Import Steam interactions (Australia subset)

data/.gitignore

@@ -6,5 +6,6 @@
 *.zst
 ml-*/
 ms-web/
+steam
 !ml-latest-small/
 *.parquet

docs/releases/2026.rst

@@ -46,6 +46,7 @@ fixed issues and merged PRs is on `milestone 2026.3`_.
 - Certain per-request warnings in batch inference are now aggregated (:issue:`1113`, :pr:`1168`).
 - Added :class:`lenskit.metrics.UniqueItemCount` (:issue:`1005`, :pr:`1169`).
 - Added support for ``interaction="default"`` to :meth:`~lenskit.data.DatasetBuilder.add_relationship_class` (:pr:`1174`).
+- Fix Steam dataset import to properly import interactions (previously was incomplete and broken, :issue:`1176`, :pr:`1179`).
 
 .. _milestone 2026.3: https://github.com/lenskit/lkpy/milestone/28?closed=1
 

src/lenskit/data/sources/steam.py

@@ -94,7 +94,7 @@ def load_steam(*files: Path, reviews: bool = False) -> Dataset:
         return _load_au_steam(au_interactions, au_reviews if reviews else None)
 
     elif all_reviews is not None:
-        _log.debug("looking for full-data inteactions")
+        _log.debug("looking for full-data interactions")
         if au_reviews is not None:
             _log.error("cannot specify both Australian and overall input files")
             raise DataError("invalid combination of Steam input files")
@@ -142,7 +142,30 @@ def _load_au_steam(interactions: Path, reviews: Path | None) -> Dataset:
     dsb.add_entities("user", users)
 
     _log.info("adding user-item interactions")
-    # TODO: make DSB work better with CSR-shaped data
+    dsb.add_relationship_class(
+        "plays", ["user", "item"], allow_repeats=False, interaction="default"
+    )
+    for chunk in items.chunks:
+        assert isinstance(chunk, pa.ListArray)
+        # get the user that goes with each (nested) list element
+        chunk_user_nums = pc.list_parent_indices(chunk)
+        chunk_user_ids = ui_data.column("steam_id").take(chunk_user_nums)
+        # get the item IDs
+        assert isinstance(chunk.values, pa.StructArray)
+        chunk_item_ids = chunk.values.field("item_id")
+        assert len(chunk_user_ids) == len(chunk_item_ids)
+        chunk_playtime = chunk.values.field("playtime_forever")
+        chunk_playtime2 = chunk.values.field("playtime_2weeks")
+        chunk_tbl = pa.table(
+            {
+                "user_id": chunk_user_ids,
+                "item_id": chunk_item_ids,
+                "playtime": chunk_playtime,
+                "playtime_2weeks": chunk_playtime2,
+            }
+        )
+        _log.debug("adding %d interactions", chunk_tbl.num_rows)
+        dsb.add_interactions("plays", chunk_tbl, missing="error")
 
     return dsb.build()
 

tests/data/test_load_steam.py

@@ -0,0 +1,34 @@
+# This file is part of LensKit.
+# Copyright (C) 2018-2023 Boise State University.
+# Copyright (C) 2023-2026 Drexel University.
+# Licensed under the MIT license, see LICENSE.md for details.
+# SPDX-License-Identifier: MIT
+
+import os
+from pathlib import Path
+
+import pandas as pd
+import pyarrow as pa
+import pyarrow.compute as pc
+
+from pytest import mark
+
+from lenskit.data import Vocabulary
+from lenskit.data.sources.steam import load_steam
+
+STEAM_DIR = Path("data/steam")
+AU_FILE = STEAM_DIR / "australian_users_items.json.gz"
+
+
+@mark.skipif(not AU_FILE.exists(), reason="input data does not exist")
+@mark.realdata
+def test_steam_australia():
+    data = load_steam(AU_FILE)
+
+    # do we have the right number of entities?
+    assert 87000 < data.user_count < 89000
+    assert 10000 < data.item_count < 12000
+
+    # do we about the right number of interactions?
+    ints = data.interactions()
+    assert ints.count() >= 1_000_000