pest-management-opendata.py

import itertools as it
import collections as cl
from pathlib import Path
from dataclasses import dataclass, asdict
from urllib.parse import urlparse, urlunparse

import cv2
import boto3
import numpy as np
import pandas as pd
import awswrangler as wr
from datasets import (
    Split,
    Image,
    Value,
    Features,
    Sequence,
    ClassLabel,
    DatasetInfo,
    SplitGenerator,
    GeneratorBasedBuilder,
)
from shapely.wkt import loads

__version__ = '20220912-2056'

SplitInfo = cl.namedtuple('SplitInfo', 'dtype, basename, split')

@dataclass
class SplitPayload:
    split: str
    path: Path

    def __post_init__(self):
        self.path = Path(self.path)

    def to_frame(self):
        return (pd
                .read_csv(self.path, compression='gzip')
                .query(f'split == "{self.split}"'))

#
#
#
class SplitManager:
    _splits = tuple(it.starmap(SplitInfo, (
        (Split.TRAIN, 'dev', 'train'),
        (Split.VALIDATION, 'dev', 'val'),
        (Split.TEST, 'test', 'test'),
    )))

    @staticmethod
    def custom_download(url, path):
        remote = urlparse(url)
        name = Path(remote.path)
        if name.is_absolute():
            name = name.relative_to(name.parents[-1])

        s3 = boto3.client(remote.scheme)
        s3.download_file(remote.netloc, str(name), path)

    @property
    def labels(self):
        path = self.url('dev')
        df = wr.s3.read_csv(path, compression='gzip')
        yield from df['label'].dropna().unique()

    def __init__(self, bucket):
        self.bucket = bucket
        self.path = Path('metadata', __version__)

    def __call__(self, dl_manager):
        for i in self._splits:
            url = self.url(i.basename)
            path = dl_manager.download_custom(url, self.custom_download)
            payload = SplitPayload(i.split, path)

            yield SplitGenerator(name=i.dtype, gen_kwargs=asdict(payload))

    def url(self, split):
        path = self.path.joinpath(split).with_suffix('.csv.gz')
        source = self.bucket._replace(path=str(path))
        return urlunparse(source)

#
#
#
class ExampleManager:
    _decode_flags = cv2.IMREAD_COLOR | cv2.IMREAD_IGNORE_ORIENTATION
    # _Pest = cl.namedtuple('_Pest', 'label, geometry')
    # _Feature = cl.namedtuple('_Feature', 'image, pests')

    @staticmethod
    def features(labels):
        return Features({
            'image': Image(),
            'pests': Sequence({
                'label': ClassLabel(names=labels),
                'geometry': Value('binary'),
            }),
        })

    @staticmethod
    def pests(df):
        if 'geometry' in df.columns:
            for i in df.dropna().itertuples(index=False):
                geometry = loads(i.geometry)
                yield {
                    'label': i.label,
                    'geometry': geometry.wkb,
                }

    def __init__(self, payload):
        self.payload = payload

    def __iter__(self):
        df = self.payload.to_frame()
        for (i, g) in df.groupby('url', sort=False):
            value = {
                'image': self.load(urlparse(i)),
                'pests': list(self.pests(g)),
            }

            yield (i, value)

    def load(self, url):
        path = Path(url.path)
        if path.is_absolute():
            (*_, root) = path.parents
            path = path.relative_to(root)

        data = (boto3
                .resource(url.scheme)
                .Bucket(url.netloc)
                .Object(str(path))
                .get()
                .get('Body')
                .read())
        image = np.asarray(bytearray(data))

        return cv2.imdecode(image, self._decode_flags)

#
#
#
class PestManagementOpendata(GeneratorBasedBuilder):
    _bucket = urlparse('s3://wadhwaniai-agri-opendata')

    def _info(self):
        data = SplitManager(self._bucket)
        labels = sorted(data.labels)
        features = ExampleManager.features(labels)

        return DatasetInfo(
            homepage='https://github.com/WadhwaniAI/pest-management-opendata',
            # citation=_CITATION,
            # description=_DESCRIPTION,
            license='CC-BY 4.0',
            features=features,
        )

    def _split_generators(self, dl_manager):
        splits = SplitManager(self._bucket)
        return list(splits(dl_manager))

    def _generate_examples(self, **kwargs):
        payload = SplitPayload(**kwargs)
        examples = ExampleManager(payload)
        yield from examples