ChEMBL-SM.py

import json

import datasets
import pandas as pd

class ChemblSMConfig(datasets.BuilderConfig):

    def __init__(self, features, data_url, **kwargs):
        super(ChemblSMConfig, self).__init__(**kwargs)
        self.features = features
        self.data_url = data_url

class ChemblSM(datasets.GeneratorBasedBuilder):

    BUILDER_CONFIGS = [
        ChemblSMConfig(
            name="pairs",
            features={
                "column1": datasets.Value("string"),
                "column2": datasets.Value("string"),
                "label"  : datasets.Value("bool") 
            },
            data_url="https://huggingface.co/datasets/matchbench/ChEMBL-SM/resolve/main/matches.txt"
        ),
        ChemblSMConfig(
            name="source", 
            features={
                "json": datasets.Value("string")
            }, 
            data_url="https://huggingface.co/datasets/matchbench/ChEMBL-SM/resolve/main/chembl-sm-1.csv"
        ),
        ChemblSMConfig(
            name="target", 
            features={
                "json": datasets.Value("string")
            }, 
            data_url="https://huggingface.co/datasets/matchbench/ChEMBL-SM/resolve/main/chembl-sm-2.csv"
        ),
    ]

    def _info(self):
        return datasets.DatasetInfo(
            features=datasets.Features(self.config.features)
        )

    def _split_generators(self, dl_manager):
        path = dl_manager.download(self.config.data_url)
        if self.config.name == "pairs":
            return [
                datasets.SplitGenerator(
                    name=split,
                    gen_kwargs={
                        "file_path": path
                    }
                )
                for split in [
                    datasets.Split.TEST
                ]
            ]
        elif self.config.name == "source":
            return [
                datasets.SplitGenerator(
                    name="source",
                    gen_kwargs={
                        "file_path": path
                    }
                )
            ]
        elif self.config.name == "target":
            return [
                datasets.SplitGenerator(
                    name="target",
                    gen_kwargs={
                        "file_path": path
                    }
                )
            ]

    def _generate_examples(self, file_path):
        if self.config.name == "pairs":
            with open(file_path, "r") as f:
                md = {}
                for _, line in enumerate(f):
                    t = line.strip().split(',')
                    if t[0] not in md:
                        md[t[0]] = {t[1]}
                    else:
                        md[t[0]].add(t[1])
            id = -1
            for k, v in md.items():
                for i in v:
                    id = id + 1
                    yield id, {"column1": k, "column2": i, "label": True}
        elif self.config.name == "source":
            with open(file_path, "r") as f:
                source = pd.read_csv(file_path)
                source = source.to_json()
            yield 0, {"json": source}
        elif self.config.name == "target":
            with open(file_path, "r") as f:
                target = pd.read_csv(file_path)
                target = target.to_json()
            yield 0, {"json": target}