In this lab, you explore the use of the What-If Tool (WIT) for image recognition models. Your task is to predict whether a person is smiling. The lab provides a CNN (Convolutional Neural Network) that is trained on a subset of CelebA dataset and visualizes the results on a separate test subset.

Objectives

In this lab, you will perform the following tasks:

• Launch an AI Platform Notebook.
• Download the pre-trained Keras model.
• Define helper functions for dataset conversion from csv to tf.
• Load the csv file into pandas dataframe and process it for WIT.
• Load the Keras models.
• Define the custom predict function for WIT.

Start a JupyterLab Notebook instance

1. In the GCP Console, in the Search bar, type in Notebook.
2. Select Notebooks for AI Platform.

3. On the Notebook instances page, click New Instance.

4. In the Customize instance menu, select the version of TensorFlow Enterprise 2.1 Without GPUs.

5. In the New notebook instance dialog, for Region, select us-central1, for Zone, select a zone within the selected region, leave all other fields with their default options, and click Create.

After a few minutes, the AI Platform console will display your instance name, followed by Open JupyterLab.

6. Click Open JupyterLab.

Clone the sample code