相關人員
透過集合功能整理內容
你可以依據偏好儲存及分類內容。
機器學習專案有多位利害關係人,參與程度和責任各不相同。及早與利害關係人合作,並進行有效協作,是開發合適解決方案、管理期望,以及最終成功導入 ML 的關鍵。
盡可能提早定義專案的利害關係人、預期交付項目和偏好的溝通方式。
請務必將他們納入利害關係人清單,以及任何需要核准 ML 解決方案各個層面的團隊。
交付項目
在專案的每個階段,利害關係人可能期望不同的交付項目。以下列出常見的交付項目。
設計文件:在編寫任何程式碼之前,您很可能會先建立設計文件,說明問題、建議的解決方案、可能的做法和潛在風險。通常,設計文件可做為接收意見回饋的管道,並解決專案利害關係人的問題和疑慮。
實驗結果。您必須傳達實驗階段的結果。您通常會加入下列項目:
- 實驗記錄,包括超參數和指標。
- 訓練堆疊,以及模型在特定檢查點儲存的版本。
可投入正式環境的實作方式。訓練及提供模型的完整管道是主要交付項目。在這個階段,請為日後的工程師建立文件,說明模型決策、部署和監控細節,以及資料特徵。
您應盡早與專案關係人溝通,瞭解他們對專案各階段的期望。
注意事項
在某些情況下,利害關係人可能不瞭解 ML 的複雜性和挑戰。這可能會導致專案難以獲得優先處理和執行。舉例來說,部分利害關係人可能會認為 ML 與傳統軟體工程做法類似,結果具有確定性。他們可能不瞭解專案進度停滯的原因,或專案里程碑為何並非線性。
為管理利害關係人的期望,請務必清楚說明專案各階段的複雜度、時程和交付項目。
除非另有註明,否則本頁面中的內容是採用創用 CC 姓名標示 4.0 授權,程式碼範例則為阿帕契 2.0 授權。詳情請參閱《Google Developers 網站政策》。Java 是 Oracle 和/或其關聯企業的註冊商標。
上次更新時間:2025-07-27 (世界標準時間)。
[null,null,["上次更新時間:2025-07-27 (世界標準時間)。"],[[["\u003cp\u003eMachine learning (ML) projects require early and consistent collaboration with stakeholders who have varying levels of involvement and expectations.\u003c/p\u003e\n"],["\u003cp\u003eClearly define project deliverables like design documents, experimental results, and production-ready implementations, aligning with stakeholder expectations for each project phase.\u003c/p\u003e\n"],["\u003cp\u003eProactively communicate the unique complexities and potential challenges inherent in ML projects to manage stakeholder expectations and ensure project success.\u003c/p\u003e\n"],["\u003cp\u003eEstablish clear communication channels and involve all necessary teams, including those requiring approval, for efficient project execution.\u003c/p\u003e\n"]]],[],null,["# Stakeholders\n\nML projects have multiple stakeholders with varying levels of involvement and\nresponsibilities. Early involvement and effective collaboration with\nstakeholders is essential for developing the right solution, managing\nexpectations, and ultimately for a successful ML implementation.\n\nAs early as possible, define your project's stakeholders, the expected\ndeliverables, and the preferred communication methods.\n\nBe sure to include them in your list of stakeholders, as well as any other\nteams who need to approve aspects of your ML solution.\n\nDeliverables\n------------\n\nEach stakeholder might expect different deliverables at each phase of the\nproject. Here's a list of common deliverables.\n\n- **Design doc.** Before you write a line of code, you'll most likely create a\n design doc that explains the problem, the proposed solution, the potential\n approaches, and possible risks. Typically, the design doc functions as a way\n to receive feedback and address questions and concerns from the project's\n stakeholders.\n\n \u003cbr /\u003e\n\n- **Experimental results.** You must communicate the outcomes from the\n experimentation phase. You'll typically include the following:\n\n - The record of your experiments with their hyperparameters and metrics.\n - The training stack and saved versions of your model at certain checkpoints.\n- **Production-ready implementation.** A full pipeline for training and\n serving your model is the key deliverable. At this phase, create\n documentation for future engineers that explain modeling decisions,\n deployment and monitoring specifics, and data peculiarities.\n\nYou should align early with your stakeholders on their expectations\nfor each phase of the project.\n\n### Keep in mind\n\nIn some cases, stakeholders might not understand the complexities and challenges\nof ML. This can make getting projects prioritized and executed difficult. For\nexample, some stakeholders might assume that ML is similar to traditional\nsoftware engineering practices with deterministic outcomes. They might not\nunderstand why the project's progress is stalled or why a project's milestones\nare non-linear.\n\nTo manage stakeholder expectations, it's critical to be clear about the\ncomplexities, timeframes, and deliverables at each stage of your project."]]