SalesforceDialogStudio / code /preprocess_data_DialSum.py
mzozulia's picture
Upload 339 files
2ea1065
"""
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 sys, os, pdb
import json
import shutil, errno
from tqdm import tqdm
import pandas as pd
from utils.constant import *
class PreProcessData(object):
"""docstring for PreProcessData"""
def __init__(self):
super(PreProcessData, self).__init__()
self.data_dir = "/path/to/where/the/raw/dataset/is"
self.save_dir = "/path/to/store/the/processed/dataset/" # e.g. ./data/processed/Dialogue-Summarization
def _load_json(self, path=None):
if path is None or not os.path.exists(path):
raise IOError('File does not exist: %s' % path)
# return None
with open(path) as df:
data = json.loads(df.read())
return data
def _load_txt(self, path=None, split_tok="\n"):
if path is None or not os.path.exists(path):
raise IOError('File does not exist: %s' % path)
with open(path) as df:
data = df.read().strip().split(split_tok)
return data
def _load_csv(self, path=None, sep="\t"):
if path is None or not os.path.exists(path):
raise IOError('File does not exist: %s' % path)
with open(path) as df:
data = pd.read_csv(df, sep=sep)
return data
def _load_jsonl(self, path=None):
if path is None or not os.path.exists(path):
raise IOError('File does not exist: %s' % path)
data = []
with open(path) as df:
for line in df.readlines():
data.append(json.loads(line))
return data
def _load_dir_json(self, dir_path=None):
if dir_path is None or not os.path.exists(dir_path): return None
total_data = [] # assume data is a list of dialogs
for filename in sorted(os.listdir(dir_path)):
if filename in ["schema.json"]: continue
if not filename.endswith(".json"): continue
file_path = os.path.join(dir_path, filename)
data = self._load_json(path=file_path)
if type(data) == list:
total_data.extend(data)
else:
total_data.append(data)
return total_data
def _load_dir_txt(self, dir_path=None, file_type="txt"):
if dir_path is None or not os.path.exists(dir_path): return None
total_data = [] # assume data is a list of dialogs
for filename in sorted(os.listdir(dir_path)):
if not filename.endswith(file_type): continue
file_path = os.path.join(dir_path, filename)
data = self._load_txt(path=file_path)
if type(data) == list:
total_data.extend(data)
else:
total_data.append(data)
return total_data
def _load_dir_tsv(self, dir_path=None, sep="\t"):
if dir_path is None or not os.path.exists(dir_path): return None
total_data = None
for filename in sorted(os.listdir(dir_path)):
file_path = os.path.join(dir_path, filename)
data = self._load_csv(path=file_path, sep=sep)
total_data = pd.concat([total_data, data], ignore_index=True)
return total_data
def _save_json(self, data, path):
with open(path, "w") as tf:
json.dump(data, tf, indent=4)
def init_dial(self, dial_idx=0, ori_dial_id=""):
dial = {
ORI_DIAL_ID: ori_dial_id,
DIAL_IDX: dial_idx,
ORI_DIAL_INFO: {},
LOG: [],
PROMPT: [],
}
return dial
def init_turn(self, turn_id=0, dial_hist=[]):
turn = {
TURN_ID: turn_id,
USR_UTT: "",
SYS_UTT: "",
DIAL_HIST: " ".join(dial_hist),
ORI_USR_ANN: {},
ORI_SYS_ANN: {},
}
return turn
def save_dial(self, data, data_name="", file_idx=0, mode="train"):
save_name = f"dialogues_{file_idx}.json"
folder_path = os.path.join(self.save_dir, data_name, mode)
if not os.path.exists(folder_path): os.makedirs(folder_path)
path = os.path.join(folder_path, save_name)
self._save_json(data, path)
def copy_general(self, src, dst):
try:
shutil.copytree(src, dst, dirs_exist_ok=True)
except OSError as exc: # python >2.5
if exc.errno in (errno.ENOTDIR, errno.EINVAL):
shutil.copy(src, dst)
else: raise
def copy_related_files(self, data_name, exp_list=[], extra_dir=""):
source_dir = os.path.join(self.data_dir, data_name, extra_dir)
target_dir = os.path.join(self.save_dir, data_name)
for filename in os.listdir(source_dir):
if filename.startswith("."): continue # ignore hidden files
if filename.startswith("__"): continue # ignore hidden files
if filename in exp_list: continue
if filename.endswith(".py"): continue
source_path = os.path.join(source_dir, filename)
target_path = os.path.join(target_dir, filename)
self.copy_general(source_path, target_path)
def save_original_examples(self, examples, data_name):
"""
save 5 original data points just for reference and check
data would be a list of length 5, each entry is a dialog
in the form of dictionary
"""
path = os.path.join(self.save_dir, data_name, "original_examples.json")
self._save_json(examples, path)
print("original examples saved")
def save_converted_examples(self, data_name):
"""
extract the first 5 examples from the train set of the
already processed data, just for reference and check
"""
data = self._load_json(os.path.join(self.save_dir, data_name, "train/dialogues_1.json"))
examples = {key: data[key] for key in list(data.keys())[:5]}
self._save_json(examples, os.path.join(self.save_dir, data_name, "converted_examples.json"))
print("converted examples saved")
def _import_system_file(self, filename="", module_name=""):
import importlib, sys
spec = importlib.util.spec_from_file_location(module_name, filename)
module = importlib.util.module_from_spec(spec)
sys.modules[module_name] = module
spec.loader.exec_module(module)
return module
def tweetsum(self):
"""
real data store in kaggle, need to download and preprocess first
"""
data_name = "TweetSumm"
# prepare data
Modules = self._import_system_file(os.path.join(self.data_dir, data_name, "tweet_sum_processor.py"), "TweetSumProcessor")
processor = Modules.TweetSumProcessor(os.path.join(self.data_dir, data_name, "archive/twcs/twcs.csv"))
exp_list = ["tweet_sum_data_files", "archive", "tweet_sum_processor.py"]
for mode in ["train", "val", "test"]:
real_name = f"final_{mode}_tweetsum.jsonl" if mode != "val" else "final_valid_tweetsum.jsonl"
path = os.path.join(self.data_dir, data_name, "tweet_sum_data_files", real_name)
# split = self._load_jsonl(path)
new_data = {}
file_idx = 1
original_data_sample = []
with open(path) as f:
dialog_with_summaries = processor.get_dialog_with_summaries(f.readlines())
for dial_idx, dialog_with_summary in tqdm(enumerate(dialog_with_summaries)):
new_dial_id = f"{data_name}--{mode}--{dial_idx+1}"
json_format = dialog_with_summary.get_json()
dial = json.loads(json_format)
if mode == "train" and dial_idx < 5:
original_data_sample.append(dial)
new_dial = self.init_dial(dial_idx=dial_idx+1, ori_dial_id=dial["dialog"]["dialog_id"]) # idx starts from 1
new_dial[ORI_DIAL_INFO] = {
"summaries" : dial["summaries"]
}
turn_id, dial_hist = 1, []
new_turn = self.init_turn(turn_id=turn_id)
for idx, turn in enumerate(dial["dialog"]["turns"]):
utt = " ".join(turn["sentences"])
if turn["is_agent"]:
new_turn[SYS_UTT] += f" {utt}"
new_turn[SYS_UTT] = new_turn[SYS_UTT].strip()
if idx == len(dial["dialog"]["turns"]) - 1 or \
not dial["dialog"]["turns"][idx+1]["is_agent"]:
new_dial[LOG].append(new_turn)
turn_id += 1
if new_turn[USR_UTT]:
dial_hist.append("<USER> " + new_turn[USR_UTT])
dial_hist.append("<SYSTEM> " + new_turn[SYS_UTT])
new_turn = self.init_turn(turn_id=turn_id)
new_turn[DIAL_HIST] = " ".join(dial_hist)
else:
new_turn[USR_UTT] += f" {utt}"
new_turn[USR_UTT] = new_turn[USR_UTT].strip()
new_data[new_dial_id] = new_dial
if (dial_idx+1) % 1000 == 0 or dial_idx+1 == len(dialog_with_summaries):
self.save_dial(new_data, data_name=data_name, file_idx=file_idx, mode=mode)
new_data = {} # reset
file_idx += 1
if mode == "train": self.save_original_examples(original_data_sample, data_name)
self.save_converted_examples(data_name)
self.copy_related_files(data_name, exp_list)
print("*"*10, f"finishing processing dataset {data_name}", "*"*10)
def samsum(self):
"""
1. achieved from HF datasets "samsum"
2. no sys/user, but two human being, assuming the first utterance comes from user, ignore residual
"""
data_name = "SAMSum"
# prepare data
from datasets import load_dataset
data = load_dataset("samsum")
for mode in ["train", "val", "test"]:
real_name = mode if mode != "val" else "validation"
new_data, file_idx = {}, 1
for dial_idx, dial in tqdm(enumerate(data[real_name])):
new_dial_id = f"{data_name}--{mode}--{dial_idx+1}"
new_dial = self.init_dial(dial_idx=dial_idx+1, ori_dial_id=dial["id"]) # idx starts from 1
new_dial[ORI_DIAL_INFO] = {
"summary" : dial["summary"]
}
dial_hist = []
sep = "\r\n" if "\r\n" in dial["dialogue"] else "\n"
for turn_idx, turn in enumerate(dial["dialogue"].split(sep)):
speaker, utt = turn.split(": ")[0], ": ".join(turn.split(": ")[1:])
if turn_idx % 2 == 0:
new_turn = self.init_turn(turn_id=turn_idx//2+1)
new_turn[DIAL_HIST] = " ".join(dial_hist)
new_turn[USR_UTT] = utt.strip().replace(" ", " ")
new_turn[ORI_USR_ANN]['speaker'] = speaker
else:
new_turn[SYS_UTT] = utt.strip().replace(" ", " ")
new_turn[ORI_SYS_ANN]['speaker'] = speaker
dial_hist.append("<USER> " + new_turn[USR_UTT])
dial_hist.append("<SYSTEM> " + new_turn[SYS_UTT])
new_dial[LOG].append(new_turn)
new_data[new_dial_id] = new_dial
if (dial_idx+1) % 1000 == 0 or dial_idx+1 == len(data[real_name]):
self.save_dial(new_data, data_name=data_name, file_idx=file_idx, mode=mode)
new_data = {} # reset
file_idx += 1
self.save_original_examples(data["train"][:5], data_name)
self.save_converted_examples(data_name)
print("*"*10, f"finishing processing dataset {data_name}", "*"*10)
def dialogsum(self):
"""
1. we use the data from github: https://github.com/cylnlp/dialogsum/tree/main/DialogSum_Data
but, it is also available from HF datasets "knkarthick/dialogsum"
2. no sys/user, but two human being, assuming the first utterance comes from user, ignore residual
"""
data_name = "DialogSum"
for mode in ["train", "val", "test"]:
real_name = mode if mode != "val" else "dev"
path = os.path.join(self.data_dir, data_name, f"DialogSum_Data/dialogsum.{real_name}.jsonl")
data = self._load_jsonl(path)
new_data, file_idx = {}, 1
for dial_idx, dial in tqdm(enumerate(data)):
new_dial_id = f"{data_name}--{mode}--{dial_idx+1}"
new_dial = self.init_dial(dial_idx=dial_idx+1, ori_dial_id=dial["fname"]) # idx starts from 1
for key in dial:
if key in ["fname", "dialogue"]: continue
new_dial[ORI_DIAL_INFO][key] = dial[key]
dial_hist = []
turns = dial["dialogue"].replace("PErson","Person").split("#Person")[1:]
for turn_idx, turn in enumerate(turns):
speaker, utt = turn.split("#:")
speaker = "Person" + speaker
utt = utt.replace("\n","").strip()
if turn_idx % 2 == 0:
new_turn = self.init_turn(turn_id=turn_idx//2+1)
new_turn[DIAL_HIST] = " ".join(dial_hist)
new_turn[USR_UTT] = utt.strip()
new_turn[ORI_USR_ANN]['speaker'] = speaker.replace("#","")
else:
new_turn[SYS_UTT] = utt.strip()
new_turn[ORI_SYS_ANN]['speaker'] = speaker.replace("#","")
dial_hist.append("<USER> " + new_turn[USR_UTT])
dial_hist.append("<SYSTEM> " + new_turn[SYS_UTT])
new_dial[LOG].append(new_turn)
new_data[new_dial_id] = new_dial
if (dial_idx+1) % 1000 == 0 or dial_idx+1 == len(data):
self.save_dial(new_data, data_name=data_name, file_idx=file_idx, mode=mode)
new_data = {} # reset
file_idx += 1
if mode == "train": self.save_original_examples(data[:5], data_name)
self.save_converted_examples(data_name)
self.copy_related_files(data_name, ['Baseline'])
print("*"*10, f"finishing processing dataset {data_name}", "*"*10)
def ami(self):
"""
download processed data from https://drive.google.com/drive/folders/1BbmaZnzG9WrqOO-D3h211NOJePotqwQJ
the data is separated into 6 files based on annotation
here we extract the dialog context based on file "dialogueActs"
no train/val/test split, consider all as train
no readme file needs to be copied
we use ABCD instead of USR_UTT/SYS_UTT
1. each dialog contains more than 2 speaker? yes A,B,C,D
2. speaking in any order? yes A->B->C->D
"""
data_name = "AMI"
mode = "train"
data_dir = os.path.join(self.data_dir, data_name, "dialogueActs")
new_data, dial_idx = {}, 1
for filename in os.listdir(data_dir):
dial = self._load_json(os.path.join(data_dir, filename))
new_dial = self.init_dial(dial_idx=dial_idx) # idx starts from 1
# # # save dialog log
new_dial[ORI_DIAL_INFO]["dialog history"] = []
for turn in dial:
new_dial[ORI_DIAL_INFO]["dialog history"].append(turn["speaker"] + " : " + turn["text"])
# # # save abstractive summary
if os.path.exists(os.path.join(self.data_dir, data_name, "abstractive", filename)):
abs_sum = self._load_json(os.path.join(self.data_dir, data_name, "abstractive", filename))
new_dial[ORI_DIAL_INFO]["abstractive summary"] = abs_sum
# # # save extractive summary
if os.path.exists(os.path.join(self.data_dir, data_name, "extractive", filename)):
ext_sum = self._load_json(os.path.join(self.data_dir, data_name, "extractive", filename))
new_dial[ORI_DIAL_INFO]["extractive summary"] = []
for ext_turn in ext_sum:
new_dial[ORI_DIAL_INFO]["extractive summary"].append(ext_turn["speaker"] + " : " + ext_turn["text"])
new_dial_id = f"{data_name}--{mode}--{dial_idx}"
new_dial[ORI_DIAL_ID] = filename
new_data[new_dial_id] = new_dial
dial_idx += 1
if dial_idx == 2:
self.save_original_examples(dial, data_name)
self.save_dial(new_data, data_name=data_name, file_idx=1, mode=mode)
self.save_converted_examples(data_name)
print("*"*10, f"finishing processing dataset {data_name}", "*"*10)
def icsi(self):
"""
similar as AMI
speak can last to A->J
"""
data_name = "ICSI"
mode = "train"
data_dir = os.path.join(self.data_dir, data_name, "dialogueActs")
new_data, dial_idx = {}, 1
for filename in os.listdir(data_dir):
dial = self._load_json(os.path.join(data_dir, filename))
new_dial = self.init_dial(dial_idx=dial_idx) # idx starts from 1
# # # save dialog log
new_dial[ORI_DIAL_INFO]["dialog history"] = []
for turn in dial:
new_dial[ORI_DIAL_INFO]["dialog history"].append(turn["speaker"] + " : " + turn["text"])
# # # save abstractive summary
if os.path.exists(os.path.join(self.data_dir, data_name, "abstractive", filename)):
abs_sum = self._load_json(os.path.join(self.data_dir, data_name, "abstractive", filename))
new_dial[ORI_DIAL_INFO]["abstractive summary"] = abs_sum
# # # save extractive summary
if os.path.exists(os.path.join(self.data_dir, data_name, "extractive", filename)):
ext_sum = self._load_json(os.path.join(self.data_dir, data_name, "extractive", filename))
new_dial[ORI_DIAL_INFO]["extractive summary"] = []
for ext_turn in ext_sum:
new_dial[ORI_DIAL_INFO]["extractive summary"].append(ext_turn["speaker"] + " : " + ext_turn["text"])
new_dial_id = f"{data_name}--{mode}--{dial_idx}"
new_dial[ORI_DIAL_ID] = filename
new_data[new_dial_id] = new_dial
dial_idx += 1
if dial_idx == 2:
self.save_original_examples(dial, data_name)
self.save_dial(new_data, data_name=data_name, file_idx=1, mode=mode)
self.save_converted_examples(data_name)
print("*"*10, f"finishing processing dataset {data_name}", "*"*10)
def qmsum(self):
data_name = "QMSum"
for mode in ["train", "val", "test"]:
path = os.path.join(self.data_dir, data_name, f"data/ALL/{mode}")
data = self._load_dir_json(path)
new_data, file_idx = {}, 1
for dial_idx, dial in tqdm(enumerate(data)):
new_dial_id = f"{data_name}--{mode}--{dial_idx+1}"
new_dial = self.init_dial(dial_idx=dial_idx+1)
for key_ in dial:
if key_ == "meeting_transcripts": continue
new_dial[ORI_DIAL_INFO][key_] = dial[key_]
new_dial[ORI_DIAL_INFO]["dialog history"] = []
for turn in dial["meeting_transcripts"]:
new_dial[ORI_DIAL_INFO]["dialog history"].append(turn["speaker"] + " : " + turn["content"])
new_data[new_dial_id] = new_dial
if (dial_idx+1) % 1000 == 0 or dial_idx+1 == len(data):
self.save_dial(new_data, data_name=data_name, file_idx=file_idx, mode=mode)
new_data = {} # reset
file_idx += 1
if mode == "train": self.save_original_examples(data[:5], data_name)
self.save_converted_examples(data_name)
self.copy_related_files(data_name, ['Baseline'])
print("*"*10, f"finishing processing dataset {data_name}", "*"*10)
def mediasum(self):
data_name = "MediaSum"
split_id = self._load_json(os.path.join(self.data_dir, data_name, "data/train_val_test_split.json"))
data = self._load_json(os.path.join(self.data_dir, data_name, "data/news_dialogue.json"))
split_id2mode, new_data, file_idx, dial_idx = {}, {}, {}, {}
for mode in ["train", "val", "test"]:
for dial_id in split_id[mode]:
split_id2mode[dial_id] = mode
new_data[mode], file_idx[mode], dial_idx[mode] = {}, 1, 1
for dial in tqdm(data):
new_dial = self.init_dial() # idx starts from 1
new_dial[ORI_DIAL_ID] = dial['id']
for key_ in dial:
if key_ in ["id", "utt", "speaker"]: continue
new_dial[ORI_DIAL_INFO][key_] = dial[key_]
dialog_log = []
for idx in range(len(dial["utt"])):
dialog_log.append(dial["speaker"][idx] + " : " + dial["utt"][idx])
new_dial[ORI_DIAL_INFO]["dialog history"] = dialog_log
mode = split_id2mode.get(dial["id"], "train")
new_dial_id = f"{data_name}--{mode}--{dial_idx[mode]}"
new_dial[DIAL_IDX] = dial_idx[mode]
new_data[mode][new_dial_id] = new_dial
dial_idx[mode] += 1
if len(new_data[mode]) == 1000:
self.save_dial(new_data[mode], data_name=data_name, file_idx=file_idx[mode], mode=mode)
new_data[mode] = {} # reset
file_idx[mode] += 1
# if there are some unsaved dialogs left, save it now
for mode in ["train", "val", "test"]:
if new_data[mode]:
self.save_dial(new_data[mode], data_name=data_name, file_idx=file_idx[mode], mode=mode)
self.save_original_examples(data[:5], data_name)
self.save_converted_examples(data_name)
self.copy_related_files(data_name, ["data"])
print("*"*10, f"finishing processing dataset {data_name}", "*"*10)
def crd3(self):
"""
For this dataset, we choose present only chunk_size=2 offset=0
some file are missing for chunk size = 2
"""
data_name = "CRD3"
exp_list = []
for filename in os.listdir(os.path.join(self.data_dir, data_name)):
if filename == "readme.txt": continue
if filename == "LICENSE": continue
exp_list.append(filename)
for mode in ["train", "val", "test"]:
new_data, file_idx, dial_idx = {}, 1, 1
for file_name in self._load_txt(os.path.join(self.data_dir, data_name, f"data/aligned data/{mode}_files")):
file_path = os.path.join(self.data_dir, data_name, f"data/aligned data/c=2/{file_name}_2_0.json")
if not os.path.exists(file_path): continue
data = self._load_json(file_path)
new_dial_id = f"{data_name}--{mode}--{dial_idx}"
new_dial = self.init_dial(dial_idx=dial_idx)
new_dial[ORI_DIAL_ID] = file_name
new_dial[ORI_DIAL_INFO] = data
new_data[new_dial_id] = new_dial
dial_idx += 1
if (dial_idx) % 1000 == 0:
self.save_dial(new_data, data_name=data_name, file_idx=file_idx, mode=mode)
new_data = {} # reset
file_idx += 1
if new_data: self.save_dial(new_data, data_name=data_name, file_idx=file_idx, mode=mode)
if mode == "train": self.save_original_examples([new_dial[ORI_DIAL_INFO]], data_name)
self.save_converted_examples(data_name)
self.copy_related_files(data_name, exp_list)
print("*"*10, f"finishing processing dataset {data_name}", "*"*10)
def ectsum(self):
data_name = "ECTSum"
for mode in ["train", "val", "test"]:
new_data, file_idx, dial_idx = {}, 1, 1
data_dir = os.path.join(self.data_dir, data_name, "data/final", mode)
for file_name in os.listdir(os.path.join(data_dir, "ects")):
if not file_name.endswith("txt"): pdb.set_trace()
ect_data = self._load_txt(os.path.join(data_dir, "ects", file_name))
sum_data = self._load_txt(os.path.join(data_dir, "gt_summaries", file_name))
new_dial_id = f"{data_name}--{mode}--{dial_idx}"
new_dial = self.init_dial(dial_idx=dial_idx)
new_dial[ORI_DIAL_INFO]["file_name"] = file_name
new_dial[ORI_DIAL_INFO]["ect"] = ect_data
new_dial[ORI_DIAL_INFO]["summary"] = sum_data
new_data[new_dial_id] = new_dial
dial_idx += 1
if (dial_idx) % 1000 == 0:
self.save_dial(new_data, data_name=data_name, file_idx=file_idx, mode=mode)
new_data = {} # reset
file_idx += 1
if new_data: self.save_dial(new_data, data_name=data_name, file_idx=file_idx, mode=mode)
if mode == "train": self.save_original_examples([new_dial[ORI_DIAL_INFO]], data_name)
self.save_converted_examples(data_name)
self.copy_related_files(data_name, ['codes', 'data'])
print("*"*10, f"finishing processing dataset {data_name}", "*"*10)
def run_all(self):
# self.todsum()
# self.tweetsum()
# self.samsum()
# self.dialogsum()
# self.ami()
# self.icsi()
# self.qmsum()
self.mediasum()
# self.crd3()
# self.ectsum()
pass
def copy_example(self):
source_dir = self.save_dir
target_dir = "/home/qkun/projs/TOD-Project/Datasets/Dialogue-Summarization_PROCESSED/"
file_list = ["converted_examples.json", "original_examples.json", "readme.txt", "LICENSE"]
for dir_name in sorted(os.listdir(source_dir)):
if os.path.isfile(os.path.join(source_dir, dir_name)): continue
if not os.path.exists(os.path.join(target_dir, dir_name)): os.makedirs(os.path.join(target_dir, dir_name))
for filename in file_list:
source_path = os.path.join(source_dir, dir_name, filename)
target_path = os.path.join(target_dir, dir_name, filename)
if not os.path.exists(source_path): continue
shutil.copy(source_path, target_path)
def main():
preprocess = PreProcessData()
preprocess.run_all()
preprocess.copy_example()
if __name__ == '__main__':
main()