I was testing the code, and skipped training to check the agent with random choices.
I changed the original code where the agent plays after training so to let the agent have 5 attempts with a trained model.
I also commented the command "action = random.choice(possible_actions)" to "action = random.choice(possible_actions)".
The problem is, the agent never gets a different reward than -1, even though it chooses all of the three possible actions at random. I checked the code and I realised it was suppose to give rewards as:
• +101 for a successful hit;
• -5 for a shot miss;
• -1 for every action.
But they were never programmed, nor in the code itself, not in the both the .cfg and the .wad files.
Either that or I missed something.
On top of that, the game is rendered on-screen, but it never updates according to the actions the agent is choosing.
I'm about to start checking differences between the code in the tutorial and the basic.py code that came with the VizDoom game. Anyone knows what's going on there?
I was testing the code, and skipped training to check the agent with random choices.
I changed the original code where the agent plays after training so to let the agent have 5 attempts with a trained model.
I also commented the command "action = random.choice(possible_actions)" to "action = random.choice(possible_actions)".
The problem is, the agent never gets a different reward than -1, even though it chooses all of the three possible actions at random. I checked the code and I realised it was suppose to give rewards as:
• +101 for a successful hit;
• -5 for a shot miss;
• -1 for every action.
But they were never programmed, nor in the code itself, not in the both the .cfg and the .wad files.
Either that or I missed something.
On top of that, the game is rendered on-screen, but it never updates according to the actions the agent is choosing.
I'm about to start checking differences between the code in the tutorial and the basic.py code that came with the VizDoom game. Anyone knows what's going on there?