Summary: Neural Networks
Get a recap of some of the topics discussed in this chapter.
We'll cover the following
Due to the availability of game data and the increase in computational power, the use of NNs and deep networks is on the rise in data science in general and specifically within the field of game data science. Complex deep networks are used as they can generalize to highly complex relationships over unseen data, and as a result, provide better performance than traditional models. Such networks have been used to serve many purposes within the game production cycle, including churn predicting, predicting, and measuring customer lifetime value, recommending items, and discovering and forecasting player behavior patterns. Deep learning has shown good performance and results on these problems.
Issue of interpretability
Despite the good performance shown with this approach, the issue of interpretability discussed above is still a big concern. While it’s true that the final model and prediction output is important for games, why such results are obtained is also important for different stakeholders, including design, marketing, or production teams. This issue discourages using deep NNs for problems where such interpretability is important. However, it can sometimes be beneficial to use multiple approaches with the same problem to assess which approach would work best for our problem and dataset.
Disadvantages of interpretability
While interpretability has been a problem, there are several new studies looking into the development of interpretability tools and techniques for NNs. We believe this area will continue to expand in the future. Therefore, due to the accelerated research work within this area, we advise you to read more current resources, as the discussion in this chapter can become too basic or obsolete at the given time.
Get hands-on with 1200+ tech skills courses.