Page 3 of 3

Re: RAGL through the seasons

Posted: Thu Sep 03, 2020 12:19 pm
Thanks Punsho - I've added a bit to acknowledge heavy tanks taking longer to produce than mediums:
Heavy tanks take longer to produce than mediums and so it would make sense for Soviet players to have shorter tank queues than Allies. However in Season 9 we saw the queue lengths roughly match.
I still don't really know why the tank queue lengths roughly match in Season 9, but not in Seasons 8 or 7.

Re: RAGL through the seasons

Posted: Thu Sep 03, 2020 3:12 pm
by Yara
TTTPPP wrote:
Thu Sep 03, 2020 9:00 am
Yep - the chat messages are available from the games. I did a bit of investigation into common messages in an earlier post: viewtopic.php?p=312071&sid=c4b05a909d16 ... bd#p312071 In particular I looked at the correlation between saying "gg" first and losing the game. I'm happy to look into something else in this area if you have a suggestion, although grouping games by player is quite hard since most seasons were before it was possible to log in.
Ah thanks! I always write "gl hf", I am wondering if that shows up, as you seemingly only looked for these strings? "glhf", "hf", "gl", "hfgl"?

Re: RAGL through the seasons

Posted: Thu Sep 03, 2020 4:05 pm
Yep - I actually did include "gl hf" in the "glhf" category - I stripped any non-alphabetic characters and converted to lower case. I also included "gg wp" in the "gg" category.

Re: RAGL through the seasons

Posted: Fri Sep 04, 2020 7:17 am
Next let's have a look at the light (pre-SD) vehicles. If we consider the average number build per game then we can see flaks are generally the most popular. Light tanks and APCs have some occasional late game uses, but flaks are good early game against infantry (for vision) and late game against air and artis/V2s.
021_lightVehiclesCount.png (49.36 KiB) Viewed 294 times
The average number of flaks per game has dropped over time. This is probably because they were built as a counter to radar units, and radar dome usage has dropped. Similarly we can see that light tank and APC usage has dropped. I think they used to be popular because players didn't build a service depot so often, and so they were the most durable units available.

Next let's look at how often the light vehicles are seen in a game.
021_lightVehiclesUsageRate.png (44.19 KiB) Viewed 294 times
Even though the quantity of flaks and light tanks has dropped, they are still being built by roughly the same percentage of players. Perhaps this shows that light tanks are still useful as an early vehicle, even if they're not used as part of the main army. APC usage is the most significant change - roughly half as many games saw an APC in S09 compared with S01.

Rangers are in an interesting spot. Allied players build them far more frequently than light tanks, but one is generally enough. This matches with the view that rangers are only useful early game - players get one of them for vision and never build another.

So to finish this post off, let's look at the win rates. As for medium/heavy tanks I've included the win rates for players who chose not to build each vehicle too.
021_lightVehiclesWinRate.png (69.11 KiB) Viewed 294 times
The biggest discrepancy is for rangers - particularly in Seaon 6. An Allied player building a ranger was more than twice as likely to win compared with one who didn't. More recently the gap is much smaller, and light tanks seem a reasonably valid choice. It's worth noting that building light tanks still seems slightly worse than not building light tanks though!

For Soviets then the lines are much closer. There's a small advantage for flaks over APCs, and we can see that players who don't choose APCs also do well. In S09 then Soviets did roughly 10% worse than Allies, which explains why all Soviet options are lower than the Allied ones.

In the next post we'll focus on the early game by restricting the figures to the first five vehicles.

Re: RAGL through the seasons

Posted: Sun Sep 13, 2020 7:19 am
As promised, next I wanted to look at how light vehicles are used in the early game. I've tried to keep the graphs in this post similar to the previous post to enable comparison. First up here are the average number of each light vehicle when restricted to only the first five vehicles (and averaging only over games where it was possible to build that vehicle - i.e. by Allies/Soviets).
022_lightVehiclesFirstFive.png (50.63 KiB) Viewed 215 times
The first obvious thing is the average totals are much lower. Many of the first five vehicles are harvesters, and so the average ends up somewhere near to one for each. The line for rangers is almost unchanged from the previous graph, suggesting that, while it's the most popular unit here, it's pretty much obsolete after the early game.

APC usage has dropped off, as has light tank usage to an extent. Comparing against the graph from last time we can see that the number of flaks used during the game has dropped significantly since Season 1, but the number built in the early game has slightly risen.

Next is the vehicle usage graph, which shows the percentage of Allies/Soviet players that built at least one of the given light vehicle. Again this graph is restricted to the first five vehicles built.
022_lightVehiclesUsageRateFirstFive.png (47.32 KiB) Viewed 215 times
Unsurprisingly the line for rangers is almost identical to the usage graph from the previous post. About 75% of rangers appear in the first five vehicles, so it follows that the usage of those rangers will dominate the usages of the other 25%. However since the usage percentages are about the same then we also know that the 25% of rangers are being built by players who already built one - i.e. no one is transitioning into rangers, but perhaps some players are doing a Ranger Danger build.

We can see that APC and Light Tank usage in the first five vehicles is broadly similar to the rest of the game. The raw percentages are about 5% lower though, so about 5% of players transition to them later, even though they didn't build one at the start. From experience this happens when your SD gets sniped, and you still want to build armour.

Flak usage in the early game is quite different from the whole game. I think this might be because historically they were used against prevalent radar units and APCs were better in the early game; now there are less radar units built but they're more popular in the early game.

Finally for completeness I've included the win rate graphs for building (and not building) each light vehicle in the first five units.
022_lightVehiclesWinRateFirstFive.png (73.6 KiB) Viewed 215 times
For the most part this graph looks very similar to the whole game win rates. There is one interesting point of variation that I noticed, which is that in Season 9 it was just as good to build a light tank or a ranger in the first five units. Furthermore, it was better to build a light tank as an early vehicle than not to build one. Given the often heard view that light tanks are underpowered then I thought it would be good to look at the win rates of all four options in Season 9:

Code: Select all

Ra LT:  41/ 71 (57.75%)
Ra --: 137/238 (57.56%)
-- LT:  28/ 48 (58.33%)
-- --:  11/ 29 (37.93%)
Admittedly there's a reasonable margin of error and many other factors influencing the results, but the data suggests that in Season 9 it was best to open with a light tank and not to build a ranger at all.

Re: RAGL through the seasons

Posted: Thu Sep 17, 2020 7:42 pm
Next up are harvesters and MCVs. I grouped them together because they're (almost) essential units and because they can't attack. One convenient feature of them is that they're available to all players, so it makes the percentages slightly easier to compute.
023_vehicleQueued.png (30.27 KiB) Viewed 144 times
Looking at the average number of each queued by season we end up with one of the most constant looking graphs so far. On average players have made 2.5 harvesters and 1.5 MCVs per game. The two main build orders (as detailed in the academy) each result in two harvesters and one MCV in the first five vehicles, which suggests there might be very few produced later in the game. It's possible that some players produce no harvesters or no MCVs, so let's look at how prevelent they are:
023_vehicleUsageRate.png (32.33 KiB) Viewed 144 times
So both units were standard in Season 1, and have got more ubiquitous since then.

Are they good units to build?
023_vehicleWinRate.png (47.12 KiB) Viewed 144 times
So yes they're good units to build. If you don't build an MCV or don't build an ore truck then you're likely to lose. One reason not to build either unit is if you're attempting an all-in another is because you made a mistake. Both are risky strategies.

Re: RAGL through the seasons

Posted: Tue Sep 22, 2020 6:45 am
Last time we looked at some stats relating to harvesters and MCVs. I wanted to continue that by looking at how many of each should be built.

First MCVs - I found that generally builds could be grouped into four buckets - no MCVs, one MCV, two MCVs and three or more MCVs. This obviously doesn't include the MCV that each player gets to start.
024_MCVsWinRate.png (39.46 KiB) Viewed 73 times
This shows that 2 or maybe 3 MCVs used to be the sweet spot, but by Season 9 the most reliable option was to build a single MCV. The extra utility from the first MCV is huge compared with that of subsequent ones. It allows you to continue producing buildings and defences while you drive one of them to a new location. Also often the reason for building a second MCV is that the first was destroyed, so the win rate for higher numbers of MCVs will be naturally lowered due to this.

Harvesters are a little trickier - there are more of them, and so I did the same analysis with six buckets.
024_HarvestersWinRate.png (51.35 KiB) Viewed 73 times
Here we can see that two has always been a good number, and that three has become more reasonable recently. I found it intriguing that 4 and 0 harvesters had good results in Season 1, but by Season 9 they seem much less successful.

Re: RAGL through the seasons

Posted: Tue Sep 22, 2020 7:31 am
by SirCake
Looking at the harvesters built only may be deceiving, since refs come with a harvester, too.
Can you show us how win rates look if the ref-harvesters are included?

Re: RAGL through the seasons

Posted: Wed Sep 23, 2020 6:56 pm
As suggested by SirCake I investigated including refs into the harvester count. Each refinery comes with a harvester and, as it only costs $300 more, most of the investment is in the vehicle.

Trying to repeat the analysis into harvester win rate by season I found that it was helpful to use seven buckets. It's a little hard to read, but we can see that adjacent buckets have had similar win rates.
025_HarvestersAndRefsWinRate.png (58.54 KiB) Viewed 36 times
So for example in Seasons 7, 8 and 9 the best total harvester count was 8, 9 or 10-11. In contrast in Season 1 the optimal count was slightly fewer - between 7 and 9. The worst harvester count has consistently been 0-5, followed by 6.

Games in earlier seasons were slightly longer on aveage than they are now, so perhaps it's also worth looking at the win rate of different harvester counts for games of different lengths. For every game (combining all seasons) I counted the number of harvesters a player built (refs and harvesters) and plotted it against the game length to the nearest minute. Each bubble on the graph has a size showing how many results it represents and a colour showing the win rate. Light blue means that the combination is bad, black means that the combination is good.
025_HarvestersVsGameLengthWinRate.png (64.85 KiB) Viewed 36 times
The first thing that stands out is that we can see that longer games usually have more harvesters, and that most games last around 5-20 minutes, although there's a long tail of 30 minute plus games that aren't even included on the graph.

Focusing on the middle section then there's a sweet spot of 8 harvesters for games that last 8 or 9 minutes. This seems to be considerably better than having 6-7 harvesters, which in turn is a huge improvement on having 5 or fewer. For 10 minute games then 9 harvesters seems best, and for games that finish after 11-16 minutes then 10 harvesters seems to be about optimal.