How to Use ZenML: Simple Steps to Build ML Pipelines

To use ZenML, first install it with pip install zenml, then create a pipeline by defining steps as Python functions decorated with @step. Finally, assemble these steps into a pipeline using @pipeline and run it with pipeline.run().

Syntax

ZenML uses simple Python decorators to define steps and pipelines. Each @step marks a function as a pipeline step, and @pipeline groups these steps into a workflow. You run the pipeline by calling pipeline.run().

• @step: Decorates a function to become a pipeline step.
• @pipeline: Decorates a function that connects steps.
• pipeline.run(): Executes the pipeline.

python

from zenml import step, pipeline

@step
def say_hello() -> str:
    return "Hello, ZenML!"

@pipeline
def hello_pipeline():
    greeting = say_hello()
    print(greeting)

if __name__ == "__main__":
    hello_pipeline().run()

Output

Hello, ZenML!

Example

This example shows a simple ZenML pipeline with two steps: one generates data, and the other processes it. It demonstrates how to define steps, create a pipeline, and run it.

python

from zenml import step, pipeline

@step
def generate_numbers() -> list[int]:
    return [1, 2, 3, 4, 5]

@step
def sum_numbers(numbers: list[int]) -> int:
    return sum(numbers)

@pipeline
def sum_pipeline():
    numbers = generate_numbers()
    total = sum_numbers(numbers)
    print(f"Sum is: {total}")

if __name__ == "__main__":
    sum_pipeline().run()

Output

Sum is: 15

Common Pitfalls

Common mistakes when using ZenML include:

• Not decorating functions with @step, so they won't be recognized as pipeline steps.
• Forgetting to decorate the pipeline function with @pipeline.
• Trying to print inside steps instead of returning values for better pipeline management.
• Running pipeline functions without calling .run().

Always return data from steps and use print in the pipeline or after running it.

python

from zenml import step, pipeline

# Wrong: missing @step decorator
# def generate_data():
#     return [1, 2, 3]

@step
def generate_data():
    return [1, 2, 3]

# Wrong: missing @pipeline decorator
# def my_pipeline():
#     data = generate_data()
#     print(data)

@pipeline
def my_pipeline():
    data = generate_data()
    print(data)

if __name__ == "__main__":
    my_pipeline().run()

Output

[1, 2, 3]

Quick Reference

Concept	Description	Example
@step	Marks a function as a pipeline step	@step\ndef step_func(): pass
@pipeline	Defines a pipeline connecting steps	@pipeline\ndef my_pipeline(): pass
pipeline.run()	Runs the defined pipeline	my_pipeline().run()
Return values	Steps should return data, not print	return data
Step inputs	Steps can accept inputs from other steps	def step_func(input_data):

Key Takeaways

Install ZenML and use @step to define pipeline steps.

Group steps with @pipeline and run with pipeline.run().

Always return data from steps instead of printing inside them.

Decorators @step and @pipeline are required for ZenML to recognize functions.

Use pipeline.run() to execute your ML workflow.