How to Download
Browse files
README.md
CHANGED
@@ -64,7 +64,7 @@ configs:
|
|
64 |
|
65 |
**Industrial-scale music recommendation dataset with organic/recommendation interactions and audio embeddings**
|
66 |
|
67 |
-
[π Overview](#overview) β’ [π Key Features](#key-features) β’ [π Statistics](#statistics) β’ [π Format](#data-format) β’ [π Benchmark](#benchmark) β’ [β FAQ](#faq)
|
68 |
|
69 |
## Overview
|
70 |
|
@@ -169,6 +169,62 @@ Each dataset is also available in a user-aggregated sequential format with the f
|
|
169 |
|
170 |
Code for the baseline models can be found in `benchmarks/` directory, see [Reproducibility Guide](benchmarks/README.md)
|
171 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
172 |
## FAQ
|
173 |
|
174 |
### Are test items presented in training data?
|
@@ -201,4 +257,4 @@ A track is considered "listened" if over 50% of its duration is played.
|
|
201 |
|
202 |
A played_ratio_pct greater than 100% indicates that the user rewound and
|
203 |
replayed sections of the trackβso the total time listened exceeds the original track length.
|
204 |
-
These values are expected behavior and not log noise. See [discussion](https://huggingface.co/datasets/yandex/yambda/discussions/10)
|
|
|
64 |
|
65 |
**Industrial-scale music recommendation dataset with organic/recommendation interactions and audio embeddings**
|
66 |
|
67 |
+
[π Overview](#overview) β’ [π Key Features](#key-features) β’ [π Statistics](#statistics) β’ [π Format](#data-format) β’ [π Benchmark](#benchmark) β’ [β¬οΈ Download](#download) β’ [β FAQ](#faq)
|
68 |
|
69 |
## Overview
|
70 |
|
|
|
169 |
|
170 |
Code for the baseline models can be found in `benchmarks/` directory, see [Reproducibility Guide](benchmarks/README.md)
|
171 |
|
172 |
+
### Download
|
173 |
+
|
174 |
+
Simplest way:
|
175 |
+
```python
|
176 |
+
from datasets import load_dataset
|
177 |
+
|
178 |
+
ds = load_dataset("yandex/yambda", data_dir="flat/50m", data_files="likes.parquet")
|
179 |
+
```
|
180 |
+
|
181 |
+
Also, we provide simple wrapper for convenient
|
182 |
+
usage
|
183 |
+
```python
|
184 |
+
from typing import Literal
|
185 |
+
|
186 |
+
from datasets import Dataset, DatasetDict, load_dataset
|
187 |
+
|
188 |
+
|
189 |
+
class YambdaDataset:
|
190 |
+
INTERACTIONS = frozenset(["likes", "listens", "multi_event", "dislikes", "unlikes", "undislikes"])
|
191 |
+
|
192 |
+
def __init__(
|
193 |
+
self, dataset_type: Literal["flat", "sequential"] = "flat", dataset_size: Literal["50m", "500m", "5b"] = "50m"
|
194 |
+
):
|
195 |
+
assert dataset_type in {"flat", "sequential"}
|
196 |
+
assert dataset_size in {"50m", "500m", "5b"}
|
197 |
+
|
198 |
+
self.dataset_type = dataset_type
|
199 |
+
self.dataset_size = dataset_size
|
200 |
+
|
201 |
+
def interaction(
|
202 |
+
self, type: Literal["likes", "listens", "multi_event", "dislikes", "unlikes", "undislikes"]
|
203 |
+
) -> Dataset:
|
204 |
+
assert type in YambdaDataset.INTERACTIONS
|
205 |
+
|
206 |
+
return YambdaDataset._download(f"{self.dataset_type}/{self.dataset_size}", type)
|
207 |
+
|
208 |
+
def audio_embeddings(self) -> Dataset:
|
209 |
+
return YambdaDataset._download("", "embeddings")
|
210 |
+
|
211 |
+
def album_item_mapping(self) -> Dataset:
|
212 |
+
return YambdaDataset._download("", "album_item_mapping")
|
213 |
+
|
214 |
+
def artist_item_mapping(self) -> Dataset:
|
215 |
+
return YambdaDataset._download("", "artist_item_mapping")
|
216 |
+
|
217 |
+
@staticmethod
|
218 |
+
def _download(data_dir: str, file: str) -> Dataset:
|
219 |
+
data = load_dataset("yandex/yambda", data_dir=data_dir, data_files=f"{file}.parquet")
|
220 |
+
assert isinstance(data, DatasetDict)
|
221 |
+
return data["train"]
|
222 |
+
|
223 |
+
dataset = YambdaDataset("flat", "50m")
|
224 |
+
|
225 |
+
likes = dataset.interaction("likes")
|
226 |
+
```
|
227 |
+
|
228 |
## FAQ
|
229 |
|
230 |
### Are test items presented in training data?
|
|
|
257 |
|
258 |
A played_ratio_pct greater than 100% indicates that the user rewound and
|
259 |
replayed sections of the trackβso the total time listened exceeds the original track length.
|
260 |
+
These values are expected behavior and not log noise. See [discussion](https://huggingface.co/datasets/yandex/yambda/discussions/10)
|