Add MovieLens 1M dataset

CHANGELOG.md

@@ -8,6 +8,7 @@ The format is based on [Keep a Changelog](http://keepachangelog.com/en/1.0.0/).
 
 ### Added
 
+- Added `MovieLens 1M` dataset ([#397](https://github.com/pyg-team/pytorch-frame/pull/397))
 - Added light-weight MLP ([#372](https://github.com/pyg-team/pytorch-frame/pull/372))
 - Added R^2 metric ([#403](https://github.com/pyg-team/pytorch-frame/pull/403))
 

test/datasets/test_movielens_1m.py

@@ -0,0 +1,51 @@
+import tempfile
+
+import torch
+
+import torch_frame
+from torch_frame.config.text_embedder import TextEmbedderConfig
+from torch_frame.data.stats import StatType
+from torch_frame.datasets import Movielens1M
+from torch_frame.testing.text_embedder import HashTextEmbedder
+
+
+def test_movielens_1m():
+    with tempfile.TemporaryDirectory() as temp_dir:
+        dataset = Movielens1M(
+            temp_dir,
+            col_to_text_embedder_cfg=TextEmbedderConfig(
+                text_embedder=HashTextEmbedder(10)),
+        )
+    assert str(dataset) == 'Movielens1M()'
+    assert len(dataset) == 1000209
+    assert dataset.feat_cols == [
+        'user_id', 'gender', 'age', 'occupation', 'zip', 'movie_id', 'title',
+        'genres', 'timestamp'
+    ]
+
+    dataset = dataset.materialize()
+
+    tensor_frame = dataset.tensor_frame
+    assert len(tensor_frame.feat_dict) == 4
+    assert tensor_frame.feat_dict[torch_frame.categorical].dtype == torch.int64
+    assert tensor_frame.feat_dict[torch_frame.categorical].size() == (1000209,
+                                                                      6)
+    assert tensor_frame.feat_dict[
+        torch_frame.multicategorical].dtype == torch.int64
+    assert tensor_frame.feat_dict[torch_frame.embedding].dtype == torch.float32
+    assert tensor_frame.col_names_dict == {
+        torch_frame.categorical:
+        ['age', 'gender', 'movie_id', 'occupation', 'user_id', 'zip'],
+        torch_frame.multicategorical: ['genres'],
+        torch_frame.timestamp: ['timestamp'],
+        torch_frame.embedding: ['title'],
+    }
+    assert tensor_frame.y.size() == (1000209, )
+    assert tensor_frame.y.min() == 1 and tensor_frame.y.max() == 5
+
+    col_stats = dataset.col_stats
+    assert len(col_stats) == 10
+    assert StatType.COUNT in col_stats['user_id']
+    assert StatType.MULTI_COUNT in col_stats['genres']
+    assert StatType.YEAR_RANGE in col_stats['timestamp']
+    assert StatType.EMB_DIM in col_stats['title']

torch_frame/datasets/__init__.py

@@ -15,6 +15,7 @@
 from .data_frame_benchmark import DataFrameBenchmark
 from .data_frame_text_benchmark import DataFrameTextBenchmark
 from .mercari import Mercari
+from .movielens_1m import Movielens1M
 from .amazon_fine_food_reviews import AmazonFineFoodReviews
 from .diamond_images import DiamondImages
 from .huggingface_dataset import HuggingFaceDatasetDict
@@ -34,6 +35,7 @@
     'DataFrameBenchmark',
     'DataFrameTextBenchmark',
     'Mercari',
+    'Movielens1M',
     'AmazonFineFoodReviews',
     'DiamondImages',
 ]

torch_frame/datasets/movielens_1m.py

@@ -0,0 +1,90 @@
+from __future__ import annotations
+
+import os.path as osp
+import zipfile
+
+import pandas as pd
+
+import torch_frame
+from torch_frame.config.text_embedder import TextEmbedderConfig
+
+
+class Movielens1M(torch_frame.data.Dataset):
+    r"""The MovieLens 1M rating dataset, assembled by GroupLens Research
+    from the MovieLens web site, consisting of movies (3,883 nodes) and
+    users (6,040 nodes) with approximately 1 million ratings between them.
+
+    **STATS:**
+
+    .. list-table::
+        :widths: 10 10 10 10 20
+        :header-rows: 1
+
+        * - #Users
+          - #Items
+          - #User Field
+          - #Item Field
+          - #Samples
+        * - 6040
+          - 3952
+          - 5
+          - 3
+          - 1000209
+    """
+
+    url = 'https://files.grouplens.org/datasets/movielens/ml-1m.zip'
+
+    def __init__(
+        self,
+        root: str,
+        col_to_text_embedder_cfg: dict[str, TextEmbedderConfig]
+        | TextEmbedderConfig | None = None,
+    ):
+        path = self.download_url(self.url, root)
+        folder_path = osp.dirname(path)
+
+        with zipfile.ZipFile(path, 'r') as zip_ref:
+            zip_ref.extractall(folder_path)
+
+        data_path = osp.join(folder_path, 'ml-1m')
+        users = pd.read_csv(
+            osp.join(data_path, 'users.dat'),
+            header=None,
+            names=['user_id', 'gender', 'age', 'occupation', 'zip'],
+            sep='::',
+            engine='python',
+        )
+        movies = pd.read_csv(
+            osp.join(data_path, 'movies.dat'),
+            header=None,
+            names=['movie_id', 'title', 'genres'],
+            sep='::',
+            engine='python',
+            encoding='ISO-8859-1',
+        )
+        ratings = pd.read_csv(
+            osp.join(data_path, 'ratings.dat'),
+            header=None,
+            names=['user_id', 'movie_id', 'rating', 'timestamp'],
+            sep='::',
+            engine='python',
+        )
+
+        df = pd.merge(pd.merge(ratings, users), movies) \
+               .sort_values(by='timestamp') \
+               .reset_index().drop('index', axis=1)
+
+        col_to_stype = {
+            'user_id': torch_frame.categorical,
+            'gender': torch_frame.categorical,
+            'age': torch_frame.categorical,
+            'occupation': torch_frame.categorical,
+            'zip': torch_frame.categorical,
+            'movie_id': torch_frame.categorical,
+            'title': torch_frame.text_embedded,
+            'genres': torch_frame.multicategorical,
+            'rating': torch_frame.numerical,
+            'timestamp': torch_frame.timestamp,
+        }
+        super().__init__(df, col_to_stype, target_col='rating', col_to_sep='|',
+                         col_to_text_embedder_cfg=col_to_text_embedder_cfg)