Skip to content

1. Creating Transform Graphs

1.1. Concept

A Transform Graph

  • represents a pipeline or graph of tasks that need to be executed from start to finish, with any number of intermediate tasks, and with any number of connections (flow of data from one task to another) between the tasks
  • For Machine Learning and Deep Learning workloads, the Transform Graph can be used for preprocessing data, training models, inferencing using trained models and other complex machine learning pipelines

1.2. Creating a Transform Graph

The video below shows the steps to create a simple Transform Graph. This transform graph has 2 nodes (or transforms): each node simply copies its input to its output

!!! tip Pre-requisite: Create Transforms as described in Create Transforms.

A transform graph is created by

  • defining tasks (or nodes) in the graph: Each task represents a unit of processing, which is defined using a Transform
  • each task receives its input
    • from another task in the graph or pipeline: achieved by connecting one task to another in the graph. the connection represents the flow of data from the output of one task to the input of another.
    • a task can also have its input from Infinsnap, Infinslice or an mlflow run's output.