How to write a neural network for a shooter game?

Question

Neural networks

I want to write a neural network for playing a shooter, tell me what is needed for this, I saw how opanAI, StarCraft 2, and Doom 1995 bots play, but I don’t understand how they train bots, if there are specific examples, please send them.

Answer 1

This training, in fact, is training for behavior in certain scenarios, the metrics of which can be taken from the best players. Game developers may have them.
For games, other methods can be practiced, such as decision trees.

Answer 2

In a general sense, artificial game opponents have some given goal, which they "seek" to achieve. To simplify setup and programming, the goal is decomposed into tasks and subtasks.
For example, the goal of bots in Quake 3 or Counter Strike is to dominate the match. To achieve this goal, you need to complete the following tasks: make the most kills of opponents, be killed the least number of times. To lead the list of killers, you need to own the best items, occupy the most advantageous positions on the map. In order not to be killed, it is necessary to have unpredictable movement tactics and, for example, avoid open spaces. These subtasks are decomposed into components until they are maximally accessible for programming.
Further, on the final list of subtasks, dynamically changing priorities are placed that determine the behavior of AI at a particular moment: in this situation, it is better to perform the avoidance subtask, and in the other, the subtask of finding and capturing a dominant position on the map. Priorities change based on some factors, for example: current equipment, state of health / armor / ammo, proximity of opponents, position on the map, etc.
In order to assign a certain tactic to a single bot instance, add rewards and punishments to the system - the same place in the match frag table is perfect for this.
The above is a very generalized version, and you will find more elegant implementations in the literature on the topic. Including using all kinds of neural networks.
First of all, as noted above, you need to decide on your goals: what kind of game AI you want to make, and what it should do. As soon as you formalize and structure your problem, you will quickly find the solution area and the necessary literature.