bjshnog

The people who created the WN8 rating are completely separate from XVM. I say this as the one who built half of WN8. Wargaming's API does not supply spotting/tracking damage data, and since WN8 is designed to be an account-wide rating, it can only use data from the API. In the future, if Wargaming adds that data to the API, then we may be able to use it in a rating system. Be aware that WN8 is terrible for single battles. Also, the particular thing you are looking at is a session statistics mod, which is also completely disjoint from XVM.

You could post in here. It's more likely to get his attention.

This isn't an XVM issue. Asking here won't solve anything. You will just have to try contacting the admins of the other websites again.

Well, the tool can output up to 10000 lines, and I've checked the consistency of the output. It has exactly the same form as the surrounding colour sets such as "wn6" and "kb". The last config I used, however, had one colour step every 50 points of WN8 (WN8 = 0, 50, 100, 150, etc), up to a maximum of 6000, and it just stopped everything in XVM from working. I will upload the default config and the modified config. UPDATE: I found the issue. There is a single missing " on every line. -_- Tested and working fine. :)

I am trying to make a tool which allows users to set gradient colour schemes for WN8 in XVM (meaning lots more steps), but it seems like it can't use more than 10 colours. Is this actually a limit? What can I do?

I want people to stop coming to the conclusion that we don't know how it works or that we just added things because they correlate. It's fucking infuriating when people attribute their own faults and ignorance about the entire subject of statistics to someone else.

I wrote an algorithm which attempts to predict the most likely distribution of damage, etc, across the played tanks. Link. (Is this what you mean?) This is for the API version. The more accurate version uses dossier files. I also think people should refer to the API version as "EVBE" (Estimated VBE).

Well, I actually meant just in case someone wanted to compare their own skill in Medium Tanks against Tank Destroyers, or if they wanted to look at their skill in all vehicles except arty; just for personal comparison. The actual player's rating will be overall, with every tank played included, and weighted by the number of battles in the tank.

For example, if a player wants to measure his 'skill' in his favourite tanks, he can select them using checkboxes or whatever (which would depend on the program/website used to calculate it). If a certain player's favourite tanks are the E 75, Pz B2 and T49 (for example), then he could select them from the list and get a VBE score for those tanks only. Otherwise, predefined sets will be available, such as vehicle tiers, classes, or battle tiers. There is a detailed description of how the system will work right now, just no actual final formulae. (there are formulae which show how the system will look, just nothing about how much each stat will be weighted) [link]

I should ask this: How many people here have read and understood my proposal for VBE? After reading through the replies about WN9/VBE in this thread, it looks like some people are aware of how it will work. :) Re: selection of tanks for the rating: I think I already suggested selection of tanks individually, but I may be wrong. I was definitely thinking about it, though, and I meant to post the idea in the "filter / weighting" post.

As of now, WN9 as been renamed to VBE (Vehicle-Based Efficiency) because it is vastly different from all previous WN ratings.

They aren't in XVM yet. WN8 may be implemented soon at some point, but I don't think it's decided yet. Also, VBE isn't anywhere near finished. You need macros in XVM to be able to have it work, because they link to information stored on the XVM servers. It doesn't calculate it on your end. ;) Also, WN8A is now just called WN8 and WN8D was renamed to WN9D, and now VBE. - bjshnog, VBE creator

Hello! It's good that there is a thread for WN8 and WN9 now. I just came back to this forum to check on the WN6 / WN7 thread, but then I saw a more relevant thread. I suppose I'll just leave this here: ⟪WN9D⟫ Development + ⟪WN9⟫ Information. WN9D is to be used on websites like vbaddict.net and programs like WoT Statistics, because it uses stats for each tank. The last part of the WN9 calculation algorithm will be identical to that of WN9D, but since stats per tank are not available through the API, it contains an algorithm which will attempt to reconstruct stats per tank with the limited data. That stat separating algorithm can be found here.

Just a note: WN8 won't be out for long before WN9 comes out, so maybe it's not a good idea to put it in XVM. You're better off waiting for WN9. :) (translate please)

WN8 seems to be close to completion... An important thing to say is that unicums and real super-unicums will see massive increases in WN8 from WN7, but the average will stay about the same.

⟪WN8⟫ Development / Description Post #1 is the main concept. Posts #160+ are the most important.

Should I post a link to the development thread here?

Is XVM-stat.exe still required for stats, or is that built into XVM?

I don't know if anyone has suggested this or if it is already in XVM (Google Translate is bad): Is it possible to make XVM show rating colours on a gradient? Example: In the config file, WN=1200 -> #55FF55 and WN=1600 -> #55AAFF. Can XVM be modified so that, for a player with WN=1400, it shows in the colour [(400-200)*RGB(55,FF,55)+(200)*RGB(55,AA,FF)]/(1600-1200) = #55D4AA (50% between colours for 1200 and 1600)? Can this already be done?

In order to do that, many more stats would have to be stored on the XVM servers. Right now, I think they just calculate WN6 based on the players' stats, then store the resulting number on their server(s), but I could be wrong.

Actually, as per my suggestion, it is better to use WN8. I don't mean to sound pushy, but WN8 will work well for every tank individually. Don't use the top 0.01% or the TOP100. Use the top numbers which will be determined by our analysis. :) Another idea! When showing a tank's TWN8 (Tank WN8) in battle (if it gets implemented in XVM), it should not show the value recorded in the player's Service Record. It should, instead show a different value which pretends that there are some number of extra battles played at that player's overall WN8. For example, if a player plays 5 battles in a new tank at WN8 = 1600, but the player's overall WN8 = 1000, then it acts as if there are 5 battles played at 1600 and (for example) 20 battles played at 1000. The number shown in-battle would not be 1600, but instead, (5*1600 + 20*1000)/(5+20) = 1120. This will make up for "beginner's luck coefficient" and the special MM that new tanks get. Later on, that player reaches overall WN8 = 1400 and they now have 980 battles in that "new" tank (which now has specific WN8 = 1800). The number that shows in-battle as TWN8 will be (980*1800 + 20*1400)/(980+20) = 1792. Instead of adjusting for newbie stats, we will be using all data points. However, they don't all have the same weight. Each point will be weighted as number of battles in the specific tank divided by total (data acts as if there are s/t points in the same place instead of only 1). I chose to do this after considering several sources of randomness. I multiplied each of the factors together and the result was s/t. Furthermore, as many people know, different classes/types of tanks have different roles on the battlefield. I have come up with a system which will create more balance between tanks. For example, players of light tanks will be considered to be playing the tank better if they scout well, meaning damage has less weight and spots have more. WN8 will also be split into two versions: WN8D: 'D' means 'Dossier'. This is the rating that we will use to calculate the expected stats for WN8(D) = 1000. It is what WN8 would be if WG's API provided data for every tank a player has played. It can only be used in conjunction with dossier files, which will allow programs like WoT Statistics to calculate WN8D and websites like vbaddict, which relies on dossier files being uploaded. WN8A: This is the rating that can (will?) be used in XVM and other websites such as WoTLabs and WOT-News. It is not as accurate as WN8D, but it is the best that can be done using the data provided by WG's API. When completed, it will be vastly superior to WN7 and actually allow us to accurately say which tanks are more powerful than others in actual practice. In forum signature images, this will be labeled as simply "WN8". I'm just saying this so that you guys know where WN8 is going.

I was actually thinking about that a lot today and I have a solution for it. When we calculate the line to predict the stats for WN8 = 2000, the first step is to find the mean point (the average point of the dataset for both parameters). Next, draw a line which goes straight from the origin (because 0 of all stats results in 0 WN8) through that mean point. At tier 1, that line should be multiplied by a certain percentage (for example: 96%). Every point below that line is ignored in the next step. Take the mean point of the remaining points (above the line) and draw a line through that. Ignore the first line. The second line drawn is what we can use to predict stats at WN8 = 2000. However, if the scatterplots still look weird, then we can also ignore all of the points where WN8 < 1000. As the tier value increases, that percentage slightly decreases (because there are less battles in which the player was a complete newbie).

Could you explain what you mean a bit more? I don't quite understand.

We use the current formula we have to predict what damage (or other stat) is expected for WN8 = 2000. After this, we replace all of the stats that we used in the previous WN8 formula. Then, Praetor77 can reanalyse the stats with the evolutionary algorithm and have a new formula. We repeat the process many times until the top stats don't vary as much. If you meant that skill distribution in a specific tank does not reflect skill distribution over the entire server, then I agree, but that does not matter. The players who have lower skill (decided by WN8) will have lower stats in that specific tank, and the process we use to calculate top stats will tell us what to expect from a player with WN8 = 2000. If you meant average specifically, then what I meant in that post was that XVM could show the average stats of the players (which can be calculated on a larger scale by the XVM people) AND the top players of that tank AND the stats to be expected from a player with 2000 WN8. I did not mean "average" as in the stat expected from an average player. That was only so that people could compare themselves to other players of that tank. So, for the Object 907, for example, our stats show an average damage of something like 1200. We know that is incorrect, but we also know that the skill of those players was low, so our process tells us what to expect from a player with a higher skill level. Also, if the XVM developers have a larger database of player stats they would like to share with us, that would be brilliant. :) There is actually a better way to calculate top stats than what we are doing at the moment, but it is much more complex than the current method (which is already complex enough as it is). At the present moment, we are comparing the player's aggregate stats to the expected top stats, but with the stats we have, we could find individual tank WN8 for each player and then average the results, weighted by number of battles in each tank. I think the most this would do, however, is speed up the process of calculating the top stats for each tank. A very important note (in case you didn't see it already): The current system we are using is independent of player skill distrubution in each tank. This means that if a tank has really bad stats, it is very likely that the players who played the tank are really bad at the game.

I was going to ask that you refer to me as "bjshnog" on this forum, since "shnog" has more of a chance to cause confision.