dialogstudio.py

"""
 Copyright (c) 2023, salesforce.com, inc.
 All rights reserved.
 SPDX-License-Identifier: Apache License 2.0
 For full license text, see the LICENSE file in the repo root or https://www.apache.org/licenses/LICENSE-2.0
"""


#!/usr/bin/env python3
import os
import json
import datasets
from datasets import (GeneratorBasedBuilder,
                        BuilderConfig,
                        SplitGenerator, 
                        DatasetInfo, 
                        Features,
                        Sequence,
                        Value,
                        Version)

logger = datasets.logging.get_logger(__name__)
datasets.logging.disable_progress_bar()

_VERSION = Version("1.0.0")
_CITATION = """
    @misc{zhang2023dialogstudio,
    title={DialogStudio: Towards Richest and Most Diverse Unified Dataset Collection for Conversational AI}, 
    author={Jianguo Zhang and Kun Qian and Zhiwei Liu and Shelby Heinecke and Rui Meng and Ye Liu and Zhou Yu and and Huan Wang and Silvio Savarese and Caiming Xiong},
    year={2023},
    eprint={2307.10172},
    archivePrefix={arXiv},
    primaryClass={cs.CL}
"""

DATASETS = {
    # not pass: "CANARD"
    "open_domain": [
                        "chitchat-dataset", "ConvAI2", "AntiScam", "Empathetic", "HH-RLHF", 
                        "PLACES3.5", "Prosocial", "SODA"
                    ],
    "knowledge_grounded": [
                        "CompWebQ", "CoQA", "CoSQL", "DART", "FeTaQA", 
                        "GrailQA", "HybridQA", "MTOP", "MultiModalQA", "SParC", 
                        "Spider", "SQA", "ToTTo", "WebQSP", "WikiSQL", 
                        "WikiTQ", "wizard_of_internet", "wizard_of_wikipedia"
                    ],
    "dialogue_summarization": [
                        "AMI", "CRD3", "DialogSum", "ECTSum", "ICSI", 
                        "MediaSum", "QMSum", "SAMSum", "TweetSumm", "ConvoSumm", 
                        "SummScreen_ForeverDreaming", "SummScreen_TVMegaSite"
                    ],
    "natural_language_understanding": [
                        "ATIS", "ATIS-NER", "BANKING77", "BANKING77-OOS", "CLINC-Single-Domain-OOS-banking", 
                        "CLINC-Single-Domain-OOS-credit_cards", "CLINC150", "DSTC8-SGD", "HWU64", "MIT-Movie",
                        "MIT-Restaurant", "RESTAURANTS8K", "SNIPS", "SNIPS-NER", "TOP", "TOP-NER"
                    ],    
    "task_oriented": [
                        "ABCD", "AirDialogue", "BiTOD", "CaSiNo", "CraigslistBargains", 
                        "Disambiguation", "DSTC2-Clean", "FRAMES", "GECOR", "HDSA-Dialog", 
                        "KETOD", "KVRET", "MetaLWOZ", "MS-DC", "MuDoCo", 
                        "MulDoGO", "MultiWOZ_2.1", "MULTIWOZ2_2", "SGD", "SimJointGEN", 
                        "SimJointMovie", "SimJointRestaurant", "STAR", "Taskmaster1", "Taskmaster2", 
                        "Taskmaster3", "WOZ2_0"
                    ],
    "conversational_recommendation": [
                        "Redial", "DuRecDial-2.0", "OpenDialKG", "SalesBot",
                    ]
}

_URL = "https://huggingface.co/datasets/Salesforce/dialogstudio/tree/main/"

class DialogStudioConfig(BuilderConfig):
    """BuilderConfig for DialogStudio."""

    def __init__(self, extra_features, category, data_name, citation, url, **kwargs):
        """BuilderConfig for DialogStudio.
        Args:
          extra_features: `list[string]`, list of the features that will appear in the
            feature dict. Should not include "label".
          data_url: `string`, url to download the zip file from.
          citation: `string`, citation for the data set.
          url: `string`, url for information about the data set.
          label_classes: `list[string]`, the list of classes for the label if the
            label is present as a string. Non-string labels will be cast to either
            'False' or 'True'.
          **kwargs: keyword arguments forwarded to super.
        """
        super(DialogStudioConfig, self).__init__(version=_VERSION, **kwargs)
        self.extra_features = extra_features
        self.category = category
        self.data_name = data_name
        self.compressed_file = f"{data_name}.zip"
        self.citation = citation
        self.url = url


class DialogStudio(GeneratorBasedBuilder):
    """DialogStudio"""
    BUILDER_CONFIGS = []
    for category, dataset_list in DATASETS.items():
        if category in ["task_oriented", "conversational_recommendation"]:
            extra_features = {
                    "dialog":[
                            "external knowledge non-flat",
                            "external knowledge", 
                            "dst knowledge",
                            "intent knowledge",
                        ], 
                    "turn":[
                            "dst",
                            "dst accumulated",
                            "intent",
                            "external knowledge", 
                            "external knowledge non-flat"
                        ]
                    }
        elif category in ["natural_language_understanding"]:
            extra_features = {"dialog":[], "turn":["external knowledge", "external knowledge non-flat"]}
        else:
            extra_features = {"dialog":[], "turn":[]}

        for data_name in dataset_list:
            BUILDER_CONFIGS.append(
                DialogStudioConfig(
                    name=data_name,
                    data_name=data_name,
                    description="",
                    extra_features=extra_features,
                    category=category,
                    citation=_CITATION,
                    url="https://github.com/salesforce/DialogStudio/tree/main",
            ))
    
    DEFAULT_CONFIG_NAME = "WOZ2_0"

    def _info(self):
        features = {
                    "original dialog id": Value("string"),
                    "new dialog id": Value("string"),
                    "dialog index": Value("int32"),
                    "original dialog info": Value("string"),
                    "log": [
                        {
                            "turn id": Value("int32"),
                            "user utterance": Value("string"),
                            "system response": Value("string"),
                            "dialog history": Value("string"),
                            "original user side information": Value("string"),
                            "original system side information": Value("string"),
                        }
                    ],
                    "prompt": [Value("string")]
                }
        if self.config.extra_features["dialog"]:
            for name in self.config.extra_features["dialog"]:
                features[name] = Value("string")
        if self.config.extra_features["turn"]:
            log_config = {
                            "turn id": Value("int32"),
                            "user utterance": Value("string"),
                            "system response": Value("string"),
                            "dialog history": Value("string"),
                            "original user side information": Value("string"),
                            "original system side information": Value("string"),
                        }
            for name in self.config.extra_features["turn"]:
                log_config[name] = Value("string")
            features["log"] = [log_config]
        
        return DatasetInfo(
            description="",
            features=Features(features),
            homepage=self.config.url,
            citation=_CITATION,
        )

    
    def _split_generators(self, dl_manager):
        """
        This script assume the datset is not stored in zip file
        Instead, data is stored in format:
        .
         |-task-oriented
                |-WOZ2.0
                    |-train
                        |-dialogues_1.json
                        |-dialogues_2.json
                        |-...
        this script would download the json file one-by-one
        """

        splits = []
        file_path_list = {"train":[], "val":[], "test":[]} 
        print("❤️Attention❤️: Dataset download may take some time. We appreciate your patience!")
        for mode in ["train", "val", "test"]:
            for file_idx in range(1, 1000000):
                file_to_download = os.path.join(self.config.category, self.config.data_name, mode, f"dialogues_{file_idx}.json")
                try:
                    dl_path = dl_manager.download(file_to_download)
                except:
                    break
                file_path_list[mode].append(dl_path)

        if file_path_list["train"]:
            if any(split.name == datasets.Split.TRAIN for split in splits):
                raise ValueError("Split 'train' already added.")
            splits.append(
                SplitGenerator(
                    name=datasets.Split.TRAIN,
                    gen_kwargs={
                        "file_path_list": file_path_list["train"],
                        "split": datasets.Split.TRAIN,
                    },
                )
            )
        if file_path_list["val"]:
            if any(split.name == datasets.Split.VALIDATION for split in splits):
                raise ValueError("Split 'validation' already added.")
            splits.append(
                SplitGenerator(
                    name=datasets.Split.VALIDATION,
                    gen_kwargs={
                        "file_path_list": file_path_list["val"],
                        "split": datasets.Split.VALIDATION,
                    },
                )
            )
        if file_path_list["test"]:
            if any(split.name == datasets.Split.TEST for split in splits):
                raise ValueError("Split 'test' already added.")
            splits.append(
                SplitGenerator(
                    name=datasets.Split.TEST,
                    gen_kwargs={
                        "file_path_list": file_path_list["test"],
                        "split": datasets.Split.TEST,
                    },
                )
            )
        return splits


    def _load_json(self, file_path):
        with open(file_path, encoding="utf-8") as f:
            data = json.loads(f.read())
        return data


    def _generate_examples(self, file_path_list, split):
        """This function returns the examples in the raw (text) form."""
        data = {}
        for filepath in file_path_list:
            data.update(self._load_json(filepath))
        
        logger.info(f"generating {len(data)} examples from = {split}")

        for dial_id, dial in data.items():
            if type(dial["log"]) == dict:
                dial["log"] = [dial["log"]]
            example = {
                "original dialog id": dial["original dialog id"],
                "new dialog id": dial_id,
                "dialog index": dial["dialog index"],
                "original dialog info": json.dumps(dial["original dialog info"]),
                "log": [{
                        "turn id": turn["turn id"],
                        "user utterance": turn["user utterance"],
                        "system response": turn["system response"],
                        "dialog history": turn["dialog history"],
                        "original user side information": json.dumps(turn["original user side information"]),
                        "original system side information": json.dumps(turn["original system side information"]),
                    } for turn in dial["log"]],
                "prompt": dial["prompt"] if "prompt" in dial and dial["prompt"] else [""]
            }
            if self.config.extra_features["dialog"]:
                for name in self.config.extra_features["dialog"]:
                    example[name] = json.dumps(dial[name]) if name in dial else ""
            if self.config.extra_features["turn"]:
                for idx, turn in enumerate(example["log"]):
                    for name in self.config.extra_features["turn"]:
                        example["log"][idx][name] = json.dumps(dial["log"][idx][name]) if name in dial["log"][idx] else ""
            yield dial["dialog index"], example