0Pricing
MLOps Academy · 课时

跟踪一次预测请求的完整往返

跟踪一个请求从输入到记录响应的全过程

跟踪一次预测请求的完整往返 是 CoddyKit 上的免费 MLOps Academy 课时。 这是第 4 节课，共 4 节。 你可以在下方免费阅读本课时的完整内容 — 然后在浏览器中使用内置代码编辑器和全天候 AI 导师进行实践。 这是 MLOps Academy 学习路径的一部分，你的进度在网页和 CoddyKit 应用中同步。 MLOps Academy 课程共包含 4 节课。

本课时的部分内容尚未翻译，以英文显示。

Follow One Request

Time to watch the whole loop work. You will follow a single prediction from a client request all the way to a logged response. 🔍

The Client Sends Input

A caller POSTs JSON to /predict. This is the raw input feature payload your model needs to make a decision.

curl -X POST localhost:8000/predict \
  -H "Content-Type: application/json" \
  -d '{"tenure": 12, "monthly_charges": 79.9}'

Pydantic Validates

FastAPI parses the body against your schema. If a field is missing or the wrong type, Pydantic rejects it with a 422 before the model runs.

The Model Predicts

The validated row hits the loaded champion model. It computes a result in memory, no retraining and no registry call on the hot path.

pred = app.state.model.predict([[f.tenure, f.monthly_charges]])

Log What Happened

Before responding, write a structured log line with the inputs, the prediction, and a timestamp. This is your audit trail.

import logging, json
logging.info(json.dumps({
    "input": f.model_dump(),
    "prediction": int(pred[0]),
}))

Return the Response

FastAPI serializes your dict back to JSON and sends it with a 200. The client receives the prediction in milliseconds.

{"prediction": 1}

Capture Latency

Wrap the call to measure how long it took. Logging latency per request is what later powers your monitoring dashboards.

import time
start = time.perf_counter()
# predict...
elapsed_ms = (time.perf_counter() - start) * 1000

Tie Back to the Version

Log which model produced the answer. Recording the version means you can always say exactly which model made any past prediction.

{"model": "churn-classifier", "version": 3}

The Loop Closed

Train, register, promote, serve, predict, log: one request just walked the entire pipeline you built. That is end-to-end MLOps. 🎯

Quick Check

Why log the input, prediction, and model version on every request?

Logs Feed Monitoring

Those logged round-trips become the raw material for the next stage: monitoring latency, error rates, and eventually input drift over time.

You Built a System

This is no longer a notebook. It is a traceable system where every prediction can be explained and reproduced from end to end.

Recap: The Full Round-Trip

You traced one request through validation, prediction, logging, and response, tying it back to a model version. That closes your first end-to-end flow. ✅

常见问题解答

「跟踪一次预测请求的完整往返」课时是免费的吗？

是的 — 「跟踪一次预测请求的完整往返」的完整文本可在网页上免费阅读。要进行交互式练习（内置代码编辑器和全天候 AI 导师）并解锁 MLOps Academy 课程的其余内容，请升级到 CoddyKit PRO。 MLOps Academy 课程共包含 4 节课。

「跟踪一次预测请求的完整往返」这节课中我会学到什么？

跟踪一个请求从输入到记录响应的全过程 你通过在浏览器中直接运行的动手代码来练习 MLOps Academy，全天候 AI 导师会在你学习这节课的过程中回答你的问题。

学习 MLOps Academy 需要有经验吗？

无需任何先前经验。CoddyKit 上的 MLOps Academy 课程适合初学者到高级学习者，你可以从这里开始或从头开始，按照自己的节奏学习。 这是第 4 节课，共 4 节。

「跟踪一次预测请求的完整往返」课时需要多长时间？

大多数 CoddyKit 课程大约需要 5–10 分钟。每节课都很精短且互动，所以你能稳步进步，并在网页和应用中从离开的地方继续。

我能在这节 MLOps Academy 课中编写并运行代码吗？

能。每节 MLOps Academy 课都包含内置代码编辑器，你可以在浏览器中直接编写并运行真实代码，并获得即时 AI 反馈 — 无需本地设置。

此课程中的所有课时

  1. 训练模型并记录到注册表
  2. 将最佳模型推向生产环境
  3. 提供生产模型服务
  4. 跟踪一次预测请求的完整往返
← 返回 MLOps Academy