script-generation.py

from pathlib import Path

import datasets
import pandas as pd

_VERSION = "1.2.1"

_CITATION = f"""
@dataset{{unsplash-lite-dataset,
  title  = {{Unsplash Lite Dataset {_VERSION}}},
  url    = {{\\url{{https://github.com/unsplash/datasets}}}},
  author = {{Unsplash}},
  year   = {{2023}},
  month  = {{May}},
  day    = {{02}},
}}
"""

_DESCRIPTION = """
This dataset, available for commercial and noncommercial usage,
contains 25k nature-themed Unsplash photos, 25k keywords, and 1M searches.
"""

_HOMEPAGE = f"https://github.com/unsplash/datasets/tree/{_VERSION}"

_URL = f"https://unsplash.com/data/lite/{_VERSION}"

_LICENSE = "Unsplash Dataset License"

_TSV = (
    "collections",
    "colors",
    "conversions",
    "keywords",
    "photos",
)

_FEATURES = datasets.Features(
    {
        "photo": {
            "id": datasets.Value("string"),
            "url": datasets.Value("string"),
            "image_url": datasets.Value("string"),
            "submitted_at": datasets.Value("string"),
            "featured": datasets.Value("bool"),
            "width": datasets.Value("uint16"),
            "height": datasets.Value("uint16"),
            "aspect_ratio": datasets.Value("float32"),
            "description": datasets.Value("string"),
            "blur_hash": datasets.Value("string"),
        },
        "photographer": {
            "username": datasets.Value("string"),
            "first_name": datasets.Value("string"),
            "last_name": datasets.Value("string"),
        },
        "exif": {
            "camera_make": datasets.Value("string"),
            "camera_model": datasets.Value("string"),
            "iso": datasets.Value("string"),
            "aperture_value": datasets.Value("string"),
            "focal_length": datasets.Value("string"),
            "exposure_time": datasets.Value("string"),
        },
        "location": {
            "name": datasets.Value("string"),
            "latitude": datasets.Value("float32"),
            "longitude": datasets.Value("float32"),
            "country": datasets.Value("string"),
            "city": datasets.Value("string"),
        },
        "stats": {
            "views": datasets.Value("uint32"),
            "downloads": datasets.Value("uint32"),
        },
        "ai": {
            "description": datasets.Value("string"),
            "primary_landmark_name": datasets.Value("string"),
            "primary_landmark_latitude": datasets.Value("string"),
            "primary_landmark_longitude": datasets.Value("string"),
            "primary_landmark_confidence": datasets.Value("string"),
        },
        "keywords": [
            {
                "keyword": datasets.Value("string"),
                "ai_service_1_confidence": datasets.Value("string"),
                "ai_service_2_confidence": datasets.Value("string"),
                "suggested_by_user": datasets.Value("bool"),
            },
        ],
        "collections": [
            {
                "collection_id": datasets.Value("string"),
                "collection_title": datasets.Value("string"),
                "photo_collected_at": datasets.Value("string"),
            },
        ],
        "conversions": [
            {
                "converted_at": datasets.Value("string"),
                "conversion_type": datasets.Value("string"),
                "keyword": datasets.Value("string"),
                "anonymous_user_id": datasets.Value("string"),
                "conversion_country": datasets.Value("string"),
            },
        ],
        "colors": [
            {
                "hex": datasets.Value("string"),
                "red": datasets.Value("uint8"),
                "green": datasets.Value("uint8"),
                "blue": datasets.Value("uint8"),
                "keyword": datasets.Value("string"),
                "ai_coverage": datasets.Value("float32"),
                "ai_score": datasets.Value("float32"),
            },
        ],
    },
)

def df_withprefix(df, prefix, exclude=None):
    columns = [col for col in df.columns if col.startswith(prefix)]
    if exclude is not None:
        columns = [col for col in columns if exclude not in col]
    if "photo_id" not in columns:
        columns.append("photo_id")

    return df[columns].rename(columns=lambda col: col.removeprefix(prefix))

class Unsplash(datasets.GeneratorBasedBuilder):
    """The Unsplash Lite dataset."""

    DEFAULT_WRITER_BATCH_SIZE = 100

    def _info(self):
        return datasets.DatasetInfo(
            features=_FEATURES,
            supervised_keys=None,
            description=_DESCRIPTION,
            homepage=_HOMEPAGE,
            license=_LICENSE,
            version=_VERSION,
            citation=_CITATION,
        )

    def _split_generators(self, dl_manager):
        archive_path = Path(dl_manager.download_and_extract(_URL))

        # read all tsv files
        dataframes = {}
        for doc in _TSV:
            # read all tsv files for this document type
            frames = []
            for filename in archive_path.glob(f"{doc}.tsv*"):
                frame = pd.read_csv(filename, sep="\t", header=0)
                frames.append(frame)

            # concatenate all subframes into one
            concat_frames = pd.concat(frames, axis=0, ignore_index=True)

            if doc != "photos":
                dataframes[doc] = concat_frames
            else:
                # split "photos" into "photo", "photographer", "exif", "location", "stats", "ai"
                dataframes["photo"] = df_withprefix(concat_frames, "photo_", "location")
                dataframes["photo"]["blur_hash"] = concat_frames["blur_hash"]
                dataframes["photographer"] = df_withprefix(concat_frames, "photographer_")
                dataframes["exif"] = df_withprefix(concat_frames, "exif_")
                dataframes["location"] = df_withprefix(concat_frames, "photo_location_")
                dataframes["stats"] = df_withprefix(concat_frames, "stats_")
                dataframes["ai"] = df_withprefix(concat_frames, "ai_")

        # preprocess some columns
        dataframes["photo"]["featured"] = dataframes["photo"]["featured"].map({"t": True, "f": False})
        dataframes["keywords"]["suggested_by_user"] = dataframes["keywords"]["suggested_by_user"].map({"t": True, "f": False})

        # cast columns to appropriate dtypes
        for doc in dataframes.keys():
            if doc in _TSV:
                features = _FEATURES[doc][0]
            else:
                features = _FEATURES[doc]

            dataframes[doc].astype({
                key: features[key].dtype
                for key in features.keys()
            })

        # groupby "photo_id" if not "photo" dataframe
        for key in _TSV[:-1]:
            dataframes[key] = dataframes[key].groupby("photo_id")

        return [
            datasets.SplitGenerator(
                name=datasets.Split.TRAIN,
                gen_kwargs={"dataframes": dataframes},
            ),
        ]

    def _generate_examples(self, dataframes):
        # iterate over rows of "photos" dataframe
        photo_id_frames = {}
        for index, row in dataframes["photo"].iterrows():
            photo_id = row["id"]
            photographer = dataframes["photographer"].iloc[index]
            exif = dataframes["exif"].iloc[index]
            location = dataframes["location"].iloc[index]
            stats = dataframes["stats"].iloc[index]
            ai = dataframes["ai"].iloc[index]

            for key in _TSV[:-1]:
                try:
                    photo_id_frames[key] = dataframes[key].get_group(photo_id)
                except:
                    photo_id_frames[key] = pd.DataFrame()

            data = {
                "photo": {
                    "id": photo_id,
                    "url": row["url"],
                    "image_url": row["image_url"],
                    "submitted_at": row["submitted_at"],
                    "featured": row["featured"],
                    "width": row["width"],
                    "height": row["height"],
                    "aspect_ratio": row["aspect_ratio"],
                    "description": row["description"],
                    "blur_hash": row["blur_hash"],
                },
                "photographer": {
                    "username": photographer["username"],
                    "first_name": photographer["first_name"],
                    "last_name": photographer["last_name"],
                },
                "exif": {
                    "camera_make": exif["camera_make"],
                    "camera_model": exif["camera_model"],
                    "iso": exif["iso"],
                    "aperture_value": exif["aperture_value"],
                    "focal_length": exif["focal_length"],
                    "exposure_time": exif["exposure_time"],
                },
                "location": {
                    "name": location["name"],
                    "latitude": location["latitude"],
                    "longitude": location["longitude"],
                    "country": location["country"],
                    "city": location["city"],
                },
                "stats": {
                    "views": stats["views"],
                    "downloads": stats["downloads"],
                },
                "ai": {
                    "description": ai["description"],
                    "primary_landmark_name": ai["primary_landmark_name"],
                    "primary_landmark_latitude": ai["primary_landmark_latitude"],
                    "primary_landmark_longitude": ai["primary_landmark_longitude"],
                    "primary_landmark_confidence": ai["primary_landmark_confidence"],
                },
                "keywords": [
                    {
                        "keyword": keyword["keyword"],
                        "ai_service_1_confidence": keyword["ai_service_1_confidence"],
                        "ai_service_2_confidence": keyword["ai_service_2_confidence"],
                        "suggested_by_user": keyword["suggested_by_user"],
                    }
                    for _, keyword in photo_id_frames["keywords"].iterrows()
                ],
                "collections": [
                    {
                        "collection_id": collection["collection_id"],
                        "collection_title": str(collection["collection_title"]),
                        "photo_collected_at": collection["photo_collected_at"],
                    }
                    for _, collection in photo_id_frames["collections"].iterrows()
                ],
                "conversions": [
                    {
                        "converted_at": conversion["converted_at"],
                        "conversion_type": conversion["conversion_type"],
                        "keyword": conversion["keyword"],
                        "anonymous_user_id": conversion["anonymous_user_id"],
                        "conversion_country": str(conversion["conversion_country"]),
                    }
                    for _, conversion in photo_id_frames["conversions"].iterrows()
                ],
                "colors": [
                    {
                        "hex": color["hex"],
                        "red": color["red"],
                        "green": color["green"],
                        "blue": color["blue"],
                        "keyword": color["keyword"],
                        "ai_coverage": color["ai_coverage"],
                        "ai_score": color["ai_score"],
                    }
                    for _, color in photo_id_frames["colors"].iterrows()
                ],
            }

            yield index, data