Batch Training (The Good, The Bad and The Ugly)

[Marbod]

On 26/08/2021 at 5:30 PM, alre said:

that's more or less the opposite of what I do:

Well it turned to be the better choice for me too by now after reading the posts again. I missed somehow the second table which corrected the math that assumed that batches are only good at high numbers 27+.

[Freagarach]

I wish I could like/thank the first post again, after the update! These kind of calculations are really interesting!

[Ceres]

@Stan`you please consider pinning this thread/ the first post of it? Thank you!

[Stan`]

Sure, but why?

[Ceres]

Well, personally I found this thread and @Micfild 's calcs and discussion so elaborate and helpful to understand the pros and cons of 1b1, batch, and autoQtraining that I thought it would be worth placing it at the top of the subforum about general 0ad gaming. I know that there are plenty of other highly appreciated posts and not all of them can be pinned.

[Yekaterina]

[real_tabasco_sauce]

nice handwriting!

[real_tabasco_sauce]

keep in mind that batch training allows the next batch to be trained sooner.

[Yekaterina]

6 minutes ago, real_tabasco_sauce said:

keep in mind that batch training allows the next batch to be trained sooner.

That is one thing that I forgot to consider. However, I really doubt whether anyone can sustain production sizes of 27...

It is the most suitable strategy for post war regeneration, if you are floating resources. You are left with 2 or 3 barracks, each producing 27. This will boost you from 100 pop back up to 181 in just 2 minutes and 20 seconds, allowing you to fight again.

[Yekaterina]

I would like to conclude at the following:

If you are new to the game and struggle with booming up to the standard of 200 pop in 12 minutes, then autotrain is for you. It will save you much hassle in selecting the right batch sizes and makes you more flexible (especially when a rush happens).

Pro players, carry on and do your thing. Maths can never take care of all the surprises and varied strategies that you might be playing into.

[BreakfastBurrito_007]

between autoqueue and manual batch training, the main thing is to do whatever it takes to keep your resource counts low. 
 

I like the options right now and I don’t think I will commit to either. 
 

In my opinion, the ease of autoqueue and the mental space it frees up is the most important aspect of it

[Yekaterina]

@real_tabasco_sauce mentioned that autotraining in larger batches will start the second batch faster, which is something that I did not take into account. So I am now comparing autotraining in batches of 2 units to autotraining one by one. The mathematical derivation:

Plotting the two functions onto a graph in Desmos:

 

Vertical axis: total resources gathered (arbitrary unit)

Horizontal axis: time passed (arbitrary unit)

Blue line: R(t) for 2 by 2

Red line: R(t) for 1 by 1

Black line: the amount of resources by which blue leads red. 

 

Analysis and conclusion:

Initially, red is greater than blue: 1 by 1 wins for short time lengths. If you want a quick burst of resources then training 1 by 1 is superior for a short duration. 

Then blue catches up and exceeds red at a finite time: this is the critical value; if you are planning on batch training continuously for longer than this critical value then 2 by 2 will maximise your economy. 

The black line suggests that the advantage of 2 by 2 will blow up after a significant amount of time, so if you are booming peacefully from scratch then 2 by 2 is favourable. 

 

@berhudar This may be of interest to you. Would you like to develop a build order for 2 by 2 training? Tessekurler

[Yekaterina]

This is the link to the graph for anyone who is interested:

https://www.desmos.com/calculator/vlu7yvlz8y

You can tweak the values of the parameters to see the result for each unit type. 

[Yekaterina]

An example of gameplay on 2 by 2 batch training:

metadata.jsoncommands.txt

It was my first attempt so some parts of it was poorly executed. However, you can see that it is still a fast booming strategy in spite of my poor execution. 

 

[real_tabasco_sauce]

2 hours ago, Sevda said:

The black line suggests that the advantage of 2 by 2 will blow up after a significant amount of time, so if you are booming peacefully from scratch then 2 by 2 is favourable. 

I guess we can infer that 3 by 3 and 4 by 4 etc. have steeper and steeper exponential growths, and with more delay than 2 by 2. Nice math @Sevda this is a good way of looking at it.

[Freagarach]

On 24/08/2021 at 11:34 AM, Freagarach said:

https://trac.wildfiregames.com/browser/ps/trunk/binaries/data/mods/public/simulation/components/ProductionQueue.js#L775
So what happens is that on a progresstimeout (called once every 1000 ms) that time (1000 ms) is allocated to the queue. The time needed for an item is then subtracted from that allocated time until there is no more time left, then it will be repeated on the next timeout. When an entity finishes being produced and there is still allocated time left, that is taken from the next item in the queue. However, when AQ is on, it will just ditch the remaining time and wait for the next timeout.

Example: An entity costs 5.6 time to produce. After five timeouts, both queues (with and without AQ) have 0.6 seconds left. The first entity in the row is finished on both queues on the next timeout. The rest 0.4 seconds (1 second allocated time minus the 0.6 it costs to finish the entity) will be discarded by the AQ queue, but will be used by the other queue to start on the next item. That means that after 6 timeouts (== seconds game time), the AQ queue will have 5.6 seconds left for its second entity whilst the normal queue has only 5.2 seconds left.

@Sevda have you considered this?

[Yekaterina]

10 hours ago, real_tabasco_sauce said:

I guess we can infer that 3 by 3 and 4 by 4 etc. have steeper and steeper exponential growths,

Yes. The only problem with 4 by 4 is that you cannot sustain production at this rate in early game with 300 resources. You will drain resources much faster than you can gather. 3 by 3 spamming women might be the upper limit, although attempting a combination of 2 and 3 might work. I am just able to sustain 2 by 2 production (due to my poor skills). Maybe @berhudar or @ValihrAnt can try 3 by 3 autotrain. 

5 hours ago, Freagarach said:

@Sevda have you considered this?

I wasn't aware of this mechanic. I am slightly confused, as I heard that the game refreshes 4 times per second (each turn is 0.25 second) so anything in between is rounded?

I am not sure if this delay will be significant in the long run and whether it will affect both batch sizes in the same way. 

How do you determine the number of timeouts per second? I think the more timeouts per second the less inaccuracy and delay but it might cause extra load on the CPU.  

 

[BreakfastBurrito_007]

hmm, so the main conclusion is that the famous 6 woman start is maybe not the best? I saw @berhudar's start for ptol and use that now, but I have not done 2 batches of 3 at start with other civs. I have been using autoqueue when my focus is elsewhere or where I would find myself making small batches repetitively.

I guess the new general agreement is that 1 by 1 is better unless you are floating res because of it. For example, from berries and hunt you can have high food income, but only one cc to make them from, so I have observed even the people who use autoqueue the most will make batches of sizes 3 to 10 during that period. 

If you make women to reduce excess food, then batches are better, but if you are making women to fix a wood shortage, then 1 by 1 is better.

Am I understanding this correctly?

[Yekaterina]

5 minutes ago, BreakfastBurrito_007 said:

so the main conclusion is that the famous 6 woman start is maybe not the best?

In most cases it isn't, unless you are racing against time to get units out very early. For example, doing a very early cavalry rush: autotrain will give you an eco advantage but the 6 women allows cavalry to be produced earlier to threaten the enemy more. 

7 minutes ago, BreakfastBurrito_007 said:

but if you are making women to fix a wood shortage, then 1 by 1 is better.

Correct

8 minutes ago, BreakfastBurrito_007 said:

make women to reduce excess food, then batches are better

I am not sure if spamming women just to get rid of your extra food is necessary; if your batch size is less than 27 you will still lose out to autotrain while floating food. 

Just let it float or research mining techs or perhaps switch to cavalry? In late game, floating food is very good because most people consume more food than they can produce in late game, especially if there are sustained raids. 

 

[BreakfastBurrito_007]

1 by 1 clearly has the fastest return on investment according to the above math. But its very important that you re-invest that return in order for the extra ROI to be impactful. Because of this I recommend 1 by 1 only for periods of time where you can avoid floating resources. Keep in mind that as you get more units your income will surpass expenditure after some time so its important to invest these extra resources into something (a house, an extra barracks, a tech).