Planetarion Forums - View Single Post

Mzyxptlk · 19 Jul 2016, 17:01

Here's the output of my new and improved stats analysis tool:

Short: contains only 'immune' or 'vulnerable' for each fleet to every other fleet.
Long: also contains the value change after each possible battle.
Full: also has the full brep of each possible battle.

My analysis tool works as follows.

An algoritm is used to create a reasonable (not "perfect", whatever that is!) fleet composition around each pod. For each pod that is selected, first all the escorts for that pod are added (more downward than upward firing ships) and then the remaining targetting gaps are filled with defense ships. Value distribution is roughly 2/3 attack fieet, 1/3 defense ships, for 1m value total. For example, for Ter De, 568 Demeter are built as pods, 35926 Pegasus and 11468 Chimera are added as escorts, and 10417 Phoenix, 4187 Syren, 3333 Wyvern and 3193 Dragon fulfill the role of defense ships.

Each fleet is then pitted against every other fleet, with the following modification: the value of the defender is reduced to 70%, to take into account that people mostly either attack downward or team up. When a fleet is attacked by another fleet, the defender has a choice of which ships to defend with. Because I could not figure out a way to determine the best possible outcome for the defenders in advance, I just simulate all options. Each ship that fires at the given attacker is tried individually, with and without flak. Afterwards, all defense ships that fire at the given attacker are used together, again both with and without flak. See Mosquito vs Illusion for an example where the defender has a choice between no less than 8 fleet compositions: flakked and unflakked Phantom, flakked and unflakked Ghost, flakked and unflakked Peacekeeper, and flakked and unflakked Phantom + Ghost + Peacekeeper. Flak in this context is 1) all 'useless' ships that are in the same class as a 'useful' defense ship (classic flak) and 2) all ships that the attacker doesn't kill or steal (covers 0-loss defense and EMP flak). T2 flak is not tried (sending Phantom to flak for Harpy against Ghost, for instance).

This results in a number of possible battles, which I run through a homegrown bcalc.

If all of these battles would result in the attacker choosing to land, then that means the defender is vulnerable to being roided by the attacker. On the other hand, if one or more of them would result in a reasonable attacker choosing to recall, then the defender is immune to being roided by the attacker, because that's obviously the combination of defending ships the defender would pick. Whether or not the simulated attacker is willing to land depends on how much the they lose compared to the defender, as well as the outcome of !roidcost: the more the defender loses, the more the attacker is willing to lose. However, I find that 80%-90% of lands are no contest: they are either nearly lossless for the attackers, or total massacres in the defender's favour.

In the Mosquito vs. Illusion example above, all 8 possible choices for the defender are lossless for the attacker, so Illusion is vulnerable to Mosquito. The same is not true for the very next matchup, Mosquito vs. Vampyre, in which the attacking Mosquito fleet needs about 71 ticks (roidcap of 200, Corp, 30 FCs) to recoup its losses and the defenders gain 11k value to boot (standard salvage).

The weakest aspect of my simulator is the fleet compositions. The program doesn't make straight-up stupid choices, but sometimes they could be more optimal. For example, if you're going Ter De, you might prefer building Harpy (makes ally def ETA) rather than splitting your value between Syren (T2 overlaps with Chimera) and Wyvern. Or you might not, because having both Wyvern and Dragon (which you need anyway) gives you 2 roiding fleets. There are choices to make that depend on more than just the raw data in the stats sheet and the user should be given the option to make those choices.

The format of the output also isn't the greatest.

No teamups, neither for attack nor defense: too much work.

[edit] I think I fixed all the formatting errors in my links now, plus I included the fleet compositions the program decides on.

19 Jul 2016, 17:01	#50
Mzyxptlk mz. Join Date: Aug 2005 Posts: 8,587	Re: R68 Shipstat proposals Here's the output of my new and improved stats analysis tool: Short: contains only 'immune' or 'vulnerable' for each fleet to every other fleet. Long: also contains the value change after each possible battle. Full: also has the full brep of each possible battle. My analysis tool works as follows. An algoritm is used to create a reasonable (not "perfect", whatever that is!) fleet composition around each pod. For each pod that is selected, first all the escorts for that pod are added (more downward than upward firing ships) and then the remaining targetting gaps are filled with defense ships. Value distribution is roughly 2/3 attack fieet, 1/3 defense ships, for 1m value total. For example, for Ter De, 568 Demeter are built as pods, 35926 Pegasus and 11468 Chimera are added as escorts, and 10417 Phoenix, 4187 Syren, 3333 Wyvern and 3193 Dragon fulfill the role of defense ships. Each fleet is then pitted against every other fleet, with the following modification: the value of the defender is reduced to 70%, to take into account that people mostly either attack downward or team up. When a fleet is attacked by another fleet, the defender has a choice of which ships to defend with. Because I could not figure out a way to determine the best possible outcome for the defenders in advance, I just simulate all options. Each ship that fires at the given attacker is tried individually, with and without flak. Afterwards, all defense ships that fire at the given attacker are used together, again both with and without flak. See Mosquito vs Illusion for an example where the defender has a choice between no less than 8 fleet compositions: flakked and unflakked Phantom, flakked and unflakked Ghost, flakked and unflakked Peacekeeper, and flakked and unflakked Phantom + Ghost + Peacekeeper. Flak in this context is 1) all 'useless' ships that are in the same class as a 'useful' defense ship (classic flak) and 2) all ships that the attacker doesn't kill or steal (covers 0-loss defense and EMP flak). T2 flak is not tried (sending Phantom to flak for Harpy against Ghost, for instance). This results in a number of possible battles, which I run through a homegrown bcalc. If all of these battles would result in the attacker choosing to land, then that means the defender is vulnerable to being roided by the attacker. On the other hand, if one or more of them would result in a reasonable attacker choosing to recall, then the defender is immune to being roided by the attacker, because that's obviously the combination of defending ships the defender would pick. Whether or not the simulated attacker is willing to land depends on how much the they lose compared to the defender, as well as the outcome of !roidcost: the more the defender loses, the more the attacker is willing to lose. However, I find that 80%-90% of lands are no contest: they are either nearly lossless for the attackers, or total massacres in the defender's favour. In the Mosquito vs. Illusion example above, all 8 possible choices for the defender are lossless for the attacker, so Illusion is vulnerable to Mosquito. The same is not true for the very next matchup, Mosquito vs. Vampyre, in which the attacking Mosquito fleet needs about 71 ticks (roidcap of 200, Corp, 30 FCs) to recoup its losses and the defenders gain 11k value to boot (standard salvage). The weakest aspect of my simulator is the fleet compositions. The program doesn't make straight-up stupid choices, but sometimes they could be more optimal. For example, if you're going Ter De, you might prefer building Harpy (makes ally def ETA) rather than splitting your value between Syren (T2 overlaps with Chimera) and Wyvern. Or you might not, because having both Wyvern and Dragon (which you need anyway) gives you 2 roiding fleets. There are choices to make that depend on more than just the raw data in the stats sheet and the user should be given the option to make those choices. The format of the output also isn't the greatest. No teamups, neither for attack nor defense: too much work. [edit] I think I fixed all the formatting errors in my links now, plus I included the fleet compositions the program decides on. __________________ The outraged poets threw sticks and rocks over the side of the bridge. They were all missing Mary and he felt a contented smug feeling wash over him. He would have given them a coy little wave if the roof hadn't collapsed just then. Mary then found himself in the middle of an understandably shocked family's kitchen table. So he gave them the coy little wave and realized it probably would have been more effective if he hadn't been lying on their turkey. Last edited by Mzyxptlk; 19 Jul 2016 at 17:21.