Defeat Recording Data

Main Behavior Pickle File: Overall dictionary from (mouse,defeat_day)-->data data.keys() # body_part_tracking, extracted_features, fiber_photometry data['body_part_tracking] is a dictionary from body part to x,y coordinates over time (T frame, 120 Hz) data['extracted_features'] is a 12xT matrix of feature traces (120 Hz) data['fiber_photometry'] is a dictionary from 'fpts' or 'fpnac' to raw GCaMP recordings from the tail striatum or nucleus accumbens respectively (1017 Hz) Model Pickle Files: - 2 sets of random forest classifiers (trained RF model, cutoff for probability of behavior ocurring) dictionary of behavior name: (trained sklearn.ensemble.RandomForestClassifier, classification probability cutoff) - t-SNE model: dictionary: 'training': features in training set 'projection': sklearn.neural_network.MLPClassifier maps raw features to t-SNE space 'labeled_map': where clusters are in t-SNE space

主行为Pickle文件：以（小鼠，战败天数）为键的总字典，映射至对应数据。数据的键包括：body_part_tracking、extracted_features与fiber_photometry。
其中：
- data["body_part_tracking"]：以身体部位为键的字典，存储各部位随时间变化的x、y坐标（共T帧，采样率120 Hz）
- data["extracted_features"]：12×T的特征轨迹矩阵，采样率120 Hz
- data["fiber_photometry"]：以'fpts'或'fpnac'为键的字典，分别对应尾状纹状体（tail striatum）和伏隔核（nucleus accumbens）的原始GCaMP记录，采样率1017 Hz

模型Pickle文件：
- 两组随机森林分类器（包含已训练的随机森林模型、行为发生概率截断阈值）：以行为名称为键的字典，每个键对应（已训练的sklearn.ensemble.RandomForestClassifier模型、分类概率截断阈值）
- t-SNE模型：为包含以下键的字典：
- "training"：训练集特征
- "projection"：将原始特征映射至t-SNE空间的sklearn.neural_network.MLPClassifier模型
- "labeled_map"：t-SNE空间中的簇标注位置