DataFrame 在資料科學中經常被用來處理表狀資料的概念,Python 的 pandas 函數庫是一個廣被實用的 DataFrame 實作。而因為受限於複 雜的 API 設計與缺乏靜態工具,使用者在編寫 pandas 程式時往往是容易 犯錯的,這些錯誤來自於:沒有追蹤行欄位的標籤與型別、透過字串傳 遞的旗標參數與缺乏函數參數的型別資訊。 本研究討論了為 pandas 提供靜態分析功能的難處,並提出了一個用 於靜態分析 pandas 程式的靜態型別檢查器:我們提出的作法基於使用替 代語意,並且我們討論了如何為常用的 API 進行靜態檢查。並且我們使 用 Python 完成了一個初步實作以作為概念驗證。我們透過定量分析的 方式評估我們初步實作的功能性,並以使用案例來討論我們的工具能如 何被使用。;Dataframe is a well-used concept in Data Science tasks, which makes abstract on how a programmer manipulates tabular data. pandas is a popular and widely-used Python library which implements dataframes. Due to the complexity of its API design and lacking static analysis tools, programming in pandas is considerably an error-prone task. There are three types of common error: errors due to lacking column labels and types, errors due to string-typed flag arguments, and errors due to lacking type informations of functions as arguments. In this paper, we discuss what is the challenges of providing the ability of static analysis on pandas programs. We propose a static type checker for pandas programs based on Alternative Semantic. We also dis- cuss how to statically analyze common pandas APIs. We develop a preliminary implementation based on Python as our proof of concept. We evaluate our implementation with a qualitative analysis on its functionality and we discuss a case study about how can our checker reduce error‐proneness during development.
