FSL_ECG_QA_Dataset / data /load_class.py
jialucode's picture
Upload 3 files
0fb79ec verified
Raw
History Blame Contribute Delete
16.1 kB
import os
import glob
import json
import argparse
from typing import List, Dict, Any, Union
from collections import Counter
import numpy as np
def get_temp_qa(question_type, paraphrased_path):
"""
Load all question data and return a list of filtered template IDs based on question_type.
Args:
question_type (str): Type of question to filter by ('single-query', 'single-verify', etc., or 'all').
paraphrased_path (str): Base path for the dataset.
Returns:
List[int]: A list of unique template IDs matching the filter criteria.
"""
data = []
for split in ["train", "valid", "test"]:
file_pattern = os.path.join(paraphrased_path, split, "*.json")
for fname in sorted(glob.glob(file_pattern)):
with open(fname, "r") as f:
records = json.load(f)
data.extend(records)
print(f"Loaded {len(data)} samples!")
temp_list = []
if question_type != "all":
for item in data:
if item['question_type'] == question_type:
temp_list.append(item['template_id'])
else:
for item in data:
if "single-" in item['question_type']:
temp_list.append(item['template_id'])
temp_ids = list(set(temp_list))
# Filter template IDs to avoid conflict class
temp_ids = [i for i in temp_ids if i not in [17, 14, 26, 28, 4, 40, 31]]
return temp_ids
def prepare_ecg_qa_data(args):
if args.question_type == "all":
class_qa = get_class_qa("single-verify")
ecg_qa_list_1 = change_ecg_to_qa(
class_qa, "single-verify",
paraphrased_path=args.paraphrased_path,
dif_exp=args.dif_exp
)
class_qa += get_class_qa("single-choose")
ecg_qa_list_2 = change_ecg_to_qa(
class_qa, "single-choose",
paraphrased_path=args.paraphrased_path,
dif_exp=args.dif_exp
)
class_qa += get_class_qa("single-query")
ecg_qa_list_3 = change_ecg_to_qa(
class_qa, "single-query",
paraphrased_path=args.paraphrased_path,
dif_exp=args.dif_exp
)
ecg_qa_list = {**ecg_qa_list_1, **ecg_qa_list_2, **ecg_qa_list_3}
else:
class_qa = get_temp_qa(args.question_type, args.paraphrased_path)
ecg_qa_list = change_ecg_to_qa(
class_qa, args.question_type,
paraphrased_path=args.paraphrased_path,
dif_exp=args.dif_exp
)
all_ecg_qa_temp = list(ecg_qa_list.keys())
sample_size = int(len(all_ecg_qa_temp) * 0.8)
train_temp = np.random.choice(all_ecg_qa_temp, sample_size, replace=False).tolist()
test_temp = [i for i in all_ecg_qa_temp if i not in train_temp]
return class_qa, train_temp, test_temp
def change_ecg_to_qa(
sample_ids,
question_type,
paraphrased_path,
in_template=None,
dif_exp=1,
attr="",
test_dataset="ptb-xl"
):
"""
Process ECG data to create a dictionary of question-answer pairs categorized by template.
Args:
sample_ids (List[int]): List of template IDs to include.
question_type (str): Type of question to filter.
paraphrased_path (str): Base path for all data.
data_root_path (str, optional): Path to the data root directory.
in_template (List, optional): List of templates to include.
dif_exp (int, optional): Controls return format. Default is 1.
attr (str, optional): Attribute type filter.
test_dataset (str, optional): Dataset to use ('ptb-xl' or 'mimic'). Default is 'ptb-xl'.
Returns:
Dict: Dictionary of ECG QA data categorized by template.
"""
data = []
# Define subdirectories to process
subdirectories = ["train", "valid", "test"]
# Load data from all subdirectories
for subdir in subdirectories:
directory_path = os.path.join(paraphrased_path, subdir)
if not os.path.exists(directory_path):
print(f"Warning: Directory {directory_path} does not exist!")
continue
file_pattern = os.path.join(directory_path, "*.json")
files = sorted(glob.glob(file_pattern))
if not files:
print(f"Warning: No JSON files found in {directory_path}")
continue
for fname in files:
with open(fname, "r") as f:
json_data = json.load(f)
data.extend(json_data)
# Filter by attribute type if specified
if attr != "":
sample_data = [item for item in data if item['attribute_type'] == attr]
else:
sample_data = data
# Filter samples by file existence if using MIMIC dataset
if test_dataset == "mimic":
ecg_data_path = os.path.join(paraphrased_path, "mimic_iv_ecg/processed_test_30k/")
if not os.path.exists(ecg_data_path):
print(f"Warning: MIMIC ECG data path {ecg_data_path} does not exist!")
sample_data = [sample for sample in sample_data if os.path.isfile(os.path.join(ecg_data_path, f"{sample['ecg_id'][0]}.mat"))]
ecg_qa_dict = {}
if len(sample_data) == 0:
print(f"Cannot find template_id == {sample_ids} or no data available")
else:
for sample in sample_data:
template_id = sample['template_id']
if template_id in sample_ids:
process_sample_by_type(sample, template_id, ecg_qa_dict, in_template)
filter_ecg_qa_dict_by_question_type(ecg_qa_dict, question_type)
if dif_exp == 1:
return {key: [value] for key, value in ecg_qa_dict.items()}
return filter_by_question_frequency(ecg_qa_dict)
def process_sample_by_type(sample, template_id, ecg_qa_dict, in_template):
"""
Process a sample based on its question type.
Args:
sample (Dict): The sample data.
template_id (int): The template ID.
ecg_qa_dict (Dict): Dictionary to populate with processed data.
in_template (List, optional): List of templates to include.
"""
question_type = sample['question_type']
if question_type == 'single-verify':
process_single_verify_sample(sample, template_id, ecg_qa_dict, in_template)
elif question_type == 'single-choose':
process_single_choose_sample(sample, template_id, ecg_qa_dict, in_template)
elif question_type == 'single-query':
process_single_query_sample(sample, template_id, ecg_qa_dict, in_template)
def process_single_verify_sample(sample, template_id, ecg_qa_dict, in_template):
"""
Process a single-verify type sample.
Args:
sample (Dict): The sample data.
template_id (int): The template ID.
ecg_qa_dict (Dict): Dictionary to populate with processed data.
in_template (List, optional): List of templates to include.
"""
if sample['answer'][0] in ["yes", "no"]:
answer = sample['answer'][0]
all_attributes = "_".join(sorted(sample['attribute']))
dict_key = f"{template_id}_{all_attributes}_{answer}"
if in_template is None or dict_key in in_template:
if dict_key not in ecg_qa_dict:
ecg_qa_dict[dict_key] = [sample]
else:
ecg_qa_dict[dict_key].append(sample)
def process_single_choose_sample(sample, template_id, ecg_qa_dict, in_template):
"""
Process a single-choose type sample.
Args:
sample (Dict): The sample data.
template_id (int): The template ID.
ecg_qa_dict (Dict): Dictionary to populate with processed data.
in_template (List, optional): List of templates to include.
"""
if len(sample['answer']) == 1:
answer = sample['answer'][0]
elif len(sample['answer']) == 2:
answer = "both"
sample['answer'] = ["both"]
else:
print("single-choose data have more than 2 answers!")
return
all_attributes = "_".join(sorted(sample['attribute']))
if in_template is None:
handle_single_choose_without_template(sample, template_id, answer, all_attributes, ecg_qa_dict)
else:
handle_single_choose_with_template(sample, template_id, answer, all_attributes, in_template, ecg_qa_dict)
def handle_single_choose_without_template(sample, template_id, answer, all_attributes, ecg_qa_dict):
"""
Handle a single-choose type sample when no template is provided.
Args:
sample (Dict): The sample data.
template_id (int): The template ID.
answer (str): The answer string.
all_attributes (str): The joined attributes string.
ecg_qa_dict (Dict): Dictionary to populate with processed data.
"""
if answer == "both":
dict_key = f"{template_id}_{all_attributes}_{answer}"
add_to_ecg_qa_dict(dict_key, sample, ecg_qa_dict)
elif answer == "none":
for attr in sample['attribute']:
dict_key = f"{template_id}_{attr}_{answer}"
add_to_ecg_qa_dict(dict_key, sample, ecg_qa_dict)
else:
dict_key = f"{template_id}_{answer}_{answer}"
add_to_ecg_qa_dict(dict_key, sample, ecg_qa_dict)
def handle_single_choose_with_template(sample, template_id, answer, all_attributes, in_template, ecg_qa_dict):
"""
Handle a single-choose type sample when a template is provided.
Args:
sample (Dict): The sample data.
template_id (int): The template ID.
answer (str): The answer string.
all_attributes (str): The joined attributes string.
in_template (List): List of templates to include.
ecg_qa_dict (Dict): Dictionary to populate with processed data.
"""
full_key = f"{template_id}_{all_attributes}_{answer}"
short_key = f"{template_id}_{answer}"
if full_key in in_template or short_key in in_template:
if answer == "both":
dict_key = f"{template_id}_{all_attributes}_{answer}"
add_to_ecg_qa_dict(dict_key, sample, ecg_qa_dict)
elif answer == "none":
for attr in sample['attribute']:
dict_key = f"{template_id}_{attr}_{answer}"
add_to_ecg_qa_dict(dict_key, sample, ecg_qa_dict)
else:
dict_key = f"{template_id}_{answer}_{answer}"
add_to_ecg_qa_dict(dict_key, sample, ecg_qa_dict)
def process_single_query_sample(sample, template_id, ecg_qa_dict, in_template):
"""
Process a single-query type sample.
Args:
sample (Dict): The sample data.
template_id (int): The template ID.
ecg_qa_dict (Dict): Dictionary to populate with processed data.
in_template (List, optional): List of templates to include.
"""
if len(sample['answer']) == 1:
answer = sample['answer'][0]
elif len(sample['answer']) >= 2:
answer = ", ".join(sample['answer'])
sample['answer'] = [answer]
all_attributes = "_".join(sorted(sample['attribute']))
dict_key = f"{template_id}_{all_attributes}_{answer}"
if in_template is None or dict_key in in_template:
if dict_key not in ecg_qa_dict:
ecg_qa_dict[dict_key] = [sample]
else:
ecg_qa_dict[dict_key].append(sample)
def add_to_ecg_qa_dict(key, sample, ecg_qa_dict):
"""
Add a sample to the ECG QA dictionary.
Args:
key (str): The dictionary key.
sample (Dict): The sample data.
ecg_qa_dict (Dict): Dictionary to populate with processed data.
"""
if key not in ecg_qa_dict:
ecg_qa_dict[key] = [sample]
else:
ecg_qa_dict[key].append(sample)
def filter_ecg_qa_dict_by_question_type(ecg_qa_dict, question_type):
"""
Filter the ECG QA dictionary by question type and minimum sample counts.
Args:
ecg_qa_dict (Dict): Dictionary to filter.
question_type (str): Type of question to filter by.
"""
for key in list(ecg_qa_dict.keys()):
if question_type in ["single-verify", "single-choose", "single-query"]:
qt = question_type
elif question_type == "all":
if len(ecg_qa_dict[key]) == 0:
del ecg_qa_dict[key]
continue
qt = ecg_qa_dict[key][0]['question_type']
else:
continue
if qt == "single-verify" and len(ecg_qa_dict[key]) < 140:
del ecg_qa_dict[key]
elif qt == "single-choose" and len(ecg_qa_dict[key]) < 14:
del ecg_qa_dict[key]
elif qt == "single-query" and len(ecg_qa_dict[key]) < 50:
del ecg_qa_dict[key]
def filter_by_question_frequency(ecg_qa_dict):
"""
Filter the ECG QA dictionary by question frequency.
Args:
ecg_qa_dict (Dict): Dictionary to filter.
Returns:
Dict: Filtered dictionary.
"""
for key in list(ecg_qa_dict.keys()):
question_id_counter = Counter(sample['question_id'] for sample in ecg_qa_dict[key])
frequent_question_ids = [q_id for q_id, count in question_id_counter.items() if count >= 10]
all_samples_by_question = []
for question_id in frequent_question_ids:
question_samples = [sample for sample in ecg_qa_dict[key] if sample['question_id'] == question_id]
all_samples_by_question.append(question_samples)
if len(all_samples_by_question) != 0:
ecg_qa_dict[key] = all_samples_by_question
else:
del ecg_qa_dict[key]
return ecg_qa_dict
def validate_path(path):
"""
Validate if a path exists.
Args:
path (str): Path to validate.
Returns:
bool: True if path exists, False otherwise.
"""
if not os.path.exists(path):
return False
return True
def main():
"""
Main function to run the ECG data processing.
"""
parser = argparse.ArgumentParser(description='ECG Data Processor')
parser.add_argument('--paraphrased_path', type=str,
default='path/to/paraphrased',
help='path to ./paraphrased containing trian/val/test ECG-QA json files')
parser.add_argument('--test_dataset', type=str, default="ptb-xl", choices=["ptb-xl", "mimic"],
help='Dataset to use (ptb-xl or mimic)')
args = parser.parse_args()
paraphrased_path = args.paraphrased_path
# Process each question type
for q_type in ['single-verify', 'single-choose', 'single-query', 'all']:
print(f"\nProcessing {q_type} question type...")
# Get template IDs for the question type
temp_ids = get_temp_qa(q_type, paraphrased_path)
print(f"{q_type} temp_ids: {temp_ids} (total: {len(temp_ids)})")
# Get the QA dictionary
ecg_qa_dict = change_ecg_to_qa(
temp_ids,
q_type,
paraphrased_path=paraphrased_path,
test_dataset=args.test_dataset
)
print(f"ECG QA dictionary length: {len(ecg_qa_dict)}")
# Calculate total QA count
total_qa_count = 0
for category_key in ecg_qa_dict.keys():
category_values = ecg_qa_dict.get(category_key)
category_total = sum(len(item) for item in category_values)
total_qa_count += category_total
print(f"Total QA count for '{q_type}': {total_qa_count}")
print("-" * 40)
if __name__ == "__main__":
main()