Every great company started with a great vision. Wildlife is no different.
When our founders started working on this crazy dream, in 2011, they didn’t have all the answers, but they were focused on making great games that would make people happy. And because we're committed to that until now, things worked out great for us. So, this is an example of how starting something from scratch needs the vision to guide you towards your goal.
Today, we have a very ambitious goal in Wildlife’s history: creating our first specialized Machine Learning (ML) team. So we had to write down our vision statement, to make sure we would have our hearts and minds in the right place. This was the result:
"Wildlife will be a reference in the Machine Learning area by the work we have done and the results we have accomplished. We’ll be known by using the best practices, have some of the best experts in the world, and achieving incredible results. We set the bar very high, to motivate us to do our best work, and to bring the people who will help us achieve it."
But writing beautiful words is quite easy. Thus, in this article, I want to share with you, as an ML team member, how we plan to transform our mission in reality in terms of responsibilities and duties. Let’s break it down.
“Wildlife will be recognized as a reference in the ML area.”
Wildlife is already one of the largest game companies in the world. Our games were downloaded more than 2 billion times in our 9-years old history. Also, in 2019, only one of our games (Sniper 3D) was the 14th most downloaded game in the world.
In terms of data, we generate terabytes every day.
That means we have a great opportunity of being known as an ML reference in the business — the next step is designing the strategy to get there.
Google, Uber, and Netflix are some examples of big companies with a good ML reputation. And what do they have in common? An end-to-end (E2E) pipeline. That includes data acquisition, feature selection, feature engineering, training, deploy, and monitoring.
That’s what we're aiming for at Wildlife. We must develop a fully automated pipeline that everyone in the company can easily apply in our games. It must be easy to use, maintain, and integrate.
However, I believe it must not be developed from scratch. By using third-party components from well-known libraries/frameworks whenever possible, we can avoid “reinventing the wheel”, making it flexible and extensible. Moreover, a fully automated ML pipeline must be reproducible at any point in time for any model. It allows for faster development and easy ways to debug/test each step of the model.
In the future, we can make the components of our ML pipeline as services. In other words, develop a Machine Learning as a Service (MLAAS) platform available for anyone who needs it. Once we have that, we’ll write another post explaining our solution. So, stay tuned!
We want to take Machine Learning in Wildlife not only to the next level but also to the highest and beyond.
“We will be known by using the best practices.”
One of the purposes of the ML Engineers is to deliver tools and a platform for Machine Learning development. Most of our teams already apply Machine Learning in their own way, testing different tools and frameworks in the development of their products — and that’s fine.
However, sometimes, these tools can not be shared with other projects, or they are not efficient enough. As an MLE team, we must help some of these developments and integrate a more effective version of them. It will scale any technique for the whole company, ensuring consistency and maintenance.
Secondly, we want to set outstanding standards covering the whole ML lifecycle. It includes the best practices of software engineering and the best way to accomplish different goals. With faster machine learning development, we can quickly start using it in more areas and measuring its effectiveness.
In the future, I hope our job would be reduced to add new models to the pipeline and to maintain/replace/improve the tools we use.
“Because humans may stay with the intellectual job and machines may do the hard work.”
“We will achieve incredible results with ML.”
Although we're a new team, Machine Learning has been used widely at Wildlife for years. Today, we have around 30 models for different kinds of tasks in production.
To give you an example, our Data Scientists developed a model to predict the Life Time Value (LTV). The LTV represents how much a user will spend in a game during his entire life. This model can predict the 2-year LTV by looking only at three days of game data! Can you imagine how important (and amazing) it is?
If we know in advance the revenue of a game in the next two years, we can allocate better the resources for that game. It includes development, UA, marketing, and so on. As far as we know, this is already a top-level result in the industry of games. And the MLE team wants to keep this high bar and accomplish even more.
This is only one example. Wildlife is a company that benefits with ML applications and still has a massive number of possible applications to benefit from. We want to take advantage of these opportunities to learn, make an impact, and help us reduce risks in this endeavor. We also believe that models can help mitigate arbitrary, inconsistent, or faulty decisions, improving the decision making process.
We'll not stop being a game-first company, but I hope we become an ML-driven one.
For me, this is more than just another Medium article available on the internet. It’s an open letter to remind us what we, as a Machine Learning team, expect to accomplish in the next five years or less. I hope we can come back here in the future and realize that we achieved our mission or even more.
Also, it’s an invitation for those who have the same vision and believe that can help us achieve it. Furthermore, if you love Machine Learning as much as we do, and you want to be part of one of the world’s largest game companies, how about joining our MLE team? Take a look at our available positions and come explore the future with us!