v1 / v1.py
# -*- coding: utf-8 -*-
"""CLUTRR_Dataset Loading Script.ipynb
Automatically generated by Colaboratory.
Original file is located at
https://colab.research.google.com/drive/1q9DdeHA5JbgTHkH6kfZe_KWHQOwHZA97
"""
# coding=utf-8
# Copyright 2019 The CLUTRR Datasets Authors and the HuggingFace Datasets Authors.
#
# CLUTRR is CC-BY-NC 4.0 (Attr Non-Commercial Inter.) licensed, as found in the LICENSE file.
#
# 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.
# Lint as: python3
"""The CLUTRR (Compositional Language Understanding and Text-based Relational Reasoning) benchmark."""
import csv
import os
import textwrap
import numpy as np
import datasets
import json
_CLUTRR_CITATION = """\
@article{sinha2019clutrr,
Author = {Koustuv Sinha and Shagun Sodhani and Jin Dong and Joelle Pineau and William L. Hamilton},
Title = {CLUTRR: A Diagnostic Benchmark for Inductive Reasoning from Text},
Year = {2019},
journal = {Empirical Methods of Natural Language Processing (EMNLP)},
arxiv = {1908.06177}
}
"""
_CLUTRR_DESCRIPTION = """\
CLUTRR (Compositional Language Understanding and Text-based Relational Reasoning),
a diagnostic benchmark suite, is first introduced in (https://arxiv.org/abs/1908.06177)
to test the systematic generalization and inductive reasoning capabilities of NLU systems.
"""
_URL = "https://raw.githubusercontent.com/kliang5/CLUTRR_huggingface_dataset/main/"
_TASK = ["gen_train23_test2to10", "gen_train234_test2to10", "rob_train_clean_23_test_all_23", "rob_train_disc_23_test_all_23", "rob_train_irr_23_test_all_23","rob_train_sup_23_test_all_23"]
class v1(datasets.GeneratorBasedBuilder):
"""BuilderConfig for CLUTRR."""
BUILDER_CONFIGS = [
datasets.BuilderConfig(
name=task,
version=datasets.Version("1.0.0"),
description="",
)
for task in _TASK
]
def _info(self):
return datasets.DatasetInfo(
description=_CLUTRR_DESCRIPTION,
features=datasets.Features(
{
"id": datasets.Value("string"),
"story": datasets.Value("string"),
"query": datasets.Value("string"),
"target": datasets.Value("int32"),
"target_text": datasets.Value("string"),
"clean_story": datasets.Value("string"),
"proof_state": datasets.Value("string"),
"f_comb": datasets.Value("string"),
"task_name": datasets.Value("string"),
"story_edges": datasets.Value("string"),
"edge_types": datasets.Value("string"),
"query_edge": datasets.Value("string"),
"genders": datasets.Value("string"),
"task_split": datasets.Value("string"),
}
),
# No default supervised_keys (as we have to pass both premise
# and hypothesis as input).
supervised_keys=None,
homepage="https://www.cs.mcgill.ca/~ksinha4/clutrr/",
citation=_CLUTRR_CITATION,
)
def _split_generators(self, dl_manager):
"""Returns SplitGenerators."""
# dl_manager is a datasets.download.DownloadManager that can be used to
# download and extract URLs
task = str(self.config.name)
urls_to_download = {
"test": _URL + task + "/test.csv",
"train": _URL + task + "/train.csv",
"validation": _URL + task + "/validation.csv",
}
downloaded_files = dl_manager.download_and_extract(urls_to_download)
return [
datasets.SplitGenerator(
name=datasets.Split.TRAIN,
# These kwargs will be passed to _generate_examples
gen_kwargs={
"filepath": downloaded_files["train"],
"task": task,
},
),
datasets.SplitGenerator(
name=datasets.Split.VALIDATION,
# These kwargs will be passed to _generate_examples
gen_kwargs={
"filepath": downloaded_files["validation"],
"task": task,
},
),
datasets.SplitGenerator(
name=datasets.Split.TEST,
# These kwargs will be passed to _generate_examples
gen_kwargs={
"filepath": downloaded_files["test"],
"task": task,
},
),
]
def _generate_examples(self, filepath, task):
"""Yields examples."""
with open(filepath, encoding="utf-8") as f:
reader = csv.reader(f)
for id_, data in enumerate(reader):
if id_ == 0:
continue
# yield id_, data
# id_ += 1
yield id_, {
"id": data[1],
"story": data[2],
"query": data[3],
"target": data[4],
"target_text": data[5],
"clean_story": data[6],
"proof_state": data[7],
"f_comb": data[8],
"task_name": data[9],
"story_edges": data[10],
"edge_types": data[11],
"query_edge": data[12],
"genders": data[13],
"task_split": data[14],
}