Clan war 2 projection - implementation differences

First of all, thank you very much for your improvements in clan war tag. you have added new values to see, and we are very happy with your changes.

We have seen an error with values. i attach an example. i think you use average from 2 days, and the result is under his actual real medals. we think you have to show only average from current day, to adjust better since where they can go.

thanky for all!

1 Like

This is not an error. We use average from all days as it shows the true strength of a clan. It further accounts for the nuance where some clans had their best players play first vs last, and that some clans don’t play all 200 decks. IF you make your best players play first then your projection will always be unrealistically higher than what you really will get.

I will give a simple example to illustrate my point.

Method 1: Avg medals of the entire week

Avg = 13325 / (200 + 44)
= 54.6106557377
Projection = 10,922.131147541

Method 2: Avg medals of today so far

Avg = (13325-8250) / 44
= 115.3409090909
Projection = 23,068.1818181818

What seemed more likely to you? The clan which has gotten 8250 points on Day 1 will get 10k points on Day 2, or 23k on Day 2?

Using Method 2 will only make sense for clans who play all decks. Using Method 1 (our method) will make sense and give a realistic projection for high-performing clan as well as low-performing clan. It gives realistic expectations to how a clan has been performing.

I understand that some people / “services” had implemented Method 2, and that leads you to believe that our method is “wrong” — when in fact our method is far superior than the alternative.

well, i understand what you try to do.

But it is very strange for clans where everybody plays. And for unstable clans i think it is very difficult too, because each day they play different number of battles. I think it is impossible for these clans to adapt a correct value. And i think, too, that it is very easy to know who wins in that categories (who plays more, take first position).

In our league, we know who is the strongest one, but we need to know if today he has a better or worst projection than usual. We know that is difficult to know if the best players plays first or not, but we use for a long time this values and it works very well to know all projections.

I was writing a lot of formulas to try help you and adjust the better possible, but i have deleted them, because I need first to ask some questions:

  • can you know how many war days clans have played in past days?
    → it could be good to know if we are in the 3rd o 4th war day, and use it to calculate average_medals_for_day in past days
  • can you know how many battles they played in war past days ?
    → it could be very interesting to calculate average_medals_for_battle in past days

There is no way to calculate either of those values reliably, maybe there are ways to approximate them though.

Regarding the 2 projection options, I’d basically summarise it like this:

  • M1: uses the aggregate performance of all clan members
  • M2: uses the aggregate performance of the players who already played

The key here is that we are trying to predict the performance of the players who HAVEN’T played today.
Meaning that if it was possible, you’d want to subtract the performance of those who already played, and the remaining performance would refer to the remaining players.

With M2, you are counting twice the performance measurement that you’d want to remove. I don’t understand why you’d want to do that.

An even more summarised version would be:
M1: use (AUB) to calculate B
M2: use A to calculate B

M1’s correlation should be significantly stronger

1 Like

This is not possible btw — the data from Supercell specifically does not isolate decks used on war days only. This severely limits what we can use to calculate the projection (and we won’t be able to include player’s individual historic performances as basis). To understand this limitation, you would need to know the API endpoints well.

We have already lobbied to have this change for 2 years but so far it has not been touched by Supercell.

Regarding the rest of the data, please see Alpe123’s reply. Unless you can articulate what what benefits your alternative formula provides, we stand by our opinion that our method is far more accurate than the alternative.

I also understand that it is easy to simply think that “the other method” is better because you are used to it — but being used to something does not mean that it is better.

Method 1 use basically an average. It’s interesting but not is important to daily war projection

Method 2 gives you a projection that day. Do we have possibilities today to win the war?

I’m sure that players in clans where everybody plays have asked for this improvement several times here. For clans in low categories i think they don’t help really to know if they will win that day or not. I’m sure they know wich capabilities have their opoonents. in many cases, you only needs if you play with more players than the others ones. But if you have similar values, i understand that method 1 is interesting.

But in high categories, everybody plays with 50 players, and the day ends very tied

A good formula for us is:
PROJECTION_200PLAYERS = MEDALS_TODAY + MEDALS_FOR_BATTLE*(200-BATTLES_TODAY)

What do you think if you add the two values? the first one is interesting. You see wich average have this clan in week (when we begin a week, we see his results to know how ill be our new opponents).
But you don’t know if a clan with bad/good average will have a good/bad one the current day

I suppose is difficult to explain a lot of values, but you could add a little help in that page to explain it. Or show a little help when you put your mouse over some values. You could use only this value if average is over 100 medals, for example, and empty for low average.

In my example:
MEDALS_TODAY = 31400
PLAYERS = 193 / 200
MEDALS_FOR_BATTLE = 154.58
AVERAGE = 30916
SIMULATION_200PLAYERS = 31400 + 154.58 * 7 = 32482

CLAN WEEK_POINTS MEDALS_TODAY CLAN_CUPS
197/200 154.58 → 30916 / 32482 (average/simulation)

I’m very happy with information that you already added. Now, i use above formula (PROJECTION_200PLAYERS) to calculate which possibilities we have before the day ends. But it would be very good if i you would add this value on your page.

It seems you that is not good to add this value. But if you use some help to explain it, then i think everybody can understand why are two values calculated

Thanks for all your answers. Best Regards,
Beni

First of all, thanks for adding those projections to Clan War 2, that’s very useful.

I was also surprised by the given projections, and happy to see now the explanation.

I think your formula makes sense somehow, but I also think that having both values (ie. yours and the simplified formula given by BeniJmnz) would be very valuable too. That’s the problem with projections, is that they are hard to relate to reality. But having them both would really make them better.

Kind regards,
Clash

Right now we are sampling data from a few hundred clans and plan to analyse it whenever we have time. These theoretical discussions can be useful, but seeing which approach is more consistent in the real world is what will help us decide. Maybe we can even devise another with even better performance.

Ultimately having several projection methods can be confusing, specially for users who aren’t aware of these discussions. So if it can be avoided, it’s better for user experience. We’ll see, maybe we find out that each method is useful for some scenarios.

Team - I was the original requestor of this enhanced feature, and I ran into this same problem doing things manually - how best to Project a clan’s medals for the day. In the end, I decided not to do it, and instead, I simply calculated the Average Medals earned by clan member for the day. This gave me a truer sense of where we stood in the day’s medal race…certainly better than just total medals. That would be my suggestion…just show us average by dividing the Clan’s Total Medals for the day only divided by Clan’s Total Battles played for the day only.

In BeniJmnz’s initial screenshot, the average would be 162.69 (31400 medals divided by 193 battles). The screenshot shows 154.58 as the average, which I believe takes the entire week into account…but a week’s average is irrelevant in the River Race (it matters in the Colosseum week). In River Race each day is unique as your Medal position translates to Boat Points each day, and then is completely reset to 0. So telling me the weekly average doesn’t help at all, and if anything is misleading.

We’ll be sampling this week’s results to be able to provide a more reliable read soon, but preliminary results show that the current method is better than other projections.

If everything goes as expected we’ll be able to analyse error rates of the projections as war days progress, and you can use that to see if results are reliable enough for your use case.

So far we’ve mostly received positive feedback, but I understand that it might not be useful for everyone.

Hi everyone. I think two values can be good to show
What do you think about this attached image?

In first place:
193 / 200 tell you about today (193 games played today, and limited to 200 games)
It is not difficult to understand that value is average today, and result → is about 200 games

In second place:
Weekly average (or perhaps there is a better description for this). In that, you keep this method that you consider better

I delete actual progress bar to add this all 2 lines (i think progress bar is difficult to understand, and this information help more.

If you don’t see good this image, i’m sure we can change some tags to clearly explain what these lines try to explain. It could be great to add daily information, and adjust it to 200 games. And I think all 2 methods can live in the same place.

I gave you a positive feedback for this improvement (i was happy when i saw it, because we could see remaining games and weekly average). Finally, if you add daily projection, we would be happy x 2

The casual player is going to find the two projections extremely difficult to comprehend. So this will likely not be implemented.

Right now, you know what this is and it seems intuitive to you. For the casual player, they don’t even know what the projections are, let alone trying to comprehend two different methodologies.

As Alpe has written, we will soon post a blog post illustrating why weekly average is better — our preliminary analysis shows that it is far more accurate and less error-prone than the daily average method.