How We Used Biology for Creating an Innovative Automatic Scheduling

Key peculiarities of algorithm

The generic algorithm on its own is not an innovation at all. We were inspired by one of the ideas found on GitHub, so it was not just taken to work, but shaped properly in accordance with the project needs.  

The thing is, the initial algorithm could just work under the standardized conditions, counting the time when the lesson should start and end. So we worked over it a lot, making the algorithm what it is now - a unique solution with diverse functionality, applicable in many fields and niches. 

There are a few of our developers who were working on implementing this algorithm. Backend engineer, Pavel is one of them. And he’s gonna share its features:

“We used a genetic algorithm, being a substring of the evolutionary ones when math from biology is used. So to say, “under the hood” there's a simulation of mutation occurring and chromosome forming. As soon as the simulations are completed, the optimal option is picked, or the one, which is matching the inserted criteria best. 

In alternative solutions, presented in the market, the planning is realized with simple iterations. The final result has a lot of flaws and faults, the number of events for the teachers is not balanced, so the teachers are often overloaded. In our case, the algorithm creates a collection of a couple of schedule options. After that, it picks the best match. It looks like a survival of the fittest in natural selection. A genetic algorithm makes this selection possible 

But there are still a lot of points to rework. Nowadays, the algorithms we use do not include the setting and limitation of the beginning and the end of the classes for both students and teachers. The teachers’ load and so-called “hours for academic development” are not included in the calculation as well".

Snags and difficulties

In the very beginning, we’ve tried a few ordinary decisions, with open-source algorithms for them. But there were hitches with equal distribution of the teachers' and students' workload and the need to generate a different schedule for every interaction. As a rule, the schedule is composed based on common iteration. The end result hardly satisfies the needs of the end-users - there are “gaps” in it, some events can occur being scheduled in a row, the workload for the teachers is not distributed properly. 

In our case, the algorithm produces 5 options of the might-be schedule. After that the selections start, picking the most appropriate ones as the natural selection works, finding the best option to survive under certain conditions. It is the work generic algorithms do. 

Moreover, there are variables such as teachers' days off, weekends, and public holidays (which are different in different regions).

During the production process, another problem was pointed out. We need to have an opportunity to edit the schedule. Therefore, we’ve got the one that takes into account the overwork / non-completion of teachers. Also, the annual schedule is duplicated in the current one which can be edited and saved back to the annual.

This project and the challenges that have been overcome is another proof of the fact that our team always tries to find the best solution for each and every task, pushing the limits, applying all the skill and knowledge,  going beyond the standard view to deliver the product of high quality and in accordance with the customers’ needs.