重點:code,software 和 OS version
- 理念的轉(zhuǎn)變:從研究本身的質(zhì)量向研究過程的質(zhì)量轉(zhuǎn)變
- the economic literature focuses on the related problem of transparency and selection bias in methods and results in academic journals
關(guān)于組織工作(organized your work)
One mandatory principle for achieving reproducible research is thus to organize the whole process and, specifically, to organize all the tasks needed and involved in the process leading to publication. These ingredients need to be properly organized if the pie is to be cooked again. It should be precisely known at
which steps of the recipe
(phase and task of the project)掠归,which ingredients
(e.g., data, methods) andwhat recipe are used
(e.g., codes, documentation) andwhat are the interactions and relationships between each element to the resulting pie
(e.g., project results). This process involves addressing related topics: task and documentation writing, file organization, workflow management and file manipulation. Many organizational forms can be considered: Some are relevant for individual research projects, while others are better suited for projects involving many researchers or a team.
翻譯:實現(xiàn)可重復(fù)性研究的一個強制性原則是:必須時刻通盤考慮整個過程硝枉,特別是組織好所有需要和參與的過程,直到最終出版来屠。如果餡餅要重新烹調(diào)燕雁,這些原料就需要適當(dāng)?shù)亟M織起來岖妄。它應(yīng)該確切地知道每種配方所在的步驟
(項目的階段和任務(wù))宫仗,不同配方需要不同的成分
(例如挤庇,數(shù)據(jù)钞速,方法)和使用什么樣的配方
(例如,代碼嫡秕,文檔)導(dǎo)致的各種成分相互作用后所產(chǎn)生可能的不同的結(jié)果
(例如渴语,項目結(jié)果)。這個過程涉及到以下主題:組織任務(wù)和文檔編寫昆咽、文件夾管理驾凶、時刻關(guān)注整體工作流程與進(jìn)度和文件操作。許多組織形式可以被考慮:一些是與個人研究項目相關(guān)的掷酗,而另一些則更適合于涉及許團(tuán)隊的項目
step1:Plan
step2:Organize
step3:Compute
step4:Document
組織任務(wù)與文檔編寫 Organizing tasks and documentation
從粘貼到任務(wù)管理系統(tǒng)
從評論到任務(wù)文檔
Information that cannot be explained within the task (documents relating to a specific ingredient such as the code) should also be documented at the task level: general choices about the project (hypothesis and decisions such as the type of modelization, the population under study, and abandoned tested directions) and technical specifications that can have an impact on the results.在任務(wù)中無法解釋的信息(與特定的成分有關(guān)的文件调违,如代碼)也應(yīng)在任務(wù)級別上記錄,包括:關(guān)于項目的一般選擇(假設(shè)和決定汇在,如:模型化的類型翰萨、參與研究的人員、和已經(jīng)放棄的研究方向糕殉,避免重復(fù))以及技術(shù)規(guī)范會對結(jié)果產(chǎn)生影響亩鬼。例如殖告,可以包括如下:
包含/排除觀測值的標(biāo)準(zhǔn);為最優(yōu)化選擇的隨機化方法雳锋、隨機種子黄绩、初始值和參數(shù)穩(wěn)健性的檢驗;呈現(xiàn)結(jié)果所使用的算法(插值玷过、平滑等)
文件夾管理 Organizing file
集中在:目錄結(jié)構(gòu)和命名約定(directory structure and naming convention)
- 目錄結(jié)構(gòu):
首先爽丹,在建構(gòu)每個項目的目錄結(jié)構(gòu)時候,必須保證是一致的(consistency)辛蚊。其次粤蝎,對項目的輸入和項目的產(chǎn)出間(input and output),必須清楚劃分袋马。
- 文檔命名:
修正初澎、改進(jìn)、增添虑凛、評論碑宴、新想法—————每個研究者的日常。
時刻關(guān)注整體工作流程與進(jìn)度 Keeping track of the workflow
- 圖形的方式
- 文件夾命名的方式
- 個人采用
處理文件夾 Handling files
- 文檔分享:Google Drive桑谍、OneDrive
- 版本控制:Github延柠、Gitee
- 寫作環(huán)境:Latex、WS
- 文獻(xiàn)管理:NoteExpress(中文)锣披、Endnote(英文)
為他人編程 Code for others (including your future self)
- 使用通用的編程方式
-
記錄代碼
代碼記錄
盡量自動化 Automate as much as you can
-
可以自動生成文件的代碼
能夠生成文件的代碼 -
BAT文件做獨立分區(qū)塊運行——本人還不懂贞间!
Batch 或者 Makefile
區(qū)塊化能夠保持獨立性 -
跨語言支持
推薦:Jupyter Notebool
Figer-201807234
一些有價值的參考
-
STATA圖表與代碼
STATA圖表與代碼
STATA圖表與代碼 -
可參考文獻(xiàn)及代碼來源
可參考文獻(xiàn)及代碼來源 -
軟件細(xì)節(jié)
軟件細(xì)節(jié)