Transform Widget

The Transform Widget is a powerful tool for processing data using the TransformsV2 library. It allows you to build a sequence of operations (a pipeline) to manipulate your data, convert types, and extract features.

Overview

The widget provides a visual interface to:

  1. Select Input Data: Choose the data you want to transform.
  2. Define a Pipeline: Add, remove, and reorder transformation steps.
  3. Configure Parameters: Adjust settings for each step.
  4. Execute: Run the pipeline and save the result.

Using the Widget

1. Selecting Input

Use the Input Selector at the top of the widget to choose a data item from your project.

  • When you select data in the Data Manager or other widgets, the Transform Widget will automatically update to focus on that item.
  • The valid transformation steps available to you depend on the type of data you have selected.

2. Building a Pipeline

The pipeline is a list of steps that run in order.

  • Add Step: Click the + button or drag a transform from the library (if available) into the step list.
  • Remove Step: Select a step and click the - button.
  • Reorder: Drag and drop steps to change their execution order.

Each step transforms the data from the previous step. The widget ensures that the output of one step is compatible with the input of the next.

3. Pre-Reductions (Advanced)

Some transformations require global statistics about the data before they can run. For example, a Z-Score Normalization needs the Mean and Standard Deviation of the entire dataset.

  • Use the Pre-Reduction Panel to calculate these values.
  • These values are stored in a temporary “Value Store” and can be bound to the parameters of your transformation steps.

4. Step Configuration

When you select a step in the list, its parameters appear in the Configuration Panel.

  • Parameters can be simple values (numbers, text) or bindings to Pre-Reduction values.
  • Adjust these settings to customize the transformation.

5. Execution

Once your pipeline is valid:

  1. Output Name: Enter a name for the result. The widget may suggest one for you.
  2. Execute: Click the Run button.
  3. The result will be created and added to the Data Manager.

Integration with Tensor Inspector

The Transform Widget’s pipeline definition is portable. You can design a pipeline here and use it in the Tensor Inspector to populate columns of a tensor.