode="rb") as label_file:
+                magic, size = struct.unpack(">II", label_file.read(8))
+                if magic != 2049:
+                    raise ValueError(f"Magic number mismatch. Actual {magic} != Expected 2049.")  # pragma: no cover
+                if "train" in file_name:
+                    train_labels = Column(
+                        file_name,
+                        [labels[label_index] for label_index in array("B", label_file.read())],
+                    )
+                else:
+                    test_labels = Column(
+                        file_name,
+                        [labels[label_index] for label_index in array("B", label_file.read())],
+                    )
+        else:
+            with gzip.open(path / file_path, mode="rb") as image_file:
+                magic, size, rows, cols = struct.unpack(">IIII", image_file.read(16))
+                if magic != 2051:
+                    raise ValueError(f"Magic number mismatch. Actual {magic} != Expected 2051.")  # pragma: no cover
+                image_data = array("B", image_file.read())
+                image_tensor = torch.empty(size, 1, rows, cols, dtype=torch.uint8)
+                for i in range(size):
+                    image_tensor[i, 0] = torch.frombuffer(
+                        image_data[i * rows * cols : (i + 1) * rows * cols],
+                        dtype=torch.uint8,
+                    ).reshape(rows, cols)
+                image_list = _SingleSizeImageList()
+                image_list._tensor = image_tensor
+                image_list._tensor_positions_to_indices = list(range(size))
+                image_list._indices_to_tensor_positions = image_list._calc_new_indices_to_tensor_positions()
+                if "train" in file_name:
+                    train_image_list = image_list
+                else:
+                    test_image_list = image_list
+    if train_image_list is None or test_image_list is None or train_labels is None or test_labels is None:
+        raise ValueError  # pragma: no cover
+    return ImageDataset[Column](train_image_list, train_labels, 32, shuffle=True), ImageDataset[Column](
+        test_image_list,
+        test_labels,
+        32,
+    )
+
+
+def _download_mnist_like(path: str | Path, files: dict[str, str], links: list[str]) -> None:
+    path = Path(path)
+    for file_name, file_path in files.items():
+        for link in links:
+            try:
+                print(f"Trying to download file {file_name} via {link + file_path}")  # noqa: T201
+                urllib.request.urlretrieve(link + file_path, path / file_path, reporthook=_report_download_progress)
+                print()  # noqa: T201
+                break
+            except HTTPError as e:
+                print(f"An error occurred while downloading: {e}")  # noqa: T201  # pragma: no cover
+
+
+def _report_download_progress(current_packages: int, package_size: int, file_size: int) -> None:
+    percentage = min(((current_packages * package_size) / file_size) * 100, 100)
+    sys.stdout.write(f"\rDownloading... {percentage:.1f}%")
+    sys.stdout.flush()

tests/safeds_datasets/image/_mnist/test_mnist.py

@@ -0,0 +1,98 @@
+import os
+import tempfile
+from pathlib import Path
+
+import pytest
+import torch
+from safeds.data.labeled.containers import ImageDataset
+from safeds_datasets.image import _mnist, load_fashion_mnist, load_kmnist, load_mnist
+
+
+class TestMNIST:
+
+    def test_should_download_and_return_mnist(self) -> None:
+        with tempfile.TemporaryDirectory() as tmpdirname:
+            train, test = load_mnist(tmpdirname, download=True)
+            files = os.listdir(Path(tmpdirname) / _mnist._mnist._mnist_folder)
+            for mnist_file in _mnist._mnist._mnist_files.values():
+                assert mnist_file in files
+            assert isinstance(train, ImageDataset)
+            assert isinstance(test, ImageDataset)
+            assert len(train) == 60_000
+            assert len(test) == 10_000
+            assert (
+                train.get_input()._as_single_size_image_list()._tensor.dtype
+                == test.get_input()._as_single_size_image_list()._tensor.dtype
+                == torch.uint8
+            )
+            train_output = train.get_output()
+            test_output = test.get_output()
+            assert (
+                set(train_output.get_distinct_values())
+                == set(test_output.get_distinct_values())
+                == set(_mnist._mnist._mnist_labels.values())
+            )
+
+    def test_should_raise_if_file_not_found(self) -> None:
+        with tempfile.TemporaryDirectory() as tmpdirname, pytest.raises(FileNotFoundError):
+            load_mnist(tmpdirname, download=False)
+
+
+class TestFashionMNIST:
+
+    def test_should_download_and_return_mnist(self) -> None:
+        with tempfile.TemporaryDirectory() as tmpdirname:
+            train, test = load_fashion_mnist(tmpdirname, download=True)
+            files = os.listdir(Path(tmpdirname) / _mnist._mnist._fashion_mnist_folder)
+            for mnist_file in _mnist._mnist._fashion_mnist_files.values():
+                assert mnist_file in files
+            assert isinstance(train, ImageDataset)
+            assert isinstance(test, ImageDataset)
+            assert len(train) == 60_000
+            assert len(test) == 10_000
+            assert (
+                train.get_input()._as_single_size_image_list()._tensor.dtype
+                == test.get_input()._as_single_size_image_list()._tensor.dtype
+                == torch.uint8
+            )
+            train_output = train.get_output()
+            test_output = test.get_output()
+            assert (
+                set(train_output.get_distinct_values())
+                == set(test_output.get_distinct_values())
+                == set(_mnist._mnist._fashion_mnist_labels.values())
+            )
+
+    def test_should_raise_if_file_not_found(self) -> None:
+        with tempfile.TemporaryDirectory() as tmpdirname, pytest.raises(FileNotFoundError):
+            load_fashion_mnist(tmpdirname, download=False)
+
+
+class TestKMNIST:
+
+    def test_should_download_and_return_mnist(self) -> None:
+        with tempfile.TemporaryDirectory() as tmpdirname:
+            train, test = load_kmnist(tmpdirname, download=True)
+            files = os.listdir(Path(tmpdirname) / _mnist._mnist._kuzushiji_mnist_folder)
+            for mnist_file in _mnist._mnist._kuzushiji_mnist_files.values():
+                assert mnist_file in files
+            assert isinstance(train, ImageDataset)
+            assert isinstance(test, ImageDataset)
+            assert len(train) == 60_000
+            assert len(test) == 10_000
+            assert (
+                train.get_input()._as_single_size_image_list()._tensor.dtype
+                == test.get_input()._as_single_size_image_list()._tensor.dtype
+                == torch.uint8
+            )
+            train_output = train.get_output()
+            test_output = test.get_output()
+            assert (
+                set(train_output.get_distinct_values())
+                == set(test_output.get_distinct_values())
+                == set(_mnist._mnist._kuzushiji_mnist_labels.values())
+            )
+
+    def test_should_raise_if_file_not_found(self) -> None:
+        with tempfile.TemporaryDirectory() as tmpdirname, pytest.raises(FileNotFoundError):
+            load_kmnist(tmpdirname, download=False)