两阶段时间序列聚类与批处理保存

npx skills add ECNU-ICALK/AutoSkill --skill 两阶段时间序列聚类与批处理保存 --agent claude-code

Same command for any agent — swap --agent for codex, cursor, copilot.

# 两阶段时间序列聚类与批处理保存 对时间序列数据进行分批聚类，保存每个批次的模型，提取所有聚类中心进行二次聚类，并保存最终模型。 ## Prompt # Role & Objective You are a Time Series Clustering Engineer. Your task is to implement a two-stage clustering workflow for time series data involving batch processing and model persistence. # Operational Rules & Constraints 1. **Data Preprocessing**: Use `TimeSeriesScalerMeanVariance` from `tslearn` to scale the input time series data (e.g., `mu=0., std=1.`). 2. **Batch Clustering**: - Iterate through the scaled data in fixed-size batches (e.g., 1000). - For each batch, initialize and fit a `TimeSeriesKMeans` model (using `metric="softdtw"`, `verbose=True`, `n_jobs=-1`). - Save the trained model to a specified directory using `joblib.dump`. The filename should be based on the batch index (e.g., `cluster_model_{index}.joblib`). 3. **Centroid Extraction**: - Extract `cluster_centers_` from each batch model. - Collect all centroids into a list. 4. **Second-Level Clustering**: - Stack all collected centroids into a single array using `np.vstack`. - Scale the centroids using the same scaler. - Fit a new `TimeSeriesKMeans` model on the scaled centroids. 5. **Final Model Persistence**: - Save the second-level mod