Delete build_dataset.py

build_dataset.py

@@ -1,431 +0,0 @@
-import json
-import re
-
-from datasets import Dataset, DatasetDict
-import os
-
-from PIL import Image
-import hashlib
-import shutil
-
-import openai
-
-
-VERSION = "20250515"
-LANG = "zh"
-OUTPUT_PATH = "/fs-computility/ai4sData/earth-shared/SFE/sfe_dataset/v0"
-
-chemistry_dir = "/fs-computility/ai4sData/earth-shared/SFE/v0/chem"
-life_dir = "/fs-computility/ai4sData/earth-shared/SFE/v0/life"
-astro_dir = "/fs-computility/ai4sData/earth-shared/SFE/v0/astro"
-earth_dir = "/fs-computility/ai4sData/earth-shared/SFE/v0/earth"
-mat_dir = "/fs-computility/ai4sData/earth-shared/SFE/v0/mat"
-
-VERSION = f"{VERSION}-{LANG}"
-
-ENABLE_FILETER = False
-FILTERS = ["e014"]
-
-def enumerate_question_types(question_type):
-    if question_type.lower() in ["exact match"]:
-        return "exact_match"
-    elif question_type.lower() in ["multiple choice"]:
-        return "mcq"
-    elif question_type.lower() in ["open question"]:
-        return "open_ended"
-    
-    raise ValueError(f"Unknown question type: {question_type}")
-
-
-def parse_float_sequence_within(input_str):
-    pattern_in_bracket = r"\[(.*)\]"
-    match = re.search(pattern_in_bracket, input_str)
-
-    if not match:
-        return None
-
-    inside_str = match.group(1)
-    groups = inside_str.split(";")
-
-    bboxs = []
-    for group in groups:
-        floats = group.split(",")
-        if len(floats) != 4:
-            continue
-        try:
-            bboxs.append([float(f) for f in floats])
-        except Exception as e:
-            continue
-
-    if len(bboxs) == 0:
-        return None
-
-    return bboxs
-
-
-if __name__ == "__main__":
-    question_ids = []
-    question_texts = []
-    question_options = []
-    question_images = []
-    question_answers = []
-    question_types = []
-    question_fields = []
-
-    # load astro data
-    for folder_name in os.listdir(astro_dir):
-        if not folder_name.startswith("A"):
-            continue
-
-        if ENABLE_FILETER and not folder_name.lower() in FILTERS:
-            continue
-
-        folder_path = os.path.join(astro_dir, folder_name)
-        if not os.path.isdir(folder_path):
-            continue
-        
-        file_processed = False
-        for file_name in os.listdir(folder_path):
-            if not file_name.endswith(".jsonl"):
-                continue
-            if not file_name.startswith(LANG):
-                continue
-            file_processed = True
-            file_path = os.path.join(folder_path, file_name)
-            print(file_path)
-            with open(file_path, "r") as f:
-                for line in f:
-                    data = json.loads(line)
-                    data = {k.lower(): v for k, v in data.items()}
-
-                    images = data["images"]
-                    assert len(images) != 0
-
-                    abs_images = []
-                    for image in images:
-                        image_path = os.path.join(astro_dir, image)
-                        assert os.path.exists(image_path), f"Image not found: {image_path}"
-                        md5hash = hashlib.md5(Image.open(image_path).tobytes()).hexdigest()
-                        new_image_path = os.path.join(OUTPUT_PATH, "images", f"{image.split('.')[0].replace('/', '_')}_{md5hash}.{image.split('.')[1]}")
-                        if not os.path.exists(new_image_path):
-                            shutil.copy(image_path, new_image_path)
-                        # abs_images.append(new_image_path)
-                        abs_images.append(f"{image.split('.')[0].replace('/', '_')}_{md5hash}.{image.split('.')[1]}")
-                    
-                    question_text = data["text"]
-                    missing_images = len(images) - question_text.count("<image>") - sum([option.count("<image>") for option in data["answer choices"]])
-                    question_text = " ".join(["<image>" for _ in range(missing_images)]) + " " + question_text
-                    question_text = question_text.strip()
-                    assert len(images) == question_text.count("<image>") + sum([option.count("<image>") for option in data["answer choices"]])
-
-                    question_options_ = data["answer choices"]
-                    question_type = enumerate_question_types(data["question type"])
-                    question_answer = str(data["ground truth"])
-
-                    question_ids.append(data["question_id"])
-                    question_texts.append(question_text)
-                    question_options.append(question_options_)
-                    question_images.append(abs_images)
-                    question_answers.append(question_answer)
-                    question_types.append(question_type)
-                    question_fields.append("astronomy")
-        if not file_processed:
-            print(folder_path)
-            raise
-    print("astro data loaded")
-
-    # load chem data
-    for folder_name in os.listdir(chemistry_dir):
-        if not folder_name.startswith("C"):
-            continue
-
-        if ENABLE_FILETER and not folder_name.lower() in FILTERS:
-            continue
-
-        folder_path = os.path.join(chemistry_dir, folder_name)
-        if not os.path.isdir(folder_path):
-            continue
-        
-        file_processed = False
-        for file_name in os.listdir(folder_path):
-            if not file_name.endswith(".jsonl"):
-                continue
-            if not file_name.startswith(LANG):
-                continue
-            file_processed = True
-            file_path = os.path.join(folder_path, file_name)
-            print(file_path)
-            with open(file_path, "r") as f:
-                for line in f:
-                    data = json.loads(line)
-                    data = {k.lower(): v for k, v in data.items()}
-
-                    images = data["images"]
-                    assert len(images) != 0
-
-                    abs_images = []
-                    for image in images:
-                        image_path = os.path.join(chemistry_dir, image)
-                        if not os.path.exists(image_path):
-                            image_path = os.path.join(folder_path, image)
-                        assert os.path.exists(image_path), f"Image not found: {image_path}"
-                        md5hash = hashlib.md5(Image.open(image_path).tobytes()).hexdigest()
-                        new_image_path = os.path.join(OUTPUT_PATH, "images", f"{image.split('.')[0].replace('/', '_')}_{md5hash}.{image.split('.')[1]}")
-                        if not os.path.exists(new_image_path):
-                            shutil.copy(image_path, new_image_path)
-                        abs_images.append(new_image_path)
-                    
-                    question_text = data["text"]
-                    missing_images = len(images) - question_text.count("<image>") - sum([option.count("<image>") for option in data["answer choices"]])
-                    question_text = " ".join(["<image>" for _ in range(missing_images)]) + " " + question_text
-                    question_text = question_text.strip()
-                    assert len(images) == question_text.count("<image>") + sum([option.count("<image>") for option in data["answer choices"]])
-
-                    question_options_ = data["answer choices"]
-                    question_type = enumerate_question_types(data["question type"])
-                    question_answer = str(data["ground truth"])
-
-                    question_ids.append(data["question_id"])
-                    question_texts.append(question_text)
-                    question_options.append(question_options_)
-                    question_images.append(abs_images)
-                    question_answers.append(question_answer)
-                    question_types.append(question_type)
-                    question_fields.append("chemistry")
-        if not file_processed:
-            print(folder_path)
-            raise
-    print("chem data loaded")
-
-    # load earth data
-    for folder_name in os.listdir(earth_dir):
-        if not folder_name.startswith("E"):
-            continue
-
-        if ENABLE_FILETER and not folder_name.lower() in FILTERS:
-            continue
-
-        folder_path = os.path.join(earth_dir, folder_name)
-        if not os.path.isdir(folder_path):
-            continue
-        
-        file_processed = False
-        for file_name in os.listdir(folder_path):
-            if not file_name.endswith(".jsonl"):
-                continue
-            if not file_name.startswith(LANG):
-                continue
-            file_processed = True
-            file_path = os.path.join(folder_path, file_name)
-            print(file_path)
-            with open(file_path, "r") as f:
-                for line in f:
-                    if line.strip() == "":
-                        continue
-                    data = json.loads(line)
-                    data = {k.lower(): v for k, v in data.items()}
-
-                    images = data["images"] if "images" in data else data["image"] 
-                    assert len(images) != 0
-
-                    abs_images = []
-                    for image in images:
-                        image_path = os.path.join(earth_dir, image)
-                        if not os.path.exists(image_path):
-                            image_path = os.path.join(folder_path, image)
-                        assert os.path.exists(image_path), f"Image not found: {image_path}"
-                        md5hash = hashlib.md5(Image.open(image_path).tobytes()).hexdigest()
-                        new_image_path = os.path.join(OUTPUT_PATH, "images", f"{image.split('.')[0].replace('/', '_')}_{md5hash}.{image.split('.')[1]}")
-                        if not os.path.exists(new_image_path):
-                            shutil.copy(image_path, new_image_path)
-                        abs_images.append(new_image_path)
-                    
-                    question_text = data["text"]
-                    missing_images = len(images) - question_text.count("<image>") - sum([option.count("<image>") for option in data["answer choices"]])
-                    question_text = " ".join(["<image>" for _ in range(missing_images)]) + " " + question_text
-                    question_text = question_text.strip()
-                    assert len(images) == question_text.count("<image>") + sum([option.count("<image>") for option in data["answer choices"]])
-
-                    if folder_name in ["E011", "E012"]:
-                        print("adding task-specific prompts")
-                        task_prompt = open(os.path.join(folder_path, f"{LANG}.prompt.txt")).read()
-                        question_text = f"{question_text}\n\n{task_prompt}"
-
-                        assert parse_float_sequence_within(str(data["ground truth"])) is not None
-
-                    question_options_ = data["answer choices"]
-                    question_type = enumerate_question_types(data["question type"])
-                    question_answer = str(data["ground truth"])
-
-                    question_ids.append(data["question_id"])
-                    question_texts.append(question_text)
-                    question_options.append(question_options_)
-                    question_images.append(abs_images)
-                    question_answers.append(question_answer)
-                    question_types.append(question_type)
-                    question_fields.append("earth")
-        if not file_processed:
-            print(folder_path)
-            raise
-    print("earth data loaded")
-
-    # load life data
-    for folder_name in os.listdir(life_dir):
-        if not folder_name.startswith("L"):
-            continue
-
-        if ENABLE_FILETER and not folder_name.lower() in FILTERS:
-            continue
-
-        folder_path = os.path.join(life_dir, folder_name)
-        if not os.path.isdir(folder_path):
-            continue
-        
-        file_processed = False
-        for file_name in os.listdir(folder_path):
-            if not file_name.endswith(".jsonl"):
-                continue
-            if not file_name.startswith(LANG):
-                continue
-            file_processed = True
-            file_path = os.path.join(folder_path, file_name)
-            print(file_path)
-            with open(file_path, "r") as f:
-                for line in f:
-                    if line.strip() == "":
-                        continue
-                    data = json.loads(line)
-                    data = {k.lower(): v for k, v in data.items()}
-
-                    images = data["images"] if "images" in data else data["image"] 
-                    assert len(images) != 0
-
-                    abs_images = []
-                    for image in images:
-                        image_path = os.path.join(life_dir, image)
-                        if not os.path.exists(image_path):
-                            image_path = os.path.join(folder_path, image)
-                        assert os.path.exists(image_path), f"Image not found: {image_path}"
-                        md5hash = hashlib.md5(Image.open(image_path).tobytes()).hexdigest()
-                        new_image_path = os.path.join(OUTPUT_PATH, "images", f"{image.split('.')[0].replace('/', '_')}_{md5hash}.{image.split('.')[1]}")
-                        if not os.path.exists(new_image_path):
-                            shutil.copy(image_path, new_image_path)
-                        abs_images.append(new_image_path)
-                    
-                    question_text = data["text"]
-                    missing_images = len(images) - question_text.count("<image>") - sum([option.count("<image>") for option in data["answer choices"]])
-                    question_text = " ".join(["<image>" for _ in range(missing_images)]) + " " + question_text
-                    question_text = question_text.strip()
-                    assert len(images) == question_text.count("<image>") + sum([option.count("<image>") for option in data["answer choices"]])
-
-                    question_options_ = data["answer choices"]
-                    question_type = enumerate_question_types(data["question type"])
-                    question_answer = str(data["ground truth"])
-
-                    question_ids.append(data["question_id"])
-                    question_texts.append(question_text)
-                    question_options.append(question_options_)
-                    question_images.append(abs_images)
-                    question_answers.append(question_answer)
-                    question_types.append(question_type)
-                    question_fields.append("life")
-        if not file_processed:
-            print(folder_path)
-            raise
-    print("life data loaded")
-
-    # load mat data
-    for folder_name in os.listdir(mat_dir):
-        if not folder_name.startswith("M"):
-            continue
-
-        if ENABLE_FILETER and not folder_name.lower() in FILTERS:
-            continue
-
-        folder_path = os.path.join(mat_dir, folder_name)
-        if not os.path.isdir(folder_path):
-            continue
-        
-        file_processed = False
-        for file_name in os.listdir(folder_path):
-            if not file_name.endswith(".jsonl"):
-                continue
-            if not file_name.startswith(LANG):
-                continue
-            file_processed = True
-            file_path = os.path.join(folder_path, file_name)
-            print(file_path)
-            with open(file_path, "r") as f:
-                for line in f:
-                    if line.strip() == "":
-                        continue
-                    data = json.loads(line)
-                    data = {k.lower(): v for k, v in data.items()}
-
-                    images = data["images"] if "images" in data else data["image"] 
-                    assert len(images) != 0
-
-                    abs_images = []
-                    for image in images:
-                        image_path = os.path.join(mat_dir, image)
-                        if not os.path.exists(image_path):
-                            image_path = os.path.join(folder_path, image)
-                        assert os.path.exists(image_path), f"Image not found: {image_path}"
-                        md5hash = hashlib.md5(Image.open(image_path).tobytes()).hexdigest()
-                        new_image_path = os.path.join(OUTPUT_PATH, "images", f"{image.split('.')[0].replace('/', '_')}_{md5hash}.{image.split('.')[1]}")
-                        if not os.path.exists(new_image_path):
-                            shutil.copy(image_path, new_image_path)
-                        abs_images.append(new_image_path)
-                    
-                    question_text = data["text"]
-                    missing_images = len(images) - question_text.count("<image>") - sum([option.count("<image>") for option in data["answer choices"]])
-                    question_text = " ".join(["<image>" for _ in range(missing_images)]) + " " + question_text
-                    question_text = question_text.strip()
-                    assert len(images) == question_text.count("<image>") + sum([option.count("<image>") for option in data["answer choices"]])
-
-                    question_options_ = data["answer choices"]
-                    question_type = enumerate_question_types(data["question type"])
-                    question_answer = str(data["ground truth"])
-
-                    question_ids.append(data["question_id"])
-                    question_texts.append(question_text)
-                    question_options.append(question_options_)
-                    question_images.append(abs_images)
-                    question_answers.append(question_answer)
-                    question_types.append(question_type)
-                    question_fields.append("material")
-        if not file_processed:
-            print(folder_path)
-            raise
-    print("mat data loaded")
-    
-    dataset_dict = {
-        "id": question_ids,
-        "question": question_texts,
-        "options": question_options,
-        "answer": question_answers,
-        "images": question_images,
-        "question_type": question_types,
-        "field": question_fields,
-        "lang": [LANG for _ in range(len(question_ids))],
-    }
-
-    dataset = Dataset.from_dict(dataset_dict)
-
-    dataset_split = DatasetDict({
-        "test": dataset
-    })
-    print(dataset_split)
-    # get the first row
-    print(dataset_split["test"][0])
-
-    # save dataset
-    os.makedirs(os.path.join(OUTPUT_PATH, VERSION), exist_ok=True)
-    test_split_path = os.path.join(OUTPUT_PATH, VERSION, "test.parquet")
-    dataset_split["test"].to_parquet(test_split_path)
-
-    metadata_path = os.path.join(OUTPUT_PATH, VERSION, "dataset_info.json")
-    with open(metadata_path, "w") as f:
-        json.dump({"splits": ["test"]}, f)
-    # dataset.to_parquet(os.path.join(OUTPUT_PATH, VERSION, "data.parquet"))
-