机器学习世界的研究重点稍微从探索“模型 能够理解什么”，转向了研究“模型是如何理解的”。由于人们担心引入偏差，或者过度泛化模型的适用性，因此开发出了What-If Tool（WIT）这样的工具。这个工具可帮助数据科学家深入研究模型的行为，并将各种功能和数据集对输出的影响进行可视化。WIT由Google引入，简化了比较模型、切片数据集、可视化构面和编辑单个数据点等任务，并可以在Tensorboard 或Jupyter notebooks中使用。尽管WIT可以帮助执行分析，但研究者仍然需要对模型背后的数学 和理论有深刻的理解。WIT只是数据科学家 用来深入了解模型行为的工具，对于使用不当或缺乏训练不佳的算法，初级用户不应奢望有任何工具可以消除或减轻其风险或造成的损害。

在数据中我们对What-If充满了好奇，总是想从上下文关联推断中获得对未来假设的展望。

wit-smile-intro.png

What-If Tool(WIT)为扩展对黑箱分类或回归ML模型的理解提供了一个易于使用的接口。使用该插件，您可以对大量示例执行推理，并立即以各种方式可视化结果。此外，可以手动或以编程方式编辑示例并重新运行模型，以查看更改的结果。它包含用于研究模型性能和数据集子集公平性的工具。

To build the web demos yourself:

  • Binary classifier for UCI Census dataset salary prediction
    • Dataset: UCI Census
    • Task: Predict whether a person earns more or less than $50k based on their census information
    • To build and run the demo from code: bazel run tensorboard/plugins/interactive_inference/tf_interactive_inference_dashboard/demo:demoserver then navigate to http://localhost:6006/tf-interactive-inference-dashboard/demo.html
  • Binary classifier for smile detection in images
    • Dataset: CelebA
    • Task: Predict whether the person in an image is smiling
    • To build and run the demo from code: bazel run tensorboard/plugins/interactive_inference/tf_interactive_inference_dashboard/demo:imagedemoserver then navigate to http://localhost:6006/tf-interactive-inference-dashboard/image_demo.html
  • Multiclass classifier for Iris dataset
    • Dataset: UCI Iris
    • Task: Predict which of three classes of iris flowers that a flower falls into based on 4 measurements of the flower
    • To build and run the demo from code: bazel run tensorboard/plugins/interactive_inference/tf_interactive_inference_dashboard/demo:irisdemoserver then navigate to http://localhost:6006/tf-interactive-inference-dashboard/iris_demo.html
  • Regression model for UCI Census dataset age prediction
    • Dataset: UCI Census
    • Task: Predict the age of a person based on their census information
    • To build and run the demo from code: bazel run tensorboard/plugins/interactive_inference/tf_interactive_inference_dashboard/demo:agedemoserver then navigate to http://localhost:6006/tf-interactive-inference-dashboard/age_demo.html
    • This demo model returns attribution values in addition to predictions (through the use of vanilla gradients) in order to demonstate how the tool can display attribution values from predictions.

https://pair-code.github.io/what-if-tool/index.html 洞悉数据的利器，Google PAIR（People + AI Research initiative (PAIR) ）出品。