Creating advanced custom constraints with Python

This Decision Optimization Modeling Assistant example shows you how to create advanced custom constraints that use Python.

Before you begin

Open any Decision Optimization model in the Decision Optimization Modeling Assistant. This example uses Shift Assignment sample that is available in the DO-samples.

About this task

The Modeling Assistant provides you with many constraint suggestions for your problem domain which can be customized. You might, however, want to express constraints beyond those that are predefined for the given domains. You can achieve this by using more advanced custom constraints that use Python DOcplex. This example illustrates how you can create these.

This video provides a visual method as an alternative to following the written steps in this documentation. Some minor steps and graphical steps in this video might differ as the user interface is frequently improved. After you load the example in your Decision Optimization experiment you can follow the video.

Note: To create and run Optimization models you must have both a Machine Learning service added to your project and a deployment space associated with your experiment:
  1. Add a Machine Learning service to your project. You can either do this at the project level (see Creating a Watson Machine Learning Service instance), or this can be done when you first create a new Decision Optimization experiment: click Add a service, select or create a New service, click Associate service in the item bar, then close the pane.
  2. Associate a deployment space with your Decision Optimization experiment (see Deployment spaces). A deployment space can be created or selected when you first create a new Decision Optimization experiment: click Create a deployment space, enter a name for your deployment space and click Create. For existing models, you can also create or select a space in the Overview information pane.

Procedure

  1. In the Build model view of your open Modeling Assistant model, look at the Suggestions pane. If you have Display by category selected, expand the Others section to locate New custom constraint, and click it to add it to your model. Alternatively, without categories displayed, you can enter, for example, custom in the search field to find the same suggestion and click it to add it to your model.
    A new custom constraint is added to your model.

    New custom constraint in model, with elements highlighted to be completed by user.

  2. Click Enter your constraint. Use [brackets] for data, concepts, variables, or parameters and enter the constraint you want to specify. For example, type No [employees] has [onCallDuties] for more than [2] consecutive days and press enter.
    The specification is displayed with default parameters (parameter1, parameter2, parameter3) for you to customize. These parameters will be passed to the Python function that implements this custom rule.

    Custom constraint expanded to show default parameters and function name.

  3. Edit the default parameters in the specification to give them more meaningful names. For example, change the parameters to employees, on_call_duties, and limit and click enter.
  4. Click function name and enter a name for the function. For example, type limitConsecutiveAssignments and click enter.
    Your function name is added and an Edit Python button appears.

    Custom rule showing customized parameters and Edit Python button.

  5. Click the Edit Python button.
    A new window opens showing you Python code that you can edit to implement your custom rule. You can see your customized parameters in the code as follows:

    Python code showing block to be customized

    Notice that the code is documented with corresponding data frames and table column names as you have defined in the custom rule. The limit is not documented as this is a numerical value.
  6. Optional: You can edit the Python code directly in this window, but you might find it useful to edit and debug your code in a notebook before using it here. In this case, close this window for now and in the Scenario pane, expand the three vertical dots and select Generate a notebook for this scenario that contains the custom rule. Enter a name for this notebook.
    The notebook is created in your project assets ready for you to edit and debug. Once you have edited, run and debugged it you can copy the code for your custom function back into this Edit Python window in the Modeling Assistant.
  7. Edit the Python code in the Modeling Assistant custom rule Edit Python window.
    For example, you can define the rule for consecutive days in Python as follows:
        def limitConsecutiveAssignments(self, mdl, employees, on_call_duties, limit):
            global helper_add_labeled_cplex_constraint, helper_get_index_names_for_type, helper_get_column_name_for_property
            print('Adding constraints for the custom rule')
            for employee, duties in employees.associated(on_call_duties):
                duties_day_idx = duties.join(Day)  # Retrieve Day index from Day label
                for d in Day['index']:
                    end = d + limit + 1  # One must enforce that there are no occurence of (limit + 1) working consecutive days
                    duties_in_win = duties_day_idx[((duties_day_idx['index'] >= d) & (duties_day_idx['index'] <= end)) | (duties_day_idx['index'] <= end - 7)]
                    mdl.add_constraint(mdl.sum(duties_in_win.onCallDutyVar) <= limit)
  8. Click the Run button to run your model with your custom constraint.
    When the run is completed you can see the results in the Explore solution view.