nle_hf_dataset.py

# Copyright 2020 The HuggingFace Datasets Authors and the current dataset script contributor.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
#     http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.

""" Dungeons and Data: A Large-Scale NetHack Dataset. """

import h5py
import json
import os
import datasets


_CITATION = """\
@article{hambro2022dungeons,
  title={Dungeons and Data: A Large-Scale NetHack Dataset},
  author={Hambro, Eric and Raileanu, Roberta and Rothermel, Danielle and Mella, Vegard and Rockt{\"a}schel, Tim and K{\"u}ttler, Heinrich and Murray, Naila},
  journal={arXiv preprint arXiv:2211.00539},
  year={2022}
}
"""

_DESCRIPTION = """\
3 billion state-action-score transitions from 100,000 trajectories collected from the symbolic bot winner of the NetHack Challenge 2021.
"""

_HOMEPAGE = ""

_LICENSE = ""

_URLS = {
    "data": [f"data/{i}.hdf5" for i in range(1, 6)],
    "metadata": [f"metadata/{i}.json" for i in range(1, 6)],
}

class NleHfDataset(datasets.GeneratorBasedBuilder):
    """Dungeons and Data: A Large-Scale NetHack Dataset."""
    VERSION = datasets.Version("1.0.0")

    BUILDER_CONFIGS = [
        datasets.BuilderConfig(name="data", version=VERSION, description="Data for all episodes"),
        datasets.BuilderConfig(name="metadata", version=VERSION, description="Metadata for all episodes"),
    ]
    DEFAULT_CONFIG_NAME = "metadata"

    def _info(self):
        if self.config.name == "metadata":
            features = datasets.Features(
                {
                    "gameid": datasets.Value("int32"), 
                    "version": datasets.Value("string"), 
                    "points": datasets.Value("int32"), 
                    "deathdnum": datasets.Value("int32"), 
                    "deathlev": datasets.Value("int32"), 
                    "maxlvl": datasets.Value("int32"), 
                    "hp": datasets.Value("int32"), 
                    "maxhp": datasets.Value("int32"), 
                    "deaths": datasets.Value("int32"), 
                    "deathdate": datasets.Value("int32"), 
                    "birthdate": datasets.Value("int32"), 
                    "uid": datasets.Value("int32"), 
                    "role": datasets.Value("string"), 
                    "race": datasets.Value("string"), 
                    "gender": datasets.Value("string"), 
                    "align": datasets.Value("string"), 
                    "name": datasets.Value("string"), 
                    "death": datasets.Value("string"), 
                    "conduct": datasets.Value("string"), 
                    "turns": datasets.Value("int32"), 
                    "achieve": datasets.Value("string"), 
                    "realtime": datasets.Value("int64"), 
                    "starttime": datasets.Value("int64"), 
                    "endtime": datasets.Value("int64"), 
                    "gender0": datasets.Value("string"), 
                    "align0": datasets.Value("string"), 
                    "flags": datasets.Value("string")
                }
            )
        else:
            features = datasets.Features(
                {
                    "tty_chars": datasets.Array3D(shape=(None, 24, 80), dtype="uint8"),
                    "tty_colors": datasets.Array3D(shape=(None, 24, 80), dtype="int8"),
                    "tty_cursor": datasets.Array2D(shape=(None, 2), dtype="int16"),
                    "actions": datasets.Sequence(datasets.Value("int16")),
                    "rewards": datasets.Sequence(datasets.Value("int32")),
                    "dones": datasets.Sequence(datasets.Value("bool")),
                }
            )
        
        return datasets.DatasetInfo(
            description=_DESCRIPTION,
            features=features,
            homepage=_HOMEPAGE,
            license=_LICENSE,
            citation=_CITATION,
        )

    def _split_generators(self, dl_manager):
        if self.config.data_files is None:
            urls = _URLS[self.config.name]
        else:
            urls = self.config.data_files["train"]
        # print(self.config.data_files["train"])

        filepaths = [dl_manager.download(url) for url in urls]
        return [
            datasets.SplitGenerator(
                name=datasets.Split.TRAIN, gen_kwargs={"filepaths": filepaths})
        ]

    def _generate_examples(self, filepaths):
        for i, filepath in enumerate(filepaths):
            if self.config.name == "metadata":
                with open(filepath, "r") as f:
                    data = json.load(f)
                    yield i, data
            else:
                with h5py.File(filepath, "r") as f:
                    yield i, {
                        "tty_chars": f["tty_chars"][()],
                        "tty_colors": f["tty_colors"][()],
                        "tty_cursor": f["tty_cursor"][()],
                        "actions": f["actions"][()],
                        "rewards": f["rewards"][()],
                        "dones": f["dones"][()]
                    }