AI too obstinate in sending his all his Corps despite the odds?

[Drakken]

As an aside, if you look at the Luxemburg picture it seems the google in the middle of the battle screen isn't centered on the region where the battle is located.

IIRC, isn't it supposed to be fixed with the 1.806E patch?

[calvinus]

Ok, so let's consider that all these values are used to calculated a balance between your losses and enemy losses.

So taking your first example:

France:
5 Wounded (LR1) : 1.0 x 5 = 5.0
1 Shaken (LR3) : 0.0 x 1 = 0.0
4 Disorganized (LR2): 2.0 x 4 = 8.0
1 Out of combat (LR5) : 3.0 x 1 = 3.0
1 Destroyed (LR4) : 3.0 x 1 = 3.0
TOT FRANCE = 5.0 + 0.0 + 8.0 + 3.0 + 3.0 = 19.0

Germany:
2 Shaken (LR1) : 0.0 x 1 = 0.0
2 Disorganized (LR2): 2.0 x 2 = 4.0
1 Out of combat (LR5) : 3.0 x 1 = 3.0
1 Destroyed (LR4) : 3.0 x 1 = 3.0
TOT GERMANY = 0.0 + 4.0 + 3.0 + 3.0 = 10.0

BALANCE = TOT GERMANY - TOT FRANCE = 10.0 - 19.0 = -9.0

Look here:

LR0 8.0 1 Land retreat threshold: if difference between friendly summary tactical situation and enemy summary tactical situation is superior or equal to this value, a retreat is attempted. Tactical situation is calculated using lines from LR1 to LR6.

That means the AI retreats as soon the negative balance is lower than -8.0.
Indeed your AI retreated as soon he reached -9.0...

[rattlesnake]

I am sure, here is what I have when I open the AIParamsTable.csv file
LR1;1.0;1;Land retreat tactical situation calculation: multiplier applied to each wounded unit
LR2;2.0;1;Land retreat tactical situation calculation: multiplier applied to each disorganized unit
LR3;0.0;1;Land retreat tactical situation calculation: multiplier applied to each shaken unit
LR4;3.0;1;Land retreat tactical situation calculation: multiplier applied to each destroyed unit
LR5;3.0;1;Land retreat tactical situation calculation: multiplier applied to each out-of-combat unit
LR6;4.0;1;Land retreat tactical situation calculation: multiplier applied to each unit in panic :

what are you going to say ?The parameters are related with the RP?

[calvinus]

As an aside, if you look at the Luxemburg picture it seems the google in the middle of the battle screen isn't centered on the region where the battle is located.

IIRC, isn't it supposed to be fixed with the 1.806E patch?

It happens once in a while, if you go to Breakthrough/Retreat sub-window, then the map is properly updated.

[calvinus]

what are you going to say ?The parameters are related with the RP?

The row that refers to RPs is:

LR7 3.0 1 Land retreat Recruit Points threshold: if RP level is inferior or equal to a random integer between 1 and this value, a retreat is attempted.

[rattlesnake]

Ok, so let's consider that all these values are used to calculated a balance between your losses and enemy losses.

So taking your first example:



France:
5 Wounded (LR1) : 1.0 x 5 = 5.0
1 Shaken (LR3) : 0.0 x 1 = 0.0
4 Disorganized (LR2): 2.0 x 4 = 8.0
1 Out of combat (LR5) : 3.0 x 1 = 3.0
1 Destroyed (LR4) : 3.0 x 1 = 3.0
TOT FRANCE = 5.0 + 0.0 + 8.0 + 3.0 + 3.0 = 19.0

Germany:
2 Shaken (LR1) : 0.0 x 1 = 0.0
2 Disorganized (LR2): 2.0 x 2 = 4.0
1 Out of combat (LR5) : 3.0 x 1 = 3.0
1 Destroyed (LR4) : 3.0 x 1 = 3.0
TOT GERMANY = 0.0 + 4.0 + 3.0 + 3.0 = 10.0

BALANCE = TOT GERMANY - TOT FRANCE = 10.0 - 19.0 = -9.0

Look here:


That means the AI retreats as soon the negative balance is lower than -8.0.
Indeed your AI retreated as soon he reached -9.0...

So France 12 vs German 6?
From that formula,the more the troops ,the easier they will retreat?(if all of them participate in the battle)
IF there is only one troop vs 12 troops,it will never retreat?

[calvinus]

So France 12 vs German 6

No, France 19 "AI loss points" vs Germay 10 "AI loss points". Net balance -9.

[calvinus]

Other factors:
AI fights to the death when defending a colony or a target of Grand Offensive.
AI is more willing to retreat when outnumbered by the enemy (numerical superiority).

[rattlesnake]

Other factors:
AI fights to the death when defending a colony or a target of Grand Offensive.
AI is more willing to retreat when outnumbered by the enemy (numerical superiority).

It is complex.Only the AI developer knows what it is .
Here comes anothor situation:
1 elite with attack of 6 ,defend of 7,movement of 6
VS
3 mobilized with attack of 1,defend of 2,movement of 2
will it retreat?

[Drakken]

Other factors:
AI fights to the death when defending a colony or a target of Grand Offensive.
AI is more willing to retreat when outnumbered by the enemy (numerical superiority).

Odd, as I have seen the AI send waves after waves of attack on a numerically superior enemy force.

What are the AiParam factor lines for:
- Retreating when the defender has the numerically superior force modifier?
- When the attacker has a net disadvantage position (many of my attackers had a -2 vs 4 differiencial because of their respective position, but I can let it pass since these were scripted offensives)?

[calvinus]

Odd, as I have seen the AI send waves after waves of attack on a numerically superior enemy force.

What are the AiParam factor lines for:
- Retreating when the defender has the numerically superior force modifier?
- When the attacker has a net disadvantage position (many of my attackers had a -2 vs 4 differiencial because of their respective position, but I can let it pass since these were scripted offensives)?

Evaluation of Numerical Superiority and Terrain factor are hard coded, no moddable parameter, sorry.

[Drakken]

Evaluation of Numerical Superiority and Terrain factor are hard coded, no moddable parameter, sorry.

Well, in that case I can only leave it to you to see if there's something that can be tweaked.

What I can attempt, however, is to tweak LR1 slightly up to 1.5, or even all the way to 2.0, and see if the AI will tend to retreat if there are too many flipped (wounded) Corps compared to what was inflicted to the Defender.

My aim here is that the AI should be less likely to continue if it has too many wounded Corps compared to lesser enemy casualties, but more likely to continue if the defender has sustained significant casualties.

Each flipped Corps vs non-flipped Corps would make it likelier that the AI will retreat to preserve its force, but will continue longer if it inflicts similar damage to the opponent (which would keep the balance of LR1 casualties equal, caeteris paribus, at 1-for-1, but increase the importance of LR1 imbalances).

[calvinus]

Now your losses and enemy losses are counted using the same factors.
I'm wondering if we should introduce a multiplier/divider instead...

Example:

your "AI loss points" 13
enemy "AI loss points" 10 multiplied by Y

If Y = 0.5 (so halved down), 10 becomes 5.
Thus, 5 - 13 = -8 that means retreat!

[Drakken]

Now your losses and enemy losses are counted using the same factors.
I'm wondering if we should introduce a multiplier/divider instead...

Example:

You mean for LR1?

I'm not sure I understand.

[calvinus]

You can also reduce the LR0 threshold from 8.0 to... let's say... 6.0.

[calvinus]

You mean for LR1?

I'm not sure I understand.

Read carefully post #32 and you'll see that AI calculates the two sums (for the balance) using the same params.

[Drakken]

You can also reduce the LR0 threshold from 8.0 to... let's say... 6.0.

I like 8.0, as it is, as threshold for Retreat. At 6.0, I feel that maybe the AI will start to retreat too easily!

[Drakken]

Read carefully post #32 and you'll see that AI calculates the two sums (for the balance) using the same params.

Ah, you mean that instead, it would be (LR4+LR5/2).

Gotcha.

This is really a fine-tuning issue, really. I fear putting any divider because I want to avoid seeing the AI retreating too fast, nor do I want to see one of the condition take too much weight.

[rattlesnake]

Now your losses and enemy losses are counted using the same factors.
I'm wondering if we should introduce a multiplier/divider instead...

Example:

What is the meaning of Y in the war?
I don't think there is any need for that.

[Drakken]

So, in short, want me to test the following AI settings?

LR1;1.5;1;Land retreat tactical situation calculation: multiplier applied to each wounded unit
LR2;2.0;1;Land retreat tactical situation calculation: multiplier applied to each disorganized unit
LR3;0.0;1;Land retreat tactical situation calculation: multiplier applied to each shaken unit
LR4;3.0;1;Land retreat tactical situation calculation: multiplier applied to each destroyed unit
LR5;3.0;1;Land retreat tactical situation calculation: multiplier applied to each out-of-combat unit
LR6;4.0;1;Land retreat tactical situation calculation: multiplier applied to each unit in panic

AND

LR0 set from 8.0 to 6.0

[rattlesnake]

Well, in that case I can only leave it to you to see if there's something that can be tweaked.

My aim here is that the AI should be less likely to continue if it has too many wounded Corps compared to lesser enemy casualties, but more likely to continue if the defender has sustained significant casualties.

Yes,the formula indicates that.Even one side has many wounded corps,but it gets the information that the enemy also has many casualties,so it will sustain the battle.It is true in the real war.

[calvinus]

Yes,the formula indicates that.Even one side has many wounded corps,but it gets the information that the enemy also has many casualties,so it will sustain the battle.It is true in the real war.

The AI actually works exactly so.

[rattlesnake]

Ah, you mean that instead, it would be (LR4+LR5/2).

Gotcha.

This is really a fine-tuning issue, really. I fear putting any divider because I want to avoid seeing the AI retreating too fast, nor do I want to see one of the condition take too much weight.

No.
He says that the loses of offensive side and the loses of the defensive side should not be the same. There should be a parameters to multiply.
Not the value of LR4 and LR5.

[Drakken]

Yes,the formula indicates that.Even one side has many wounded corps,but it gets the information that the enemy also has many casualties,so it will sustain the battle.It is true in the real war.

Hence why I want to increase the factor. The more information that the attack has been repelled and ineffective, the more likely the attack should stop before there are too much damage.

[rattlesnake]

The AI actually works exactly so.

That makes the meat grinder,for example battle of Verdun. Then the battle will be very fierce,that does not make any sense,if it is not battle for a strategy place.
And actually whether it retreats depends on the decision of the general.

[Tamas]

I would be wary of changing these values too much...

I feel the AI to be just about right...
Especially in movement war, and in trench war the problem right now that it can be too careful... Let's wait on calvinus' fix on that one.

[calvinus]

In 1.08G I added a new moddable parameter:

LRA 0.8 1 Land retreat threshold modifier: multiplier applied to enemy losses summary value before checking LR0

It means the "AI loss points" of the opponent are counted only 80%

Let's see if with 1.08G AI improves to your taste.

[Drakken]

In 1.08G I added a new moddable parameter:



It means the "AI loss points" of the opponent are counted only 80%

Let's see if with 1.08G AI improves to your taste.

If I understand correctly, it means that whatever the AI sums up in casualties, it will be multiplied by a factor of 0.8 against the player?

What would happen a AI vs AI fight?

[rattlesnake]

If I understand correctly, it means that whatever the AI sums up in casualties, it will be multiplied by a factor of 0.8 against the player?

What would happen a AI vs AI fight?

The opponent of the AI's loss point will be multiplied by a factor of 0.8.Thus,If you play with the AI ,your lose point will be multiplied by a factor of 0.8.Your loss point is lowered.You will see the AI is more likely to sustain the war.That is what you want.

[Drakken]

Well, tested a little the following settings:

LR1;1.5;1;Land retreat tactical situation calculation: multiplier applied to each wounded unit
LR2;2.0;1;Land retreat tactical situation calculation: multiplier applied to each disorganized unit
LR3;0.0;1;Land retreat tactical situation calculation: multiplier applied to each shaken unit
LR4;3.0;1;Land retreat tactical situation calculation: multiplier applied to each destroyed unit
LR5;3.0;1;Land retreat tactical situation calculation: multiplier applied to each out-of-combat unit
LR6;4.0;1;Land retreat tactical situation calculation: multiplier applied to each unit in panic

AND

LR0 set from 8.0 to 6.0

And I didn't see the end of the world. French pursued and took Thionville and Mauberge, in Thionville the Germans defender retreated after a few turns, just a little beyond it's Obstinacy rating.