Open multi_step_pipeline.ipynb and follow the instructions in the notebook.
❓ Question: How can we make our pipeline accept parameters like Datasets, numbers, or strings?
✅ See solution!
We can use the PipelineParameter class. In this case, we always have to specify a default value, for the case that the caller does not specify a value. Here is a shot example:
default_training_dataset = Dataset.get_by_name(ws, "my_dataset")
training_dataset_parameter = PipelineParameter(name="training_dataset", default_value=default_training_dataset)❓ Question: What are some pros and cons for separating a pipeline into multiple steps?
✅ See solution!
It depends on the use case, but here are some ideas:
Pros for having multiple steps:
- Step results can be chached via
allow_reuse=True - Better modularity of code possible
- Easier to troubleshoot individual steps
Cons against having multiple steps:
- More overhead to pass data between steps
- Long pipelines with many steps might complicated