[Requirement] UI components for selecting, configuring, and running workflows

[SimonHeybrock]

Executive summary

For the "80%" of basic workflow use, we need a simple interface that reduces the need for Python knowledge.

Context and background knowledge

Users need to create and configure workflows. In many cases the workflow are pre-built in the ESS* packages and need configuration. Simple GUI tools based on (probably?) ipywidgets are needed for all ESS* packages. These GUI components will, step by step (e.g., in a series of tabs), allow for:

1. Select a workflow. Assume that we can get a list of workflows from the package, i.e., as input. Each workflow will come with a description (e.g., the docstring) which should be displayed.
2. Set workflow params in widgets.
3. Optionally, define tables of params.
4. Select targets for computation, could be a single final result, or multiple, including intermediate results.
5. Run the workflow.
6. Plot or save outputs.

This requirement is a high-level overview, individual parts will be described in more detail in sub-requirements.

Have a look at, e.g., https://docs.mantidproject.org/v6.1.0/algorithms/LoadEventNexus-v1.html. The widget and the table of properties in the docs page was generated programmatically. This corresponds to step 2.) above.

Inputs

• A package can provide a list of workflows.
• Input params can be obtained from the workflow. There will be a mapping between input keys of the workflow and implementations of a Param class. The latter includes things such as a description (to be shown in the GUI), bounds, validators, ...
  • Need a way of marking parameters as advanced, so they can be hidden in default view of widgets.
• Targets could be any key in the pipeline, but likely we will want some customization, i.e., the workflow may specify this.

Methodology

Python function in a helper module that can be used to auto-generate a widget.

Outputs

A widget that can be used in a Jupyter notebook. Depending on where we stop in the listing, this might return a configured pipeline, a task graph, results, or even run, plot, and save directly.

Which interfaces are required?

Python module / function

Test cases

ESSsans is a good candidate for prototyping and trying this out

Comments

No response

[SimonHeybrock]

Widget team initial goal:

• What is needed from Python package beyond Sequence[Workflow]?
• What does Workflow need beyond Sequence[Parameter] to generate widgets?
• Find out if draft of Parameter class in branch param-class has all necessary info.

Parameter/workflow team initial goal:

• How to interface parameters with pipeline. How do we get or define the Sequence[Parameter]?

See https://github.com/scipp/esssans/blob/a614f49492453cf6ea0f1615e63898b77b7303f6/src/ess/loki/general.py#L60-L84 for an example of a current workflow, which also defines default params.

[YooSunYoung]

• What is needed from Python package beyond Sequence[Workflow]?

• Interesting targets, things to be computed - we tried to detect all the output leaves, but then there are providers that are not used, i.e. SamplePosition[EmptyRun] due to how generic provider works.
• Root leaves, things to be set as parameters - more precisely, a handle that widget object can access all the parameters to be set.

Description helpers

• New Types docstrings.

[SimonHeybrock]

Whiteboards from today: