[[["容易理解","easyToUnderstand","thumb-up"],["確實解決了我的問題","solvedMyProblem","thumb-up"],["其他","otherUp","thumb-up"]],[["缺少我需要的資訊","missingTheInformationINeed","thumb-down"],["過於複雜/步驟過多","tooComplicatedTooManySteps","thumb-down"],["過時","outOfDate","thumb-down"],["翻譯問題","translationIssue","thumb-down"],["示例/程式碼問題","samplesCodeIssue","thumb-down"],["其他","otherDown","thumb-down"]],["上次更新時間：2025-07-27 (世界標準時間)。"],[[["\u003cp\u003eML solutions are developed in iterative phases: ideation and planning, experimentation, pipeline building, and productionization.\u003c/p\u003e\n"],["\u003cp\u003eEach phase has specific goals, tasks, and outcomes that contribute to the overall success of the project.\u003c/p\u003e\n"],["\u003cp\u003eIdentifying the right problem for an ML solution and ensuring its feasibility is crucial in the initial phase.\u003c/p\u003e\n"],["\u003cp\u003eRigorous experimentation is essential for building an effective ML model and may involve numerous iterations.\u003c/p\u003e\n"],["\u003cp\u003eProductionizing an ML solution requires robust pipelines for data processing, model training, serving predictions, and ongoing monitoring.\u003c/p\u003e\n"]]],[],null,["# ML development phases\n\nML projects progress in phases with specific\ngoals, tasks, and outcomes. A clear understanding of the ML development\nphases helps to establish engineering\nresponsibilities, manage stakeholder expectations, and efficiently allocate\nresources.\n\nSuccessfully moving through the phases (often iteratively) is foundational\nfor designing, assembling, and building ML\n[models](/machine-learning/glossary#model) that solve business problems\nover the long term.\n\nAt a high level, implementing an ML solution consists of the following phases:\n\n1. Ideation and planning\n2. Experimentation\n3. Pipeline building\n4. Productionization\n\nIdeation and planning\n---------------------\n\nDuring the ideation and planning phase, you frame your problem in terms of an\nML solution and estimate the project's feasibility.\n\n- **Goal**: To determine if ML is the best solution to your problem.\n- **Tasks**: Analyze the business problem to understand your project's constraints.\n- **Outcome**: A design doc outlining how to solve a problem with a ML solution.\n\n**Important:** During the ideation and planning stage, it can take a long time to understand the data and metrics required for a production-ready system. \n\nExperimentation\n---------------\n\nExperimentation is the core of machine learning. During this phase,\nyou verify that an ML solution is viable. Finding a solution is an\niterative process. It's not uncommon to try hundreds of experiments\nbefore finding the right combination of\n[features](/machine-learning/glossary#feature),\n[hyperparameters](/machine-learning/glossary#hyperparameter),\nand model architecture that solves the problem.\n\n- **Goal**: To build a model that solves the business problem.\n- **Tasks**: Experiment with features, hyperparameters, and model architectures.\n- **Outcome**: A model with good enough quality to put into production.\n\n**Important:** During the experimentation phase, new-to-ML practitioners often underestimate the challenges of designing and implementing the appropriate experimentation tooling and processes. \n\nPipeline building and productionization\n---------------------------------------\n\nDuring the [pipeline](/machine-learning/glossary#pipeline) building and\nproductionization phase, you build pipelines\nfor processing data, training a model, and serving\n[predictions](/machine-learning/glossary#prediction). You then\ndeploy the model and pipelines into production with the necessary monitoring and\nlogging infrastructure.\n\n- **Goal**: To build and implement the infrastructure for scaling, monitoring, and maintaining models in production.\n- **Tasks**: Build pipelines to automate many of the tasks for keeping up-to-date models in production.\n- **Outcomes**: Validated ML pipelines.\n\n**Important:** During the productionization phase, it's easy to underestimate the complexity of productionizing data pipelines and evaluating models, especially as features evolve. For instance, not only do you have to implement all the monitoring infrastructure required for a non-ML project, but also all the ML-specific monitoring. \n\nEnd-to-end ML workflow\n----------------------\n\nThe following diagram illustrates the entire end-to-end ML workflow, listing\neach phase and its tasks and outcomes:\n\n**Figure 1**. The four main phases of an ML workflow.\n\nKeep in mind\n------------\n\nMultiple challenges exist at each phase.\nNot realizing---and planning for---them may lead to missed\ndeadlines, frustrated engineers, and failed projects.\n\n### Check Your Understanding\n\nYou've just read about some ML technology that might benefit your product. What should you do next? \nBefore spending time drafting a design doc or writing code, you should first verify that ML is the right solution to your problem. \nCorrect. Before spending time drafting a design doc or writing code, you should first verify that ML is the right solution to your problem. \nDraft a design doc outlining the ML use case and the required infrastructure to implement it. \nBefore drafting a design doc, you should first verify that ML is the right solution to your problem. \nFind code examples and begin experimenting to determine if the model can make good predictions. \nBefore writing a line of code, you should first verify that ML is the right solution to your problem.\n| **Key Terms:**\n|\n| |-------------------------------------------------------|---------------------------------------------------------------|\n| | - [feature](/machine-learning/glossary#feature) | - [hyperparameter](/machine-learning/glossary#hyperparameter) |\n| | - [model](/machine-learning/glossary#model) | - [pipeline](/machine-learning/glossary#pipeline) |\n| | - [prediction](/machine-learning/glossary#prediction) |\n|"]]
