引導式測驗 · 中等 等級

用于训练可视化测验的 TensorBoard

使用 TensorBoard 標量、圖像、直方圖和配置文件,同時保持步驟、運行比較、隱私和評估定義清晰。

相關引導路徑Tensorboard 训练可视化
問題 1 的 8

What does TensorBoard display from a training program?

預覽每個問題8 問題

當您參加測驗時,就會出現答案和解釋。

  1. What does TensorBoard display from a training program?
    • It reconstructs missing metrics from the model architecture
    • It automatically logs every intermediate tensor without instrumentation
    • Logged summaries and traces emitted by training or profiling code
    • It reruns training from prior event files
  2. Why use a consistent global step for scalar logs?
    • To make each run take the same wall-clock time
    • To align measurements across training runs
    • To force each experiment to log every batch
    • To make the underlying metrics identical
  3. What can image summaries help diagnose?
    • Whether predictions are calibrated
    • Whether every class is balanced
    • Whether the evaluation split is unbiased by itself
    • Input and output preprocessing or visual error patterns
  4. What does a histogram of weights or activations show?
    • A distribution of logged tensor values
    • The sequence order of training examples
    • A guarantee that learned features are correct
    • The calibrated probabilities of every class
  5. How can scalar smoothing mislead debugging?
    • It changes the model weights
    • It removes the raw event file
    • It can hide short-lived spikes or changes
    • It corrects mislabeled data
  6. Why should run comparisons include configuration and metric definitions?
    • TensorBoard infers every hidden parameter
    • Different settings or definitions can make curves incomparable
    • It makes all runs use one dataset
    • It removes the need to log steps
  7. What risk arises when a dashboard logs raw user images or text?
    • Event files automatically strip identifiers before sharing
    • TensorBoard never stores image or text summaries
    • Only run names are exposed when an event directory is shared
    • Event files may expose sensitive content when shared
  8. What does TensorBoard not automatically provide by itself?
    • Complete experiment lineage such as data version and code revision
    • Scalar plots
    • Logged images
    • Run names