Updated Combat Odds Calculator

[rkr1958]

Borger has worked his magic again and has come up with a much better and accurate odds calculator (we think). This "update", which will be included in the GSv2.10 official release, will NOT be released in general to the testers. However; if you're interested in helping to test out this updated combat odds calculator please read the next post below.

By the way, the GSv2.10 code is in lockdown, which means no further changes except minor tweaks (e.g., language file translation updates, combat odds calculator) THAT DO NOT impact play balance will be considered.

I made an update to the odds calculator for all battles except air duels and strategic bombardment. The latter two work pretty well with the old odds method.

I think I have a feeling why the odds calculator was not performing well. The problem seems to be that the formula calculates a strength of the unit based upon the strength of each step + 4 divided by the opponent’s survivability. The unit’s morale also affects the combat strength.

This means you get one value for the attacker and another value for the defender about the firepower they use to inflict damage. That is probably not a good idea to calculate expected losses.

The combat resolution makes die rolls per step you attack with + 4. So a 10 step unit gets 14 changes to make a hit.

Let’s say you have 25% chance to make a hit per 14 roles. Then the total number of hits will not be: 0.25 * 14 = 3.5 This is how the calculator works. When units aren’t very strong then the error is not so big, but when units get quite high firepower then the simple method seems to overestimate the real losses.

Instead I actually made a new function that will make 20 real combats and average the result. The odds will therefore show the average of 20 simulated battles. That should give a much more accurate result since it actually uses the resolve unit battle function to calculate the losses. The real battle result will differ from the estimated result, but I think you will now see results sometimes better than expected and sometimes worse. Before the real result was most often worse than the expected result.

Every time you move the cursor to a new hex the expected results will be recalculated. So if you move the cursor between several targets you might see that the predicted odds may differ slightly. This is because of the spread you can get with 20 samples.

Can you please try out the updated files and run tests between estimated combat results vs real results.

[rkr1958]

If you would like to test out the update combat odds calculator and agree to the following please PM me with your email address and machine type (i.e., PC or MAC) stating such. I will send you the resources and, if you're on a PC, the minor patch necessary to test out the calculator. Below are screen caps of the resources I've pulled together to test it out. It includes both a worksheet (cap 2) and an excel spreadsheet template (cap 1). If you agree to help out what I need back from you is NOT the hand filled in worksheet (cap 3); but your data samples entered into the Excel spreadsheet (cap 4). To properly evaluate this we need samples taken over all years.

Note: both the expected odds and flashed combat results can exceed the units strength. However; the spreadsheet will not allow you to enter expected odds or losses greater than the strength of the unit.

[Peter Stauffenberg]

I just want to mention the reason we're doing this is because Slitherine told us they can't start working on the GS v2.1 installer until after Easter. So we agreed with them to deliver the final files on April 15th to people at Slitherine to build the installer. That meant we had 2 weeks free to do cosmetic fixes to the files going to the installer.

Victor was fixing the message_spa.txt file Spanish translation file and we decided to see if it was possible to improve the odds calculator since it's annoying to have such a gap between the expected and real results.

We will NOT make any changes that will require testing of game balance. We only fix typos in txt files etc. Fortunately we haven't received new bug reports requiring fixes for quite some time. The exception is the renaming of the UK Fortress Scapa Flow garrison. It was interpreted as SS and not allowed to rename. That is a very minor issue and people will usually not want to rename that unit anyway. So we decided to not fix this since it could introduce a bug somewhere.

[Cybvep]

This is not a joke, right?

[rkr1958]

This is not a joke, right?

I'm not following you ... why would you think this is a joke?

[Cybvep]

Many strange things happen on April Fools' Day.

[rkr1958]

Many strange things happen on April Fools' Day.

Oh ... I'm such a "square" that it didn't occur to me that today is, "April Fool's Day."

[pzgndr]

Minor comment. The header at the top of general.txt still read GS v2.0 in the last install I had. There may be some other admin updates elsewhere?

[Peter Stauffenberg]

Minor comment. The header at the top of general.txt still read GS v2.0 in the last install I had. There may be some other admin updates elsewhere?

Fixed. Will become part of the final version.

[ncali]

This will be a great improvement, particularly for new players. I'm glad it's not an April Fools joke (at least, I think it isn't).

[richardsd]

This will be a great improvement, particularly for new players. I'm glad it's not an April Fools joke (at least, I think it isn't).

if it is, its installed on my machine

[Schnurri]

I'm collecting lots of data for ground combat which seems better than before. The Naval combat is not working I think. It always shows 0:0 and then you get very divergent real results. Have tried so far in 41 and 43.

[Peter Stauffenberg]

Thanks, I will look into the naval combat.

[Peter Stauffenberg]

Bug found and fixed. Updated game.class sent to Ronnie and Paul for distribution to those interested in trying it out.

The reason for the bug is that the combat sequence for naval combat is different than land combat. For naval combat the attacker shoots first and the defender last. For land combat the defender shoots first and the attacker last. For air combat both sides shoot simultaneously.

I had simply forgotten to add the result of the combat for naval battle to the return variable for the function. Therefore the initial values were sent back (0, 0).

[Peter Stauffenberg]

Please notice that dogfight and strategic combat is still using the old odds prediction because these battles used a different combat resolution. That doesn't matter because stratetic combat is against a resource and not a defender with high efficiency. So the error there is small. The main error you get is that you don't know if your bomber gets intercepted or not. If it's intercepted you lose more steps than predicted and inflict less steps as well.

We couldn't simulate that through real combat or ou would know the unit would become intercepted before bombing because the odds are worse than they should have been.

The same argument is for air to air combat. You don't know if you get intercepted or not. Therefore the odds has to be shown as if you didn't get intercepted or you know the attack would not be a good one to make.

[Schnurri]

Something flaky also with the Paradrop calculator. When checking a drop it refuses to calculate the odds - screen is just blank. Then, do drop with predictable results.

[Peter Stauffenberg]

Thanks again for finding that. I altered so paradrops use the old odds calculation since it can be intercepted by enemy fighters. Now you will see paradrop odds again. Ronnie and Paul will distribute the updated mapcanvas.class file to those of you testing

[Kragdob]

Is there a chance for odds against non-adjacent units? Something similar to info about unit repair cost would be great.

[rkr1958]

Is there a chance for odds against non-adjacent units? Something similar to info about unit repair cost would be great.

Not in this release.

[rkr1958]

247-samples from Doug & me. Come on folks ... we need more samples; especially with the latest patch. I've provided a summary below which includes samples before and after the patch. There were some really bad results pre-patch #4 for armor versus armor in 1945 that I threw out. However; there are armor samples for prior years pre latest-patch, which I think contributed to the -1.1 error for armor. I suspect we'll see that error come down if we get more samples.