Pier-2 v0.2.19

v0.2.19

Note

This is the final release of v0.2 Pier-2 (駁二).
The next version will move forward to v0.3 Huashan (華山).

DataList / DataTable

• Added more missing value imputation methods.

• Added sampling, proportional sampling, shuffling, and train/test split utilities.

• Added categorical encoding tools for machine learning preprocessing:

  • One-hot encoding
  • Label encoding
  • Ordinal encoding

• Added a programmable Describe() API and CLI command for data summarization, supporting:

  • Numerical statistics
  • Categorical statistics
  • Grouped statistics
  • Custom quantiles

• Added reusable feature scalers:

  • StandardScaler
  • MinMaxScaler
  • RobustScaler
  • MaxAbsScaler

  These scalers support fitting on training data and transforming test data with the same parameters, enabling a no-leakage machine learning preprocessing workflow.
  A corresponding stateful CLI scale command is also provided.

stats

• Added Generalized Linear Models (GLM):

  • Logistic Regression
  • Poisson Regression
  • Generic GLM

---

v0.2.19

Note

這是 v0.2 Pier-2(駁二) 的最後一個版本，下個版本將進入 v0.3 Huashan(華山)。

DataList / DataTable

• 新增更多缺失值插補方法。
• 補上抽樣、比例抽樣、洗牌與 train/test split。
• 為機器學習預處理新增分類編碼工具(one-hot, label, ordinal)。
• 新增可程式化的資料摘要 Describe() API 與 CLI 指令，支援數值、類別、分組統計與自訂分位數。
• 新增可重複使用的特徵縮放器（StandardScaler / MinMaxScaler / RobustScaler / MaxAbsScaler），支援在訓練集 fit、用同一組參數轉換測試集的無洩漏（no-leakage）ML 預處理流程，並提供對應的 stateful CLI scale 指令。

stats

• 加入廣義線性模型 (Logistic / Poisson / generic GLM)。