[AL-2078] Switch random split doc section (activeloopai#2064)

* fix * changed docstring * update * fix * Update deeplake/core/dataset/dataset.py * add indra import in random split * fix Co-authored-by: Fayaz Rahman <fayazrahman4u@gmail.com>
puyuanOT · Dec 21, 2022 · 384b074 · 384b074
1 parent 51c8a1b
commit 384b074
Show file tree

Hide file tree

Showing 5 changed files with 23 additions and 1 deletion.
diff --git a/deeplake/api/tests/test_api.py b/deeplake/api/tests/test_api.py
@@ -10,6 +10,7 @@
     assert_array_lists_equal,
     is_opt_true,
     get_dummy_data_path,
+    requires_libdeeplake,
 )
 from deeplake.tests.storage_fixtures import enabled_remote_storages
 from deeplake.core.storage import GCSProvider
@@ -2252,6 +2253,7 @@ def test_iter_warning(local_ds):
             ds.abc[10]
 
 
+@requires_libdeeplake
 def test_random_split(local_ds):
     with local_ds as ds:
         ds.create_tensor("label")

diff --git a/deeplake/core/dataset/dataset.py b/deeplake/core/dataset/dataset.py
@@ -3930,8 +3930,9 @@ def max_view(self):
         )
 
     def random_split(self, lengths: Sequence[Union[int, float]]):
-        """Splits the dataset into non-overlapping new datasets of given lengths.
+        """Splits the dataset into non-overlapping :class:`~deeplake.core.dataset.Dataset` objects of given lengths.
         If a list of fractions that sum up to 1 is given, the lengths will be computed automatically as floor(frac * len(dataset)) for each fraction provided.
+        The split generated is only performant with enterprise dataloader which can be installed with ``pip install deeplake[enterprise]``.
 
         After computing the lengths, if there are any remainders, 1 count will be distributed in round-robin fashion to the lengths until there are no remainders left.
 

diff --git a/deeplake/util/scheduling.py b/deeplake/util/scheduling.py
@@ -60,6 +60,9 @@ def create_fetching_schedule(dataset, primary_tensor_name, shuffle_within_chunks
 
 
 def create_random_split_views(dataset, lengths):
+    from deeplake.enterprise.convert_to_libdeeplake import import_indra_api
+
+    import_indra_api()
     if math.isclose(sum(lengths), 1) and sum(lengths) <= 1:
         subset_lengths: List[int] = []
         for i, frac in enumerate(lengths):

diff --git a/docs/source/Random-Split.rst b/docs/source/Random-Split.rst
@@ -0,0 +1,15 @@
+.. currentmodule:: deeplake.core.dataset
+
+Random Split
+=======
+
+Splits the dataset into non overlapping new datasets of given lengths.
+The resulting datasets are generated in such a way that when creating a dataloader from the view and training on it,
+the performance impact is minimal. Using the outputs of this function with .pytorch method of dataset (instead of .dataloader) may result in poor performance.
+See the ``random_split`` method on how to use this feature:
+
+.. autosummary::
+    :toctree:
+    :nosignatures:
+
+    Dataset.random_split
diff --git a/docs/source/index.rst b/docs/source/index.rst
@@ -39,6 +39,7 @@ Deep Lake is an open-source database for AI.
    Dataloader <Dataloader>
    Sampler <Sampler>
    Tensor Query Language <Tensor-Query-Language>
+   Random Split <Random-Split>
 
 .. toctree::
    :maxdepth: 1