Create abo_listings.py

abo_listings.py

@@ -0,0 +1,105 @@
+#!/usr/bin/env python3
+# -*- coding: utf-8 -*-
+import gzip
+import json
+from functools import partial
+from pathlib import Path
+
+import datasets
+from datasets import DatasetDict, load_dataset
+
+
+AVAILABLE_DATASETS = {
+    'main': 'https://amazon-berkeley-objects.s3.amazonaws.com/archives/abo-listings.tar'
+}
+
+VERSION = datasets.Version("0.0.1")
+
+_FIELDS = [
+    "item_id",
+    "brand",
+    "bullet_point",
+    "color",
+    "item_name",
+    "model_name",
+    "model_number",
+    "model_year",
+    "product_type",
+    "style",
+    "main_image_id",
+    "other_image_id",
+    "item_keywords",
+    "country",
+    "marketplace",
+    "domain_name",
+    "node",
+]
+
+
+class AbolistingsDataset(datasets.GeneratorBasedBuilder):
+    """AbolistingsDataset dataset."""
+
+    BUILDER_CONFIGS = [
+        datasets.BuilderConfig(
+            name=data_name, version=VERSION, description=f"{data_name} abolistings dataset"
+        )
+        for data_name in AVAILABLE_DATASETS
+    ]
+
+    @staticmethod
+    def load(data_name_config: str = "main") -> DatasetDict:
+        ds = load_dataset(__file__, data_name_config)
+        return ds
+
+    def _info(self):
+        return datasets.DatasetInfo(
+            description="",
+            features=datasets.Features(
+                {
+                    "item_id": datasets.Value("string"),
+                    "brand": datasets.Sequence(datasets.Value("string")),
+                    "bullet_point": datasets.Sequence(datasets.Value("string")),
+                    "color": datasets.Sequence(datasets.Value("string")),
+                    "item_name": datasets.Sequence(datasets.Value("string")),
+                    "model_name": datasets.Sequence(datasets.Value("string")),
+                    "model_number": datasets.Sequence(datasets.Value("string")),
+                    "model_year": datasets.Sequence(datasets.Value("string")),
+                    "product_type": datasets.Sequence(datasets.Value("string")),
+                    "style": datasets.Sequence(datasets.Value("string")),
+                    "main_image_id": datasets.Value("string"),
+                    "other_image_id": datasets.Sequence(datasets.Value("string")),
+                    "item_keywords": datasets.Sequence(datasets.Value("string")),
+                    "country": datasets.Value("string"),
+                    "marketplace": datasets.Value("string"),
+                    "domain_name": datasets.Value("string"),
+                    "node": datasets.Sequence(datasets.Value("string")),
+                }
+            ),
+            supervised_keys=None,
+            homepage="https://amazon-berkeley-objects.s3.amazonaws.com/index.html#download",
+            citation="",
+        )
+
+    def _split_generators(self, dl_manager):
+        downloader = partial(
+            lambda: dl_manager.download_and_extract(AVAILABLE_DATASETS[self.config.name])
+        )
+        # There is no predefined train/val/test split for this dataset.
+        root_path = Path(downloader()) / 'listings' / 'metadata'
+        return [
+            datasets.SplitGenerator(
+                name=datasets.Split.TRAIN, gen_kwargs={"root_path": root_path}
+            ),
+        ]
+
+    def _generate_examples(self, root_path):
+        root_path = Path(root_path)
+        files = list(root_path.glob("*.json.gz"))
+        idx = 0
+        for file in files:
+            with gzip.GzipFile(file) as f_in:
+                for l in f_in:
+                    l = l.decode("utf-8")
+                    sample = json.loads(l)
+                    yield idx, {k: sample.get(k) for k in _FIELDS}
+                    idx += 1