Lesson 1: Thinking about constraints

Constraints are the rules that govern the assignment of shifts to employees. Over the course of this lesson series, you will learn how to think about converting scheduling rules into Timefold constraints.

Think of a person whose job it is to create employee schedules. Let’s call them "a planner". When assigning shifts to employees, a planner must think of all the rules that the shifts must follow in order for the schedule to be valid. For example, a planner might consider the following:

All of these rules, and many more, are referred to as constraints in a Timefold schedule. Most constraints must be activated by including the related JSON in the input dataset. The Timefold model considers all activated constraints for all employees at once when assigning shifts in a schedule.

The constraints are split into categories based on the type of rule they represent.

Hard constraints are rules that cannot be broken. There are many reasons why this could be the case. Perhaps the rule is a legal requirement, or perhaps it is a company policy. Regardless of the reason, hard constraints are rules that must be followed in order for a schedule to be valid.

It is not possible to adjust the weight of the contribution of a hard constraint to the hard score. If any hard constraint is violated by any amount, the schedule is considered invalid. The Timefold model will never assign a shift that breaks a hard constraint by any amount, so it does not matter if that amount is -1 or -1000.

It is, however, possible to have schedules which have a negative hard constraint score. This can only happen in the case where shifts have been pre-assigned to employees in the input dataset, and they have been "pinned" in place so that the Timefold model cannot move them, and those assignments violate hard constraints. In this case, since the Timefold model cannot move the offending shifts, there will be a hard score in the final schedule.

If the offending shifts are not pinned in place, then the Timefold model will attempt to improve the hard score as much as possible by un-assigning or moving them. The Timefold model will never assign, or keep assigned, a shift that breaks a hard constraint by any amount.

Soft constraints are preferences, which can be broken. If a soft constraint is violated, the schedule is still valid. However, these types of constraints can be important for employee retention and reducing the business overhead to hire and train new employees. The Timefold model will try its best to satisfy as many soft constraints as possible, but it is not guaranteed to do so.

It is possible to adjust the weight that a soft constraint contributes to the soft score. For example, it could be that an employee would prefer to work no more than 25 hours per week, but they love to work in a certain department and will take more hours there if they are available. Therefore, the employee’s preference for working in department A is more important than their preference for working 25 hours. In this case, the weight of the soft constraint for working in department A should be set higher than the weight of the soft constraint for working no more than 25 hours per week.

It is possible for a soft score to be positive. Some soft constraints penalize undesirable behavior, such as working in a department that an employee does not like. This constraint will contribute a negative (undesirable) amount to the soft score if the employee works a shift in that department. Other soft constraints reward desirable behavior, such as working with a supervisor that an employee prefers. This constraint will contribute a positive (desirable) amount to the score if the employee works at the same time as that supervisor. If the schedule satisfies more desirable behavior than undesirable behavior, the soft score will be positive.

There is only one medium constraint active at any one time. By default, this constraint tells the Timefold model to assign as many mandatory shifts as possible. Unlike employee-specific constraints such as limits to working hours, the constraint to assign as many mandatory shifts as possible is always on by default and does not need any special configuration.

It is also possible to override the default behavior of assigning all shifts, and instead assign shifts to follow a demand curve. This is a more advanced topic and will be covered in a later lesson. You can read more about assigning shifts to a demand curve in the Demand-based scheduling technical documentation.

Medium constraints are treated individually in their own level, because they are less important than hard constraints but more important than soft constraints.

Think of a planner assigning shifts to employees manually. They will never assign shifts which break legal limits or rules, but they will assign shifts which break preferences because sometimes it is necessary to fill all the mandatory shifts. The medium constraint is treated in the same way by the Timefold model. It will try to assign as many mandatory shifts as possible at the expense of preferences, but it will not do so at the expense of violating hard constraints.

By default, each mandatory shift in the input dataset which remains unassigned contributes -1 to the total medium score. As in the case of the hard and soft constraints, a negative contribution is undesirable (we do not want unassigned shifts in the dataset if possible). It is possible to rank shifts by priority and make them contribute different amounts to the medium score when they are unassigned. This will be covered in a later lesson.

In this lesson, we have learned about: