basic_sentence_transforms.py

# basic-sentence-transforms.py: the HF datasets "loading script" for the internal NC_PAT dataset (defines configurations/tasks, columns, etc.)
import os
import json
import datasets
from datasets import Split, SplitGenerator

# relative path was not working here, so switched to absolute path
DATASET_URL = "https://huggingface.co/datasets/rfernand/basic_sentence_transforms/resolve/main"

no_extra = {
    "source": datasets.Value("string"),
    "target": datasets.Value("string"),
}

samp_class = {
    "source": datasets.Value("string"),
    "target": datasets.Value("string"),
    "class": datasets.Value("string"),
}

count_class = {
    "source": datasets.Value("string"),
    "target": datasets.Value("string"),
    "count": datasets.Value("string"),
    "class": datasets.Value("string"),
}

dir_only = {
    "source": datasets.Value("string"),
    "target": datasets.Value("string"),
    "direction": datasets.Value("string"),
}

warmup_configs = [
    {"name": "car_cdr_cons",
    "desc": "small phrase translation tasks that require only: CAR, CDR, or CAR+CDR+CONS operations",
    "features": samp_class},

    {"name": "car_cdr_cons_tuc",
    "desc": "same task as car_cdr_cons, but requires mapping lowercase fillers to their uppercase tokens",
    "features": samp_class},

    {"name": "car_cdr_rcons",
    "desc": "same task as car_cdr_cons, but the CONS samples have their left/right children swapped",
    "features": samp_class},

    {"name": "car_cdr_rcons_tuc",
    "desc": "same task as car_cdr_rcons, but requires mapping lowercase fillers to their uppercase tokens",
    "features": samp_class},

    {"name": "car_cdr_seq",
    "desc": "each samples requires 1-4 combinations of CAR and CDR, as identified by the root filler token",
    "features": count_class},

    {"name": "car_cdr_seq_40k",
    "desc": "same task as car_cdr_seq, but train samples increased from 10K to 40K",
    "features": count_class},

    {"name": "car_cdr_seq_tuc",
    "desc": "same task as car_cdr_seq, but requires mapping lowercase fillers to their uppercase tokens",
    "features": count_class},

    {"name": "car_cdr_seq_40k_tuc",
    "desc": "same task as car_cdr_seq_tuc, but train samples increased from 10K to 40K",
    "features": count_class},

    {"name": "car_cdr_seq_path",
    "desc": "similiar to car_cdr_seq, but each needed operation in represented as a node in the left child of the root",
    "features": count_class},

    {"name": "car_cdr_seq_path_40k",
    "desc": "same task as car_cdr_seq_path, but train samples increased from 10K to 40K",
    "features": count_class},

    {"name": "car_cdr_seq_path_40k_tuc",
    "desc": "same task as car_cdr_seq_path_40k, but requires mapping lowercase fillers to their uppercase tokens",
    "features": count_class},

    {"name": "car_cdr_seq_path_tuc",
    "desc": "same task as car_cdr_seq_path, but requires mapping lowercase fillers to their uppercase tokens",
    "features": count_class},
]

core_configs = [

    {"name": "active_active_stb",
    "desc": "active sentence translation, from sentence to parenthesized tree form, both directions",
    "features": dir_only},
    
    {"name": "active_active_stb_40k",
    "desc": "same task as active_active_stb, but train samples increased from 10K to 40K",
    "features": dir_only},

    
    {"name": "active_logical_ssb",
    "desc": "active to logical sentence translation, in both directions",
    "features": dir_only},
    
    {"name": "active_logical_ssb_40k",
    "desc": "same task as active_logical_ssb, but train samples increased from 10K to 40K",
    "features": dir_only},
    
    {"name": "active_logical_ttb",
    "desc": "active to logical tree translation, in both directions",
    "features": dir_only},
    
    {"name": "active_logical_ttb_40k",
    "desc": "same task as active_logical_ttb, but train samples increased from 10K to 40K",
    "features": dir_only},


    {"name": "active_passive_ssb",
    "desc": "active to passive sentence translation, in both directions",
    "features": dir_only},
    
    {"name": "active_passive_ssb_40k",
    "desc": "same task as active_passive_ssb, but train samples increased from 10K to 40K",
    "features": dir_only},
    
    {"name": "active_passive_ttb",
    "desc": "active to passive tree translation, in both directions",
    "features": dir_only},
    
    {"name": "active_passive_ttb_40k",
    "desc": "same task as active_passive_ttb, but train samples increased from 10K to 40K",
    "features": dir_only},

    
    {"name": "actpass_logical_ss",
    "desc": "mixture of active to logical and passive to logical sentence translations, single direction",
    "features": no_extra},
    
    {"name": "actpass_logical_ss_40k",
    "desc": "same task as actpass_logical_ss, but train samples increased from 10K to 40K",
    "features": no_extra},
    
    {"name": "actpass_logical_tt",
    "desc": "mixture of active to logical and passive to logical tree translations, single direction",
    "features": no_extra},
    
    {"name": "actpass_logical_tt_40k",
    "desc": "same task as actpass_logical_tt, but train samples increased from 10K to 40K",
    "features": no_extra},
    
    {"name": "logical_logical_stb",
    "desc": "logical form sentence translation, from sentence to parenthesized tree form, both directions",
    "features": dir_only},
    
    {"name": "alogical_logical_stb_40k",
    "desc": "same task as logical_logical_stb, but train samples increased from 10K to 40K",
    "features": dir_only},

    
    {"name": "passive_logical_ssb",
    "desc": "passive to logical sentence translation, in both directions",
    "features": dir_only},
    
    {"name": "passive_logical_ssb_40k",
    "desc": "same task as passive_logical_ssb, but train samples increased from 10K to 40K",
    "features": dir_only},

    {"name": "passive_logical_ttb",
    "desc": "passive to logical tree translation, in both directions",
    "features": dir_only},
    
    {"name": "passive_logical_ttb_40k",
    "desc": "same task as passive_logical_ttb, but train samples increased from 10K to 40K",
    "features": dir_only},
    
    {"name": "passive_passive_stb",
    "desc": "passive sentence translation, from sentence to parenthesized tree form, both directions",
    "features": dir_only},
    
    {"name": "passive_passive_stb_40k",
    "desc": "same task as passive_passive_stb, but train samples increased from 10K to 40K",
    "features": dir_only},
]

configs = warmup_configs + core_configs

class BasicSentenceTransformsConfig(datasets.BuilderConfig):
    """BuilderConfig for basic_sentence_transforms dataset."""

    def __init__(self, features=None, **kwargs):
        # Version history:
        # 0.0.18: Initial version released to HF datasets
        # 0.0.21: release V21 of NC_PAT dataset 
        super().__init__(version=datasets.Version("0.0.21"), **kwargs)

        self.features = features
        self.label_classes = None
        #self.data_url = "./{}.zip".format(kwargs["name"])
        self.data_url ="{}/{}.zip".format(DATASET_URL, kwargs["name"])

        self.citation = None
        self.homepage = None

    def _info(self):
        return datasets.DatasetInfo(
            description=self.description,
            features=self.features,
            # No default supervised_keys (as we have to pass both question
            # and context as input).
            supervised_keys=None,
            homepage=self.homepage,
            citation=self.citation,
        )

class BasicSentenceTransforms(datasets.GeneratorBasedBuilder):
    BUILDER_CONFIGS = [BasicSentenceTransformsConfig(name=c["name"], description=c["desc"], features=c["features"]) for c in configs]
    VERSION = datasets.Version("0.0.21")

    def _info(self):
        # features are now required here, so get them from the current CONFIG (following code example from super_glue.py)
        features = {feature: datasets.Value("string") for feature in self.config.features}

        return datasets.DatasetInfo(
            description="The dataset consists of diagnostic/warm-up tasks and core tasks within this dataset." + 
                "The core tasks represent the translation of English sentences between the active, passive, and logical forms.",
            features=datasets.Features(features),
            supervised_keys=None,
            homepage=None,
            citation=None,
        )

    def _split_generators(self, dl_manager: datasets.DownloadManager):
        url = self.config.data_url
        #print("_split_generator: url={}".format(url))

        dl_dir = dl_manager.download_and_extract(url)
        task = self.config_id
        #print("task: {}, dl_dir: {}".format(task, dl_dir))

        splits = [
            SplitGenerator(name=Split.TRAIN, gen_kwargs={"data_file": os.path.join(dl_dir, "train.jsonl")}),
            SplitGenerator(name=Split.VALIDATION, gen_kwargs={"data_file": os.path.join(dl_dir, "dev.jsonl")}),
            SplitGenerator(name=Split.TEST, gen_kwargs={"data_file": os.path.join(dl_dir, "test.jsonl")}),
        ]

        if not task.startswith("car_cdr_cons") and not task.startswith("car_cdr_rcons"):
            splits += [
                SplitGenerator(name="ood_new", gen_kwargs={"data_file": os.path.join(dl_dir, "ood_new.jsonl")}),
                SplitGenerator(name="ood_long", gen_kwargs={"data_file": os.path.join(dl_dir, "ood_long.jsonl")}),
                SplitGenerator(name="ood_all", gen_kwargs={"data_file": os.path.join(dl_dir, "ood_all.jsonl")}),
            ]

        return splits

    def _generate_examples(self, data_file):
        #print("_generate_examples: data_file: {}".format(data_file))

        with open(data_file, encoding="utf-8") as f:
            for i, line in enumerate(f):
                key = str(i)
                row = json.loads(line)    
                yield key, row  

if __name__ == "__main__":
    # short test 
    builder = BasicSentenceTransforms.BUILDER_CONFIGS[0]
    print("name: {}, desc: {}".format(builder.name, builder.description))