diff --git a/docs/source/changes.rst b/docs/source/changelog.rst
similarity index 100%
rename from docs/source/changes.rst
rename to docs/source/changelog.rst
diff --git a/src/moxel/data.py b/src/moxel/data.py
new file mode 100644
index 0000000..8325e0b
--- /dev/null
+++ b/src/moxel/data.py
@@ -0,0 +1,65 @@
+r"""
+Write the docstring of the module.
+"""
+
+import json
+import torch
+from torch.utils.data import Dataset, random_split
+from . utils import load_json
+
+
+def prepare_data(source, split_ratio=(0.8, 0.1, 0.1), seed=1):
+    r"""
+    Split a set of materials into train, validation and test sets.
+
+    .. warning::
+        * You should use this function **after** :func:`utils.batch_clean`.
+        * No directory is created by :func:`prepare_data`. **All ``.json``
+        files are stored under the directory containing ``source``**.
+
+    Before the split::
+
+        voxels_data
+        ├──clean_voxels.npy
+        └──clean_names.json
+
+    After the split::
+
+        voxels_data
+        ├──clean_voxels.npy
+        ├──clean_names.json
+        ├──train.json
+        ├──validation.json
+        └──test.json
+
+    Each ``.json`` file stores the indices of ``clean_voxels.npy`` that will be
+    used for training, validation and testing.
+
+    Parameters
+    ----------
+    source: str
+        Pathname to the file holding the names of the materials
+        (``clean_names.json``).
+    split_ratio: sequence, default=(0.8, 0.1, 0.1)
+        The sizes or fractions of splits to be produced.
+        * ``split_ratio[0] == train``.
+        * ``split_ratio[1] == validation``.
+        * ``split_ratio[2] == test``.
+    seed : int, default=1
+        Controls the randomness of the ``rng`` used for splitting.
+    """
+    rng = torch.Generator().manual_seed(seed)
+    indices = range(len(load_json(source)))
+
+    train, val, test = random_split(indices, split_ratio, generator=rng)
+
+    for split, mode in zip((train, val, test), ('train', 'validation', 'test')):
+        mode_indices = list(split)
+        with open(os.path.join(path, f'{mode}.json'), 'w') as fhand:
+            json.dump(mode_indices, fhand, indent=4)
+
+    print('\033[32mData preparation completed!\033[0m')
+
+
+class VoxelsDataset(Dataset):
+    ...
diff --git a/tests/test_data.py b/tests/test_data.py
new file mode 100644
index 0000000..e69de29