ICEWS14.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.
# TODO: Address all TODOs and remove all explanatory comments
"""
TL;DR: The datasets for temporal knowledge graph reasoning task.

[[Github]](https://github.com/LinXueyuanStdio/TFLEX)
[[OpenReview]](https://openreview.net/forum?id=oaGdsgB18L)
[[arXiv]](https://arxiv.org/abs/2205.14307)

- Built over ICEWS and GDELT, which are widly used benchmarks in TKGC.
- First introduced in paper "TFLEX: Temporal Feature-Logic Embedding Framework for Complex Reasoning over Temporal Knowledge Graph"
- Please refer to the original paper for more details.


"""
from dataclasses import dataclass
from typing import List, Dict, Set, Optional, TypedDict
import json
import os

import datasets


_CITATION = """\
@inproceedings{
  xueyuan2023tflex,
  title={TFLEX: Temporal Feature-Logic Embedding Framework for Complex Reasoning over Temporal Knowledge Graph},
  author={Lin Xueyuan and Haihong E and Chengjin Xu and Gengxian Zhou and Haoran Luo and Tianyi Hu and Fenglong Su and Ningyuan Li and Mingzhi Sun},
  booktitle={Thirty-seventh Conference on Neural Information Processing Systems},
  year={2023},
  url={https://openreview.net/forum?id=oaGdsgB18L}
}\
"""

# TODO: Add description of the dataset here
_DESCRIPTION = """\
TL;DR: The datasets for temporal knowledge graph reasoning task.

[[Github]](https://github.com/LinXueyuanStdio/TFLEX)
[[OpenReview]](https://openreview.net/forum?id=oaGdsgB18L)
[[arXiv]](https://arxiv.org/abs/2205.14307)

- Built over ICEWS and GDELT, which are widly used benchmarks in TKGC.
- First introduced in paper "TFLEX: Temporal Feature-Logic Embedding Framework for Complex Reasoning over Temporal Knowledge Graph"
- Please refer to the original paper for more details.
"""

_HOMEPAGE = "https://github.com/LinXueyuanStdio/TFLEX"

_LICENSE = "[Apache License 2.0](https://github.com/LinXueyuanStdio/TFLEX/blob/main/LICENSE)"

query_name_to_args: Dict[str, List[str]] = {
    # 1. 1-hop Pe and Pt, manually
    "Pe": ['e1', 'r1', 't1'],
    "Pt": ['e1', 'r1', 'e2'],
    # 2. entity multi-hop
    "Pe2": ['e1', 'r1', 't1', 'r2', 't2'],
    "Pe3": ['e1', 'r1', 't1', 'r2', 't2', 'r3', 't3'],
    # 3. time multi-hop
    "aPt": ['s', 'r', 'o'],
    "bPt": ['s', 'r', 'o'],
    "Pt_sPe": ['e1', 'r1', 't1', 'r2', 'e2'],
    "Pt_oPe": ['e1', 'r1', 'e2', 'r2', 't1'],
    "Pe_Pt": ['e1', 'r1', 'e2', 'r2', 'e3'],
    "Pe_aPt": ['e1', 'r1', 'e2', 'r2', 'e3'],
    "Pe_bPt": ['e1', 'r1', 'e2', 'r2', 'e3'],
    "Pe_nPt": ['e1', 'r1', 'e2', 'r2', 'e3'],
    "Pt_sPe_Pt": ['s1', 'r1', 's2', 'r2', 'o1', 'r3', 'o2'],
    "Pt_oPe_Pt": ['s1', 'r1', 's2', 'r2', 's3', 'r3', 'o1'],
    # 4. entity and & time and
    "e2i": ['e1', 'r1', 't1', 'e2', 'r2', 't2'],
    "e3i": ['e1', 'r1', 't1', 'e2', 'r2', 't2', 'e3', 'r3', 't3'],
    "t2i": ['e1', 'r1', 'e2', 'e3', 'r2', 'e4'],
    "t3i": ['e1', 'r1', 'e2', 'e3', 'r2', 'e4', 'e5', 'r3', 'e6'],
    # 5. complex time and
    "e2i_Pe": ['e1', 'r1', 't1', 'r2', 't2', 'e2', 'r3', 't3'],
    "Pe_e2i": ['e1', 'r1', 't1', 'e2', 'r2', 't2', 'r3', 't3'],
    "Pt_se2i": ['e1', 'r1', 't1', 'e2', 'r2', 't2', 'r3', 'e3'],
    "Pt_oe2i": ['e1', 'r1', 'e2', 'r2', 't1', 'e3', 'r3', 't2'],
    "t2i_Pe": ['e1', 'r1', 't1', 'r2', 'e2', 'e3', 'r3', 'e4'],
    "Pe_t2i": ['e1', 'r1', 'e2', 'r2', 'e3', 'e4', 'r3', 'e5'],
    "Pe_at2i": ['e1', 'r1', 'e2', 'r2', 'e3', 'e4', 'r3', 'e5'],
    "Pe_bt2i": ['e1', 'r1', 'e2', 'r2', 'e3', 'e4', 'r3', 'e5'],
    "Pe_nt2i": ['e1', 'r1', 'e2', 'r2', 'e3', 'e4', 'r3', 'e5'],
    "between": ['e1', 'r1', 'e2', 'e3', 'r2', 'e4'],
    # 5. entity not
    "e2i_N": ['e1', 'r1', 't1', 'e2', 'r2', 't2'],
    "e3i_N": ['e1', 'r1', 't1', 'e2', 'r2', 't2', 'e3', 'r3', 't3'],
    "Pe_e2i_Pe_NPe": ['e1', 'r1', 't1', 'e2', 'r2', 't2', 'r3', 't3'],
    "e2i_NPe": ['e1', 'r1', 't1', 'r2', 't2', 'e2', 'r3', 't3'],
    "e2i_PeN": ['e1', 'r1', 't1', 'r2', 't2', 'e2', 'r3', 't3'],
    # 6. time not
    "t2i_N": ['e1', 'r1', 'e2', 'e3', 'r2', 'e4'],
    "t3i_N": ['e1', 'r1', 'e2', 'e3', 'r2', 'e4', 'e5', 'r3', 'e6'],
    "Pe_t2i_PtPe_NPt": ['e1', 'r1', 'e2', 'r2', 't2', 'r3', 'e3', 'e4', 'r4', 'e5'],
    "t2i_NPt": ['e1', 'r1', 't1', 'r2', 'e2', 'e3', 'r3', 'e4'],
    "t2i_PtN": ['e1', 'r1', 't1', 'r2', 'e2', 'e3', 'r3', 'e4'],
    # 7. entity union & time union
    "e2u": ['e1', 'r1', 't1', 'e2', 'r2', 't2'],
    "Pe_e2u": ['e1', 'r1', 't1', 'e2', 'r2', 't2', 'r3', 't3'],
    "t2u": ['e1', 'r1', 'e2', 'e3', 'r2', 'e4'],
    "Pe_t2u": ['e1', 'r1', 'e2', 'r2', 'e3', 'e4', 'r3', 'e5'],
}
query_structures: Dict[str, str] = {
    # 1. 1-hop Pe and Pt, manually
    "Pe": "def Pe(e1, r1, t1): return Pe(e1, r1, t1)",  # 1p
    "Pt": "def Pt(e1, r1, e2): return Pt(e1, r1, e2)",  # 1p, temporal
    # 2. entity multi-hop
    "Pe2": "def Pe2(e1, r1, t1, r2, t2): return Pe(Pe(e1, r1, t1), r2, t2)",  # 2p
    "Pe3": "def Pe3(e1, r1, t1, r2, t2, r3, t3): return Pe(Pe(Pe(e1, r1, t1), r2, t2), r3, t3)",  # 3p
    # 3. time multi-hop
    "aPt": "def aPt(s, r, o): return after(Pt(s, r, o))",  # a for after
    "bPt": "def bPt(s, r, o): return before(Pt(s, r, o))",  # b for before
    "Pt_lPe": "def Pt_lPe(e1, r1, t1, r2, e2): return Pt(Pe(e1, r1, t1), r2, e2)",  # l for left (as head entity)
    "Pt_rPe": "def Pt_rPe(e1, r1, e2, r2, t1): return Pt(e1, r1, Pe(e2, r2, t1))",  # r for right (as tail entity)
    "Pt_sPe": "def Pt_sPe(e1, r1, t1, r2, e2): return Pt(Pe(e1, r1, t1), r2, e2)",  # l for left (as head entity)
    "Pt_oPe": "def Pt_oPe(e1, r1, e2, r2, t1): return Pt(e1, r1, Pe(e2, r2, t1))",  # r for right (as tail entity)
    "Pe_Pt": "def Pe_Pt(e1, r1, e2, r2, e3): return Pe(e1, r1, Pt(e2, r2, e3))",  # at
    "Pe_aPt": "def Pe_aPt(e1, r1, e2, r2, e3): return Pe(e1, r1, after(Pt(e2, r2, e3)))",  # a for after
    "Pe_bPt": "def Pe_bPt(e1, r1, e2, r2, e3): return Pe(e1, r1, before(Pt(e2, r2, e3)))",  # b for before
    "Pe_nPt": "def Pe_nPt(e1, r1, e2, r2, e3): return Pe(e1, r1, next(Pt(e2, r2, e3)))",  # n for next
    "Pt_sPe_Pt": "def Pt_sPe_Pt(s1, r1, s2, r2, o1, r3, o2): return Pt(Pe(s1, r1, Pt(s2, r2, o1)), r3, o2)",
    "Pt_oPe_Pt": "def Pt_oPe_Pt(s1, r1, s2, r2, s3, r3, o1): return Pt(s1, r1, Pe(s2, r2, Pt(s3, r3, o1)))",
    # 4. entity and & time and
    "e2i": "def e2i(e1, r1, t1, e2, r2, t2): return And(Pe(e1, r1, t1), Pe(e2, r2, t2))",  # 2i
    "e3i": "def e3i(e1, r1, t1, e2, r2, t2, e3, r3, t3): return And3(Pe(e1, r1, t1), Pe(e2, r2, t2), Pe(e3, r3, t3))",  # 3i
    "t2i": "def t2i(e1, r1, e2, e3, r2, e4): return TimeAnd(Pt(e1, r1, e2), Pt(e3, r2, e4))",  # t-2i
    "t3i": "def t3i(e1, r1, e2, e3, r2, e4, e5, r3, e6): return TimeAnd3(Pt(e1, r1, e2), Pt(e3, r2, e4), Pt(e5, r3, e6))",  # t-3i
    # 5. complex time and
    "e2i_Pe": "def e2i_Pe(e1, r1, t1, r2, t2, e2, r3, t3): return And(Pe(Pe(e1, r1, t1), r2, t2), Pe(e2, r3, t3))",  # pi
    "Pe_e2i": "def Pe_e2i(e1, r1, t1, e2, r2, t2, r3, t3): return Pe(e2i(e1, r1, t1, e2, r2, t2), r3, t3)",  # ip
    "Pt_le2i": "def Pt_le2i(e1, r1, t1, e2, r2, t2, r3, e3): return Pt(e2i(e1, r1, t1, e2, r2, t2), r3, e3)",  # mix ip
    "Pt_re2i": "def Pt_re2i(e1, r1, e2, r2, t1, e3, r3, t2): return Pt(e1, r1, e2i(e2, r2, t1, e3, r3, t2))",  # mix ip
    "Pt_se2i": "def Pt_se2i(e1, r1, t1, e2, r2, t2, r3, e3): return Pt(e2i(e1, r1, t1, e2, r2, t2), r3, e3)",  # mix ip
    "Pt_oe2i": "def Pt_oe2i(e1, r1, e2, r2, t1, e3, r3, t2): return Pt(e1, r1, e2i(e2, r2, t1, e3, r3, t2))",  # mix ip
    "t2i_Pe": "def t2i_Pe(e1, r1, t1, r2, e2, e3, r3, e4): return TimeAnd(Pt(Pe(e1, r1, t1), r2, e2), Pt(e3, r3, e4))",  # t-pi
    "Pe_t2i": "def Pe_t2i(e1, r1, e2, r2, e3, e4, r3, e5): return Pe(e1, r1, t2i(e2, r2, e3, e4, r3, e5))",  # t-ip
    "Pe_at2i": "def Pe_at2i(e1, r1, e2, r2, e3, e4, r3, e5): return Pe(e1, r1, after(t2i(e2, r2, e3, e4, r3, e5)))",
    "Pe_bt2i": "def Pe_bt2i(e1, r1, e2, r2, e3, e4, r3, e5): return Pe(e1, r1, before(t2i(e2, r2, e3, e4, r3, e5)))",
    "Pe_nt2i": "def Pe_nt2i(e1, r1, e2, r2, e3, e4, r3, e5): return Pe(e1, r1, next(t2i(e2, r2, e3, e4, r3, e5)))",
    "between": "def between(e1, r1, e2, e3, r2, e4): return TimeAnd(after(Pt(e1, r1, e2)), before(Pt(e3, r2, e4)))",  # between(t1, t2) == after t1 and before t2
    # 5. entity not
    "e2i_N": "def e2i_N(e1, r1, t1, e2, r2, t2): return And(Pe(e1, r1, t1), Not(Pe(e2, r2, t2)))",  # 2in
    "e3i_N": "def e3i_N(e1, r1, t1, e2, r2, t2, e3, r3, t3): return And3(Pe(e1, r1, t1), Pe(e2, r2, t2), Not(Pe(e3, r3, t3)))",  # 3in
    "Pe_e2i_Pe_NPe": "def Pe_e2i_Pe_NPe(e1, r1, t1, e2, r2, t2, r3, t3): return Pe(And(Pe(e1, r1, t1), Not(Pe(e2, r2, t2))), r3, t3)",  # inp
    "e2i_PeN": "def e2i_PeN(e1, r1, t1, r2, t2, e2, r3, t3): return And(Pe(Pe(e1, r1, t1), r2, t2), Not(Pe(e2, r3, t3)))",  # pin
    "e2i_NPe": "def e2i_NPe(e1, r1, t1, r2, t2, e2, r3, t3): return And(Not(Pe(Pe(e1, r1, t1), r2, t2)), Pe(e2, r3, t3))",  # pni = e2i_N(Pe(e1, r1, t1), r2, t2, e2, r3, t3)
    # 6. time not
    "t2i_N": "def t2i_N(e1, r1, e2, e3, r2, e4): return TimeAnd(Pt(e1, r1, e2), TimeNot(Pt(e3, r2, e4)))",  # t-2in
    "t3i_N": "def t3i_N(e1, r1, e2, e3, r2, e4, e5, r3, e6): return TimeAnd3(Pt(e1, r1, e2), Pt(e3, r2, e4), TimeNot(Pt(e5, r3, e6)))",  # t-3in
    "Pe_t2i_PtPe_NPt": "def Pe_t2i_PtPe_NPt(e1, r1, e2, r2, t2, r3, e3, e4, r4, e5): return Pe(e1, r1, TimeAnd(Pt(Pe(e2, r2, t2), r3, e3), TimeNot(Pt(e4, r4, e5))))",  # t-inp
    "t2i_PtN": "def t2i_PtN(e1, r1, t1, r2, e2, e3, r3, e4): return TimeAnd(Pt(Pe(e1, r1, t1), r2, e2), TimeNot(Pt(e3, r3, e4)))",  # t-pin
    "t2i_NPt": "def t2i_NPt(e1, r1, t1, r2, e2, e3, r3, e4): return TimeAnd(TimeNot(Pt(Pe(e1, r1, t1), r2, e2)), Pt(e3, r3, e4))",  # t-pni
    # 7. entity union & time union
    "e2u": "def e2u(e1, r1, t1, e2, r2, t2): return Or(Pe(e1, r1, t1), Pe(e2, r2, t2))",  # 2u
    "Pe_e2u": "def Pe_e2u(e1, r1, t1, e2, r2, t2, r3, t3): return Pe(Or(Pe(e1, r1, t1), Pe(e2, r2, t2)), r3, t3)",  # up
    "t2u": "def t2u(e1, r1, e2, e3, r2, e4): return TimeOr(Pt(e1, r1, e2), Pt(e3, r2, e4))",  # t-2u
    "Pe_t2u": "def Pe_t2u(e1, r1, e2, r2, e3, e4, r3, e5): return Pe(e1, r1, TimeOr(Pt(e2, r2, e3), Pt(e4, r3, e5)))",  # t-up
    # 8. union-DM
    "e2u_DM": "def e2u_DM(e1, r1, t1, e2, r2, t2): return Not(And(Not(Pe(e1, r1, t1)), Not(Pe(e2, r2, t2))))",  # 2u-DM
    "Pe_e2u_DM": "def Pe_e2u_DM(e1, r1, t1, e2, r2, t2, r3, t3): return Pe(Not(And(Not(Pe(e1, r1, t1)), Not(Pe(e2, r2, t2)))), r3, t3)",  # up-DM
    "t2u_DM": "def t2u_DM(e1, r1, e2, e3, r2, e4): return TimeNot(TimeAnd(TimeNot(Pt(e1, r1, e2)), TimeNot(Pt(e3, r2, e4))))",  # t-2u-DM
    "Pe_t2u_DM": "def Pe_t2u_DM(e1, r1, e2, r2, e3, e4, r3, e5): return Pe(e1, r1, TimeNot(TimeAnd(TimeNot(Pt(e2, r2, e3)), TimeNot(Pt(e4, r3, e5)))))",  # t-up-DM
    # 9. union-DNF
    "e2u_DNF": "def e2u_DNF(e1, r1, t1, e2, r2, t2): return Pe(e1, r1, t1), Pe(e2, r2, t2)",  # 2u_DNF
    "Pe_e2u_DNF": "def Pe_e2u_DNF(e1, r1, t1, e2, r2, t2, r3, t3): return Pe(Pe(e1, r1, t1), r3, t3), Pe(Pe(e2, r2, t2), r3, t3)",  # up_DNF
    "t2u_DNF": "def t2u_DNF(e1, r1, e2, e3, r2, e4): return Pt(e1, r1, e2), Pt(e3, r2, e4)",  # t-2u_DNF
    "Pe_t2u_DNF": "def Pe_t2u_DNF(e1, r1, e2, r2, e3, e4, r3, e5): return Pe(e1, r1, Pt(e2, r2, e3)), Pe(e1, r1, Pt(e4, r3, e5))",  # t-up_DNF
}
union_query_structures: List[str] = [
    "e2u", "Pe_e2u",  # 2u, up
    "t2u", "Pe_t2u",  # t-2u, t-up
]
train_query_structures: List[str] = [
    # entity
    "Pe", "Pe2", "Pe3", "e2i", "e3i",  # 1p, 2p, 3p, 2i, 3i
    "e2i_NPe", "e2i_PeN", "Pe_e2i_Pe_NPe", "e2i_N", "e3i_N",  # npi, pni, inp, 2in, 3in
    # time
    "Pt", "Pt_lPe", "Pt_rPe", "Pe_Pt", "Pe_aPt", "Pe_bPt", "Pe_nPt",  # t-1p, t-2p
    "t2i", "t3i", "Pt_le2i", "Pt_re2i", "Pe_t2i", "Pe_at2i", "Pe_bt2i", "Pe_nt2i", "between",  # t-2i, t-3i
    "t2i_NPt", "t2i_PtN", "Pe_t2i_PtPe_NPt", "t2i_N", "t3i_N",  # t-npi, t-pni, t-inp, t-2in, t-3in
]
test_query_structures: List[str] = train_query_structures + [
    # entity
    "e2i_Pe", "Pe_e2i",  # pi, ip
    "e2u", "Pe_e2u",  # 2u, up
    # time
    "t2i_Pe", "Pe_t2i",  # t-pi, t-ip
    "t2u", "Pe_t2u",  # t-2u, t-up
    # union-DM
    "e2u_DM", "Pe_e2u_DM",  # 2u-DM, up-DM
    "t2u_DM", "Pe_t2u_DM",  # t-2u-DM, t-up-DM
]


# TODO: Add link to the official dataset URLs here
# The HuggingFace Datasets library doesn't host the datasets but only points to the original files.
# This can be an arbitrary nested dict/list of URLs (see below in `_split_generators` method)
_HOST = "https://huggingface.co/datasets"
_AUTHOR = "linxy"
_DATASET = "ICEWS14"
_URLS = {
    name: f"{_HOST}/{_AUTHOR}/{_DATASET}/resolve/main/zips/{name}.zip?download=true"
    for name in ["all"] + list(query_name_to_args.keys())
}


class QueryData(TypedDict):
    """
    saved in training split: query_name, query, answer
    saved in valid or test split: query_name, query, answer, easy_answer
    iterating training dataloader: query_name, query, answer, args, definition
    iterating valid or test dataloader: query_name, query, answer, easy_answer, args, definition
    """
    query_name: str
    query: List[int]
    answer: Set[int]
    easy_answer: Optional[Set[int]] = None  # may be empty, indicating that no easy answer exists in training graph.
    args: Optional[List[str]] = None
    definition: Optional[str] = None

@dataclass
class TKGRBuilderConfig(datasets.BuilderConfig):
    """BuilderConfig for TKGR (Temporal Knowledge Graph Reasoning)."""
    query_structure_name: str = "default"

class ICEWS14Dataset(datasets.GeneratorBasedBuilder):
    """TODO: Short description of my dataset."""

    VERSION = datasets.Version("1.0.0")

    # This is an example of a dataset with multiple configurations.
    # If you don't want/need to define several sub-sets in your dataset,
    # just remove the BUILDER_CONFIG_CLASS and the BUILDER_CONFIGS attributes.

    # If you need to make complex sub-parts in the datasets with configurable options
    # You can create your own builder configuration class to store attribute, inheriting from datasets.BuilderConfig
    # BUILDER_CONFIG_CLASS = MyBuilderConfig

    # You will be able to load one or the other configurations in the following list with
    # data = datasets.load_dataset('my_dataset', 'first_domain')
    # data = datasets.load_dataset('my_dataset', 'second_domain')
    STANDARD_BUILDER_CONFIGS = [
        datasets.BuilderConfig(
            name=query_name,
            version=datasets.Version("1.0.0"), 
            description=query_structures[query_name],
        )
        for query_name in list(query_name_to_args.keys())
    ]
    BUILDER_CONFIGS = [
        datasets.BuilderConfig(
            name="all",
            version=VERSION,
            description=f"All types of queries. Train: {train_query_structures}, Valid | Test: {test_query_structures}",
        )
    ] + STANDARD_BUILDER_CONFIGS

    DEFAULT_CONFIG_NAME = "all"  # It's not mandatory to have a default configuration. Just use one if it make sense.

    def _info(self):
        if self.config.name == "all":  # This is the name of the configuration selected in BUILDER_CONFIGS above
            features = datasets.Features(
                {
                    "query_name": datasets.Value("string"),
                    "definition": datasets.Value("string"),
                    "query": datasets.Sequence(feature=datasets.Value("int32")),
                    "answer": datasets.Sequence(feature=datasets.Value("int32")),
                    "easy_answer": datasets.Sequence(feature=datasets.Value("int32")),
                    "args": datasets.Sequence(feature=datasets.Value("string")),
                }
            )
        else:
            features = datasets.Features(
                {
                    "query_name": datasets.Value("string"),
                    "definition": datasets.Value("string"),
                    "query": datasets.Sequence(feature=datasets.Value("int32")),
                    "answer": datasets.Sequence(feature=datasets.Value("int32")),
                    "easy_answer": datasets.Sequence(feature=datasets.Value("int32")),
                    "args": datasets.Sequence(feature=datasets.Value("string")),
                }
            )
        return datasets.DatasetInfo(
            description=_DESCRIPTION,
            features=features,
            homepage=_HOMEPAGE,
            license=_LICENSE,
            citation=_CITATION,
        )

    def _split_generators(self, dl_manager):
        # dl_manager is a datasets.download.DownloadManager that can be used to download and extract URLS
        # It can accept any type or nested list/dict and will give back the same structure with the url replaced with path to local files.
        # By default the archives will be extracted and a path to a cached folder where they are extracted is returned instead of the archive
        urls = _URLS[self.config.name]
        data_dir = dl_manager.download_and_extract(urls)
        return [
            datasets.SplitGenerator(
                name=datasets.Split.TRAIN,
                # These kwargs will be passed to _generate_examples
                gen_kwargs={
                    "filepath": os.path.join(data_dir, "train.jsonl"),
                    "split": "train",
                },
            ),
            datasets.SplitGenerator(
                name=datasets.Split.VALIDATION,
                # These kwargs will be passed to _generate_examples
                gen_kwargs={
                    "filepath": os.path.join(data_dir, "valid.jsonl"),
                    "split": "valid",
                },
            ),
            datasets.SplitGenerator(
                name=datasets.Split.TEST,
                # These kwargs will be passed to _generate_examples
                gen_kwargs={
                    "filepath": os.path.join(data_dir, "test.jsonl"),
                    "split": "test"
                },
            ),
        ]

    def _generate_examples(self, filepath, split):
        # method parameters are unpacked from `gen_kwargs` as given in `_split_generators`
        # This method yields (key, example) tuples from the dataset.
        # The `key` is for legacy reasons (tfds) and is not important in itself, but must be unique for each example.
        if not os.path.exists(filepath):
            return
        with open(filepath, encoding="utf-8") as f:
            for key, row in enumerate(f):
                data = json.loads(row)
                query_name = data["query_name"]
                if self.config.name == "all":
                    yield key, {
                        "query_name": query_name,
                        "query": data["query"],
                        "answer": data["answer"],
                        "easy_answer": data["easy_answer"] if "easy_answer" in data else None,
                        "args": query_name_to_args[query_name],
                        "definition": query_structures[query_name],
                    }
                else:
                    yield key, {
                        "query_name": query_name,
                        "query": data["query"],
                        "answer": data["answer"],
                        "easy_answer": data["easy_answer"] if "easy_answer" in data else None,
                        "args": query_name_to_args[query_name],
                        "definition": query_structures[query_name],
                    }