cleanup

MuCUE.py

@@ -1,165 +0,0 @@
-import json
-import os
-import datasets
-
-# --- 1. 基本信息 (无改动) ---
-_CITATION = """\
-@misc{jiang2025advancingfoundationmodelmusic,
-      title={Advancing the Foundation Model for Music Understanding},
-      author={Yi Jiang and Wei Wang and Xianwen Guo and Huiyun Liu and Hanrui Wang and Youri Xu and Haoqi Gu and Zhongqian Xie and Chuanjiang Luo},
-      year={2025},
-      eprint={2508.01178},
-      archivePrefix={arXiv},
-      primaryClass={cs.SD},
-      url={https://arxiv.org/abs/2508.01178},
-}
-"""
-_DESCRIPTION = "MuCUE: A Comprehensive Benchmark for Music Understanding"
-_REPO_ID = "Yi3852/MuCUE"
-
-# --- 2. 关键部分：定义所有Split的名称 (无改动) ---
-ALL_SPLITS = [
-    'ballroom_genres', 'ballroom_tempo', 'fma_medium', 'fma_small',
-    'gs_key', 'gs_key_30s', 'gtzan', 'gtzan_key', 'gtzan_tempo',
-    'guitarset', 'ins_cls', 'mcaps', 'md4q', 'mmau_music',
-    'mtg_genre', 'mtg_ins', 'mtg_mood', 'mucho', 'nsyn_ins',
-    'nsyn_pitch', 'salami_cnt', 'salami_overall', 'salami_pred',
-    'salami_segd', 'tat'
-]
-
-
-class MuCUEDataset(datasets.GeneratorBasedBuilder):
-    """MuCUE数据集加载器，支持批量预下载音频并返回本地绝对路径。"""
-
-    _AUDIO_BASE_URL = f"https://huggingface.co/datasets/{_REPO_ID}/resolve/main/audio/"
-
-    BUILDER_CONFIGS = [
-        datasets.BuilderConfig(
-            name="default",
-            version=datasets.Version("1.0.0"),
-            description="默认配置，预下载所有音频文件。",
-        )
-    ]
-
-    DEFAULT_CONFIG_NAME = "default"
-
-    def _info(self):
-        # 此方法无改动
-        return datasets.DatasetInfo(
-            description=_DESCRIPTION,
-            features=datasets.Features(
-                {
-                    "id": datasets.Value("string"),
-                    "question": datasets.Value("string"),
-                    "audio": datasets.Audio(sampling_rate=16_000),
-                    "time_seg": datasets.Sequence(datasets.Value("float32")),
-                    "choices": datasets.Sequence(datasets.Value("string")),
-                    "choices_text": datasets.Value("string"),
-                    "correct_choice": datasets.Value("string"),
-                    "answer": datasets.Value("string"),
-                }
-            ),
-            citation=_CITATION,
-        )
-
-    def _split_generators(self, dl_manager):
-        """
-        此方法负责下载数据、定义split，并为_generate_examples准备参数。
-        """
-        data_dir = self.base_path
-
-        # 1. 扫描所有本地的jsonl文件，收集音频的相对路径并构造其下载URL (逻辑不变)
-        relative_path_to_url = {}
-        split_samples = {}
-
-        print("步骤 1: 扫描本地jsonl文件，收集音频路径...")
-        for split_name in ALL_SPLITS:
-            filepath = os.path.join(data_dir, f"{split_name}.jsonl")
-            if os.path.exists(filepath):
-                samples = []
-                with open(filepath, "r", encoding="utf-8") as f:
-                    for line in f:
-                        sample = json.loads(line)
-                        samples.append(sample)
-
-                        relative_path = sample.get("audio")
-                        if isinstance(relative_path, str) and not relative_path.startswith("http"):
-                            if relative_path not in relative_path_to_url:
-                                url_path = relative_path.replace('\\', '/')
-                                full_url = self._AUDIO_BASE_URL + url_path
-                                relative_path_to_url[relative_path] = full_url
-                split_samples[split_name] = samples
-
-        # --- 从这里开始是关键改动 ---
-
-        # 2. 使用dl_manager批量下载所有音频LFS文件
-        print(f"步骤 2: 找到 {len(relative_path_to_url)} 个唯一的音频文件，开始并行下载...")
-        
-        # 将所有唯一的URL收集到一个列表中
-        urls_to_download = list(relative_path_to_url.values())
-        
-        # 【核心优化】一次性将整个URL列表传递给dl_manager.download()
-        # datasets库会自动处理并行下载，并显示一个总的进度条。
-        # 它返回一个本地路径的列表，顺序与urls_to_download完全对应。
-        downloaded_local_paths = dl_manager.download(urls_to_download)
-        
-        # 创建一个从URL到其下载后本地路径的映射，方便后续查找
-        url_to_local_path_map = dict(zip(urls_to_download, downloaded_local_paths))
-
-        # 3. 创建从相对路径到本地绝对路径的最终映射
-        relative_path_to_local_path = {}
-        for relative_path, full_url in relative_path_to_url.items():
-            local_path = url_to_local_path_map.get(full_url)
-            if local_path:
-                relative_path_to_local_path[relative_path] = local_path
-            else:
-                # 这种情况通常发生在某个文件下载失败时
-                print(f"警告：未能下载或找到 {full_url} 的本地路径")
-
-        print("步骤 3: 所有音频文件已下载并缓存。开始生成数据集 splits...")
-        
-        # --- 关键改动结束 ---
-
-        # 4. 为每个split创建SplitGenerator (逻辑不变)
-        split_generators = []
-        for split_name in ALL_SPLITS:
-            if split_name in split_samples:
-                split_generators.append(
-                    datasets.SplitGenerator(
-                        name=split_name,
-                        gen_kwargs={
-                            "samples": split_samples[split_name],
-                            "path_map": relative_path_to_local_path,
-                        },
-                    )
-                )
-
-        return split_generators
-
-    def _generate_examples(self, samples, path_map):
-        """
-        生成样本。此部分逻辑不变，高效且正确。
-        """
-        for i, sample in enumerate(samples):
-            key = f"{i}_{sample['id']}"
-            relative_audio_path = sample.get("audio")
-            if not isinstance(relative_audio_path, str):
-                continue
-                
-            local_audio_path = path_map.get(relative_audio_path)
-            
-            if not local_audio_path:
-                # 如果文件下载失败，可以选择跳过该样本
-                # print(f"警告: 找不到样本 {sample['id']} 的本地音频路径: {relative_audio_path}")
-                continue
-
-            yield key, {
-                "id": sample["id"],
-                "question": sample["question"],
-                "audio": local_audio_path, # 提供绝对路径
-                "time_seg": sample["time_seg"],
-                "choices": sample["choices"],
-                "choices_text": sample["choices_text"],
-                "correct_choice": sample["correct_choice"],
-                "answer": sample["answer"],
-            }