Auto Shift Planner

Why creating schedules can give us headaches:

This application originated from a request of an employee to switch from a full time job to part time in a branch with continuous hours of a retail shop with a certain regulation of personnel. The question was whether it would still be possible to fulfill a schedule that is compatible with regulation constraints. Several employee shift management applications, while offering valuable additional features, were not able to address the feasibility of that request. Many of the shortcomings that emerged mainly concerned the automated part of such software: the ability to self-generate shift rosters without human intervention. In fact, the request proved to be complex.

• A typical weekly employee schedule for a retail shop open from 8:30 AM to 22:30 PM with nine employees has:

2 possible states for a half hour time unit (employee present or not)… to the power of… 9 employees × 7 days per week × 14 hours per day × 2 in half hours

…which means 1,039710311×10⁵³¹ possible solutions. This is a lot more than the minimal number of atoms in the known universe (1×10⁸⁰).

• It is impossible to evaluate every possible solution even if we compute with the fastest machine for billion of years. This is why we jokingly say that creating schedules can give us headaches and more seriously why traditional planners do not even try to really automatically make a schedule.

There are situations in which no traditional planner can automatically address a complex problem, no spreadsheet, in fact no one can. Auto Shift Planner will try to give you a possible solution if it exists.

• Auto Shift Planner version 0.1.0+:
  • Since for a work shift existence, quantity, starting and ending time (etc.), an abstraction with greater detail instead of just morning or night shift of predefined type is needed, the application uses, as planning entity, “half hour time unit for each employee” objects (we call them cells), to which Optaplanner assigns a property value from a pair of Boolean as planning variable (employee is present - true or not - false). Although this domain model (a binary variable anti-pattern) is easy to understand and the graphical visualization of the solving process can be instructive, it is not efficient.
• Auto Shift Planner version 0.2.0+:
  • TimeGrain Pattern has been implemented. We have a shift assignment as planning entity to which Optaplanner assigns two planning variables: a time grain (when the shift starts) and a duration (how long it will last). Since we don’t know how many shifts an employee will have we over provision them and make them nullable. It’s more scalable (the search space is a lot smaller) and it’s easier to navigate to feasible regions with generic moves.

Further information on this topic can be found on the Optaplanner website. This application uses Optaplanner, a professional grade constraint satisfaction solver, as a planning engine.

back to Main page