Yesterday was headache day, so no development.
Today was coding day. Found a big BUG!
I've been using a dictionary to contain my navigation data, that's to remove the need to recalculate the graph after every run of A*. I just made a deep copy of the original dictionary. I was deep copying the dictionary, then cropping the new dictionary to a bounding box around the start and finish points, by reducing the size of the graph it makes A* run faster.
Unfortunately deep copy, or even dict.copy() are very slow operations.
So while a long distance calculation of A* was taking 0.012 seconds, the deep copy operation was taking 0.12 seconds! 10 times slower than the A* function. A big drain on resources indeed. (nearly half a second for just six enemies!)
So I started by rewriting the cropping function. Now it crops data out of the original dictionary, copying entries and appending them to a new dictionary. Because it only copies entries inside the bounding area this is much faster and the resulting graph is much smaller. This takes just 0.01 seconds, much better and still somewhat faster than rebuilding the graph from scratch every time.
I've also done some work with the AI so it won't try to move if there's no route. It gets this info from the AI tactical map, though this is another area that is a little slow to calculate, but improved 10 fold since last week.
AI maps themselves are coming along rather nicely:
LOS is now calculated for each player character and also a fleeing map, so the enemies can get away to an unseen region. Archers will be able to move to a distant spot where the Player can still be seen and fire from there. If you pull out a bow and start shooting, enemies will take cover if far away, or move to engage quickly if they are close by. The more complex I make this part of the game, the slower the AI turns are going to be, so I have to find a compromise.
Again, cropping the area of calculation to include only the active players and a moderate border should speed things up. While allowing me to make the AI a little smarter. At the moment it only calculates areas which are accessible, i.e. not behind shut doors viewable. That's very fast but it means you can seal an enemy in a room and it won't move or do anything until the door is opened again. You could also avoid enemies by not looking at them. This obviously needs to be changed.
The AI maps are not very useful for the player, but they could be rewritten to use different calculations during the player's turn. LOS is useful for simulating guards, while accessibility maps can be used to make sure an A* pathfinding check isn't run if there's no valid route to that point, thus saving time.
One thing I have to look at is how opening a door during the AI turn immediately requires a re-run of the AI map generator if it makes new areas visible or accessible to the player's characters.
More experiments and refinements tomorrow.
Today was coding day. Found a big BUG!
I've been using a dictionary to contain my navigation data, that's to remove the need to recalculate the graph after every run of A*. I just made a deep copy of the original dictionary. I was deep copying the dictionary, then cropping the new dictionary to a bounding box around the start and finish points, by reducing the size of the graph it makes A* run faster.
Unfortunately deep copy, or even dict.copy() are very slow operations.
So while a long distance calculation of A* was taking 0.012 seconds, the deep copy operation was taking 0.12 seconds! 10 times slower than the A* function. A big drain on resources indeed. (nearly half a second for just six enemies!)
So I started by rewriting the cropping function. Now it crops data out of the original dictionary, copying entries and appending them to a new dictionary. Because it only copies entries inside the bounding area this is much faster and the resulting graph is much smaller. This takes just 0.01 seconds, much better and still somewhat faster than rebuilding the graph from scratch every time.
I've also done some work with the AI so it won't try to move if there's no route. It gets this info from the AI tactical map, though this is another area that is a little slow to calculate, but improved 10 fold since last week.
AI maps themselves are coming along rather nicely:
LOS is now calculated for each player character and also a fleeing map, so the enemies can get away to an unseen region. Archers will be able to move to a distant spot where the Player can still be seen and fire from there. If you pull out a bow and start shooting, enemies will take cover if far away, or move to engage quickly if they are close by. The more complex I make this part of the game, the slower the AI turns are going to be, so I have to find a compromise.
Again, cropping the area of calculation to include only the active players and a moderate border should speed things up. While allowing me to make the AI a little smarter. At the moment it only calculates areas which are accessible, i.e. not behind shut doors viewable. That's very fast but it means you can seal an enemy in a room and it won't move or do anything until the door is opened again. You could also avoid enemies by not looking at them. This obviously needs to be changed.
The AI maps are not very useful for the player, but they could be rewritten to use different calculations during the player's turn. LOS is useful for simulating guards, while accessibility maps can be used to make sure an A* pathfinding check isn't run if there's no valid route to that point, thus saving time.
One thing I have to look at is how opening a door during the AI turn immediately requires a re-run of the AI map generator if it makes new areas visible or accessible to the player's characters.
More experiments and refinements tomorrow.
Comments
Post a Comment