Matrix Games Forums

Forums  Register  Login  Photo Gallery  Member List  Search  Calendars  FAQ 

My Profile  Inbox  Address Book  My Subscription  My Forums  Log Out

RE: I need Help - AI Testing

 
View related threads: (in this forum | in all forums)

Logged in as: Guest
Users viewing this topic: none
  Printable Version
All Forums >> [New Releases from Matrix Games] >> War in the Pacific: Admiral's Edition >> Scenario Design and Modding >> RE: I need Help - AI Testing Page: <<   < prev  1 [2] 3 4 5   next >   >>
Login
Message << Older Topic   Newer Topic >>
RE: I need Help - AI Testing - 3/9/2016 11:13:02 PM   
PaxMondo


Posts: 7878
Joined: 6/6/2008
Status: offline
1/14/42 ... allies just took Kluang ... so another week before they arrive in Singers ...
Georgetwon still not taken, but units have arrived. Likely fall tomorrow ...
Sambas fell 1/13/42 ... so maybe Singkawang in the next week.
Manila and Clark both under attack (AI split forces). Both survived first attack, but should fall by the 3rd one ...
Mindinao has essentially fallen ...

Balikpapn fell on 1/10/42.
Tarakan landing today.

No signs of IJ in Burma yet ...



_____________________________

Pax

(in reply to DD696)
Post #: 31
RE: I need Help - AI Testing - 3/10/2016 12:10:02 AM   
Andy Mac

 

Posts: 14121
Joined: 5/12/2004
From: Alexandria, Scotland
Status: offline
Sorry DD696 I am testing the Scenarios above4 and specifically the new AI script tha\t I am developing to act as a baseline for future AI mods so I am looking for testers to run out c 6 months of one of the scenarios above to test out the new AI script.

Unfortuantely that means existing mod games are not really going to help at this point because of the changes I am tryin gto make to improve the Ai

In essence the Ironman series of scenarios are basically v AI only to try and mitigate AI disadvantages

Andy

quote:

ORIGINAL: DD696

Hi Andy,

I am willing to help. On 23 Feb I restarted a heavily modified version of RA 7.9 and finished turn 135 today, so that is 16 days to do 135 turns. I'm turning out turns slower than normal for a good while now, but today I have completed (hopefully) my chemotherapy sessions and will begin to end this downhill slide.

I tend to play a passive game (Allied player) against the AI at first. I learned long ago if you defeat the Empire's navy around Port Morsby the game gets kind of dull after than. I withdraw as much as possible from Malaya to carry on the war in Burma. There are times when I get a frustrated at the bull headed AI and give them a good pounding and have some fun, then I go back a few turns and let it do what it wants with minimal opposition.

In order to help the AI in my mod I do replace all Japanese land units with the ones from scenario 1 as that is what the AI is expecting to find (as well as add in the additional ships), and I get to play with all the new toys. I beef up HI, LI, oil, fuel and resource production in Japan/Korea/Manchuria in order to make it as self sufficient as possible. I don't like having to play on hard or extra hard in order to boost these resources in the pools. I monitor Japan with tracker (and turn off Japanese garrison requirements) to prevent Japan taking major hits when I have to go in and tinker with something on the Japanese side.

I haven't played any stock scenarios in years and find them rather dull after trying out some of the good mods out there. If you think I would be of any use in your testing kindly let me know.

In my humble opinion, there should be scenarios made specifically for the AI player. This would make for a much better game and mod developers wouldn't be jumping thru hoops trying to have a scenario that works for both.


(in reply to DD696)
Post #: 32
RE: I need Help - AI Testing - 3/10/2016 12:10:52 AM   
Andy Mac

 

Posts: 14121
Joined: 5/12/2004
From: Alexandria, Scotland
Status: offline
So it seems to be progressing ok without massive losses the only challenge will be has the pace allowed you time to make java impregnable will be interesting to see

Thanks

Andy
quote:

ORIGINAL: PaxMondo

1/14/42 ... allies just took Kluang ... so another week before they arrive in Singers ...
Georgetwon still not taken, but units have arrived. Likely fall tomorrow ...
Sambas fell 1/13/42 ... so maybe Singkawang in the next week.
Manila and Clark both under attack (AI split forces). Both survived first attack, but should fall by the 3rd one ...
Mindinao has essentially fallen ...

Balikpapn fell on 1/10/42.
Tarakan landing today.

No signs of IJ in Burma yet ...




(in reply to PaxMondo)
Post #: 33
RE: I need Help - AI Testing - 3/10/2016 12:17:13 AM   
DD696

 

Posts: 717
Joined: 7/9/2004
From: near Savannah, Ga
Status: offline
That's fine, Andy.

I was willing to test out your scenario 2 (as you have it now) and was just informing you of what I have been doing to improve the AI for my personal play. I simply thought that my experience with the AI to date may have been of some benefit to you.

Hope you can improve it more. It will benefit many players of this great game.

_____________________________

USMC: 1970-1977.

(in reply to Andy Mac)
Post #: 34
RE: I need Help - AI Testing - 3/10/2016 2:29:38 AM   
PaxMondo


Posts: 7878
Joined: 6/6/2008
Status: offline
Not planning to defend in Java as I know it would break the AI too soon.

Near term objectives (early '42):

Will defend North Burma and then push down. See how AI reacts to that. It will be slow as '42 Allies just don't have enough of anything. But Burma is a good place to fight for me as allies.

Then defend OZ and take back any territory lost.

Secure Aleutians.

Secure Line Islands.



1/15/42- Rabaul landing. 1st attack repulsed.
1/16/42 - Georgetown falls.
1/18/42 - JI arrives in Singers ... repulsed.
1/20/42 - Rabaul falls. Singers falls. Tarakan falls.



quote:

ORIGINAL: Andy Mac

So it seems to be progressing ok without massive losses the only challenge will be has the pace allowed you time to make java impregnable will be interesting to see

Thanks

Andy
quote:

ORIGINAL: PaxMondo

1/14/42 ... allies just took Kluang ... so another week before they arrive in Singers ...
Georgetwon still not taken, but units have arrived. Likely fall tomorrow ...
Sambas fell 1/13/42 ... so maybe Singkawang in the next week.
Manila and Clark both under attack (AI split forces). Both survived first attack, but should fall by the 3rd one ...
Mindinao has essentially fallen ...

Balikpapn fell on 1/10/42.
Tarakan landing today.

No signs of IJ in Burma yet ...







< Message edited by PaxMondo -- 3/10/2016 2:34:04 AM >


_____________________________

Pax

(in reply to Andy Mac)
Post #: 35
RE: I need Help - AI Testing - 3/10/2016 1:17:37 PM   
Andy Mac

 

Posts: 14121
Joined: 5/12/2004
From: Alexandria, Scotland
Status: offline
Apologies I misunderstood your email.

Any help on any of the three scens posted is most welcome and different play styles as well I need the broadest set of feedback on the new baseline AI.

Basically I want as many insights as possible on the AI.

I am trying to improve it - reality is it may never be fully competitive for more than 9 months in Scen 1 but if I can pinpoint why it fails better I can continue to improve the performance.

Andy
quote:

ORIGINAL: DD696

That's fine, Andy.

I was willing to test out your scenario 2 (as you have it now) and was just informing you of what I have been doing to improve the AI for my personal play. I simply thought that my experience with the AI to date may have been of some benefit to you.

Hope you can improve it more. It will benefit many players of this great game.


(in reply to DD696)
Post #: 36
RE: I need Help - AI Testing - 3/10/2016 1:19:26 PM   
Andy Mac

 

Posts: 14121
Joined: 5/12/2004
From: Alexandria, Scotland
Status: offline
Thanks Pax actually play as you want I want to see what works and doesnt work.

Re new bases how are they working out ??

quote:

ORIGINAL: PaxMondo

Not planning to defend in Java as I know it would break the AI too soon.

Near term objectives (early '42):

Will defend North Burma and then push down. See how AI reacts to that. It will be slow as '42 Allies just don't have enough of anything. But Burma is a good place to fight for me as allies.

Then defend OZ and take back any territory lost.

Secure Aleutians.

Secure Line Islands.



1/15/42- Rabaul landing. 1st attack repulsed.
1/16/42 - Georgetown falls.
1/18/42 - JI arrives in Singers ... repulsed.
1/20/42 - Rabaul falls. Singers falls. Tarakan falls.



quote:

ORIGINAL: Andy Mac

So it seems to be progressing ok without massive losses the only challenge will be has the pace allowed you time to make java impregnable will be interesting to see

Thanks

Andy
quote:

ORIGINAL: PaxMondo

1/14/42 ... allies just took Kluang ... so another week before they arrive in Singers ...
Georgetwon still not taken, but units have arrived. Likely fall tomorrow ...
Sambas fell 1/13/42 ... so maybe Singkawang in the next week.
Manila and Clark both under attack (AI split forces). Both survived first attack, but should fall by the 3rd one ...
Mindinao has essentially fallen ...

Balikpapn fell on 1/10/42.
Tarakan landing today.

No signs of IJ in Burma yet ...








(in reply to PaxMondo)
Post #: 37
RE: I need Help - AI Testing - 3/10/2016 1:35:11 PM   
PaxMondo


Posts: 7878
Joined: 6/6/2008
Status: offline
1/22/42 - Moulmein falls. IJ in Burma.

1/25/42 - Makassar and Kendari fall.

1/29/42 Pegu falls.


_____________________________

Pax

(in reply to PaxMondo)
Post #: 38
RE: I need Help - AI Testing - 3/10/2016 5:38:36 PM   
dr. smith

 

Posts: 221
Joined: 6/30/2004
From: lost in space
Status: offline
I've actually been playing your Scenario 33 since December, which I think was a precursor? How much change to the AI scripts did you make recently?

I'm in May of '42. Didn't try any fortress Java, but I've had Japan blocked at Moulmein for awhile. Batavia fell on 22 April, Soerbaja's about to. Only thing left in DEI is Sabang.

China's been very quiet.

Can get you more details if this version is of any help.

(in reply to PaxMondo)
Post #: 39
RE: I need Help - AI Testing - 3/10/2016 8:35:43 PM   
Yaab


Posts: 2748
Joined: 11/8/2011
From: Poznan, Poland
Status: offline
AndyMac, I think I found a data error.

In scen033 and scen100, almost all xAK/AK/xAP/support ships in Pearl Harbor on December 7, have unusually high crew experience in range of 50/10 for day/night. In scen001 the experience rating of these ships matches the experience of other USN and Allied ships, which is on average 15/10.


For example, AK Alchiba is 13/11 in scen 1, and 53/12 in scen 33 and 100.

OthersxAK:Laida, Florence D
AG: Antares, Aries
xAKLs: Manini, Hirondelle
xAP: St Mihael

and so on, so forth.

(in reply to dr. smith)
Post #: 40
RE: I need Help - AI Testing - 3/10/2016 10:05:26 PM   
Andy Mac

 

Posts: 14121
Joined: 5/12/2004
From: Alexandria, Scotland
Status: offline
Overall impressions and any obvious errors would be good its probably 3 generations of AI script backwards I have learned a few new tricks I am trying to incorporate now so overall issues would still be god

Thanks

Andy

(AI will always struggle in China for both sides AI is not good at continental warfare and doesn't handle non base hex defence very well so I tend to assume China will be left quiet by a player v the AI
quote:

ORIGINAL: dr. smith

I've actually been playing your Scenario 33 since December, which I think was a precursor? How much change to the AI scripts did you make recently?

I'm in May of '42. Didn't try any fortress Java, but I've had Japan blocked at Moulmein for awhile. Batavia fell on 22 April, Soerbaja's about to. Only thing left in DEI is Sabang.

China's been very quiet.

Can get you more details if this version is of any help.



(in reply to dr. smith)
Post #: 41
RE: I need Help - AI Testing - 3/10/2016 10:05:53 PM   
Andy Mac

 

Posts: 14121
Joined: 5/12/2004
From: Alexandria, Scotland
Status: offline

Many thanks will investigate

quote:

ORIGINAL: Yaab

AndyMac, I think I found a data error.

In scen033 and scen100, almost all xAK/AK/xAP/support ships in Pearl Harbor on December 7, have unusually high crew experience in range of 50/10 for day/night. In scen001 the experience rating of these ships matches the experience of other USN and Allied ships, which is on average 15/10.


For example, AK Alchiba is 13/11 in scen 1, and 53/12 in scen 33 and 100.

OthersxAK:Laida, Florence D
AG: Antares, Aries
xAKLs: Manini, Hirondelle
xAP: St Mihael

and so on, so forth.


(in reply to Yaab)
Post #: 42
RE: I need Help - AI Testing - 3/10/2016 11:03:21 PM   
traskott

 

Posts: 1333
Joined: 6/23/2008
From: Valladolid, Spain
Status: offline
Andy, count on me to test it. Email me at traskott-dot-gmail.com and will run anytest u need.

(in reply to Andy Mac)
Post #: 43
RE: I need Help - AI Testing - 3/11/2016 7:48:15 AM   
PaxMondo


Posts: 7878
Joined: 6/6/2008
Status: offline
2/1/42 - Rangoon falls.
2/5/42 - Clark falls.
2/7/42 IJ invades Java @ Kalidjati
2/9/42 - Manila falls.

< Message edited by PaxMondo -- 3/11/2016 1:29:04 PM >


_____________________________

Pax

(in reply to traskott)
Post #: 44
RE: I need Help - AI Testing - 3/11/2016 10:34:58 AM   
Yaab


Posts: 2748
Joined: 11/8/2011
From: Poznan, Poland
Status: offline

quote:

ORIGINAL: Andy Mac


Many thanks will investigate

quote:

ORIGINAL: Yaab

AndyMac, I think I found a data error.

In scen033 and scen100, almost all xAK/AK/xAP/support ships in Pearl Harbor on December 7, have unusually high crew experience in range of 50/10 for day/night. In scen001 the experience rating of these ships matches the experience of other USN and Allied ships, which is on average 15/10.


For example, AK Alchiba is 13/11 in scen 1, and 53/12 in scen 33 and 100.

OthersxAK:Laida, Florence D
AG: Antares, Aries
xAKLs: Manini, Hirondelle
xAP: St Mihael

and so on, so forth.




I took scen100 dat file and renamed it scen28 dat file for the purpose DaBigBabes C game. Seems the contents of scen100 dat file
changed the experience levels of the Pearl Harbor ships in the DaBigBabes C game as well - there is now average 50/10 experience instead of average 15/10 experience. The ships' experience levels in Editor are 0/0 day/ night so I guess, it is the dat file that assigns exp levels to the ships at the scenario's start.

This bug is a show-stopper for me right now.

(in reply to Andy Mac)
Post #: 45
RE: I need Help - AI Testing - 3/11/2016 12:17:32 PM   
Andy Mac

 

Posts: 14121
Joined: 5/12/2004
From: Alexandria, Scotland
Status: offline
Weird wasnt aware it could do that I will investigate

(in reply to Yaab)
Post #: 46
RE: I need Help - AI Testing - 3/11/2016 1:57:15 PM   
Andy Mac

 

Posts: 14121
Joined: 5/12/2004
From: Alexandria, Scotland
Status: offline
Great thnaks this is what i dont get from AI v AI testing.

the AI sends 18th Brit Div to Singapore most of the time so Malaya takes longer which delays everything else.

The AI is actually moving pretty fast now - is it giving you many opportunities to massacre convoys etc ??


quote:

ORIGINAL: PaxMondo

2/1/42 - Rangoon falls.
2/5/42 - Clark falls.
2/7/42 IJ invades Java @ Kalidjati
2/9/42 - Manila falls.


(in reply to PaxMondo)
Post #: 47
RE: I need Help - AI Testing - 3/11/2016 1:58:12 PM   
Andy Mac

 

Posts: 14121
Joined: 5/12/2004
From: Alexandria, Scotland
Status: offline
How strong was the landing on java and has it covered the Invasion with CV and or SCTF's ??

(in reply to Andy Mac)
Post #: 48
RE: I need Help - AI Testing - 3/11/2016 6:46:12 PM   
ny59giants


Posts: 8900
Joined: 1/10/2005
Status: offline
Is it possible for the AI to use more Fast Transport TFs? Against a human opponent its hard for the Allies to combat them as they are in and out so quickly. The warships can bombard the base before the troops land and then are like a SC TF afterwards.

_____________________________


(in reply to Andy Mac)
Post #: 49
RE: I need Help - AI Testing - 3/11/2016 7:53:17 PM   
Andy Mac

 

Posts: 14121
Joined: 5/12/2004
From: Alexandria, Scotland
Status: offline
I have no way of specifying for a script to use FT TF unfortunately wish I could but cannot

(in reply to ny59giants)
Post #: 50
RE: I need Help - AI Testing - 3/11/2016 11:30:10 PM   
PaxMondo


Posts: 7878
Joined: 6/6/2008
Status: offline

quote:

ORIGINAL: Andy Mac

Great thnaks this is what i dont get from AI v AI testing.

the AI sends 18th Brit Div to Singapore most of the time so Malaya takes longer which delays everything else.

The AI is actually moving pretty fast now - is it giving you many opportunities to massacre convoys etc ??


quote:

ORIGINAL: PaxMondo

2/1/42 - Rangoon falls.
2/5/42 - Clark falls.
2/7/42 IJ invades Java @ Kalidjati
2/9/42 - Manila falls.



At this point, no. Too many CV TF's that I can't compete against without putting the USN into the DEI which I don't like to do so early.

And yes, the pace now isn't too bad. If you can increase the pace of Mili/Kuching/Sinkawang with either Kuching or better Sinkawang being a air base with Netties, I would say the early going is pretty good.


_____________________________

Pax

(in reply to Andy Mac)
Post #: 51
RE: I need Help - AI Testing - 3/12/2016 7:50:02 AM   
Yaab


Posts: 2748
Joined: 11/8/2011
From: Poznan, Poland
Status: offline
High exp ships again.

I tried to change AK Alchiba exp to 11/11 (this is the value I got from Tracker for an RHS game, which has no high-exp bug for Pearl Harbor ships). After reloding the turn the ship's value changed from 52/4 to 52/11 in scen 100.

Something is mixing the exp values for ships in scen33 and scen100, which in Editor have exp 0/0 and start the game in Pearl Harbor. Other ships are unaffected, like xAKL Honomu which has 0/0 exp in Editor and starts the game in San Diego and has 9/11 exp.

EDIT: this is the info from Editor's Manual

3.3.1.1 A Note on Experience (Exp Night and Exp Day Ratings)
Note that Ships given a zero (0) in either Exp Night or Exp Day will have that Experience rating set
from a table.
 Japanese and British Exp Day ratings will be the table value plus a random number from
0-10.
 For all other Allied ships Exp Day ratings will be 4/5 of the table value in 1941 and 1942,
and the table value in 1943 and beyond, plus a random number from 0-10.
 Japanese Exp Night ratings will be the table value plus a random number from 0-5.
 Allied Exp Night ratings for Submarines and PT Boats will be the table value plus a
random number from 0-10.
 For all other Allied ship Exp Night ratings, the rating will be a percentage of the table value plus a random number from 0-10. The percentage increases year by year.

-----------

There is a table added. AK ships have value 10. How come it chnages to 50 in scen033 and scen100 on turn 2???


< Message edited by Yaab -- 3/12/2016 8:37:44 AM >

(in reply to PaxMondo)
Post #: 52
RE: I need Help - AI Testing - 3/12/2016 10:36:46 AM   
Yaab


Posts: 2748
Joined: 11/8/2011
From: Poznan, Poland
Status: offline
AnydMac, your dat files are OK. I just posted this exp 50 thing in the tech forum. Seems the code gives huge exp gains for ships with initial 10/10 exp who are disbanded in a port during an air attack.

(in reply to Yaab)
Post #: 53
RE: I need Help - AI Testing - 3/12/2016 2:10:20 PM   
Yaab


Posts: 2748
Joined: 11/8/2011
From: Poznan, Poland
Status: offline
I think there is a problem with additional bases in Burma.

Since LCU numbers in Burma stayed the same there is not enough LCUs to guard the new bases on Rangoon - Lashio rai line (Burma Road). Granted, you will have enough units if you keep the 1st Burma Division split, but I always combine this unit, which is still a weak LCU even if combined. There should be some local police battalions to help guard the new bases from paratroopes and "rowboat corps".

Consider waht happens if frontline is stabilised 1 hex east or Rangoon-Lashio rail. The empty bases on the rail line may auto-revert to Japanese thanks to their "rowboat corps".

(in reply to Yaab)
Post #: 54
RE: I need Help - AI Testing - 3/12/2016 8:11:12 PM   
PaxMondo


Posts: 7878
Joined: 6/6/2008
Status: offline
14Feb42 - Oosthaven falls.
19FEb42 - Sabang falls.
22Feb42 - Mandalay falls.

_____________________________

Pax

(in reply to Yaab)
Post #: 55
RE: I need Help - AI Testing - 3/12/2016 8:13:33 PM   
PaxMondo


Posts: 7878
Joined: 6/6/2008
Status: offline

quote:

ORIGINAL: Yaab

I think there is a problem with additional bases in Burma.

Since LCU numbers in Burma stayed the same there is not enough LCUs to guard the new bases on Rangoon - Lashio rai line (Burma Road). Granted, you will have enough units if you keep the 1st Burma Division split, but I always combine this unit, which is still a weak LCU even if combined. There should be some local police battalions to help guard the new bases from paratroopes and "rowboat corps".

Consider waht happens if frontline is stabilised 1 hex east or Rangoon-Lashio rail. The empty bases on the rail line may auto-revert to Japanese thanks to their "rowboat corps".

New bases have no garrison requirements.
They are needed to give the AI the ability to progress, otherwise it doesn't move well (like china).

So, the choice is: get the AI to fight better in Burma (and add the bases needed for that), or don't fight in Burma (and have no additional bases).

Me? I prefer the ability to have the AI fight in Burma.


_____________________________

Pax

(in reply to Yaab)
Post #: 56
RE: I need Help - AI Testing - 3/12/2016 8:18:59 PM   
Yaab


Posts: 2748
Joined: 11/8/2011
From: Poznan, Poland
Status: offline
You need to hold these bases for the Burma Road to be open. If one of them reverts to Jap, this closes the Burma Road until you recapture the lost base/hex. That is my only concern.

(in reply to PaxMondo)
Post #: 57
RE: I need Help - AI Testing - 3/12/2016 9:05:10 PM   
PaxMondo


Posts: 7878
Joined: 6/6/2008
Status: offline

quote:

ORIGINAL: Yaab

You need to hold these bases for the Burma Road to be open. If one of them reverts to Jap, this closes the Burma Road until you recapture the lost base/hex. That is my only concern.

Understood completely. But this is a setup for AI games ... which do you want? Without the bases, the AI under-performs. With them, you get much more effective AI advances.
Does it create a little more challenge for the human player? Sure. I'm not seeing that as a bad thing .... just my opinion though.



_____________________________

Pax

(in reply to Yaab)
Post #: 58
RE: I need Help - AI Testing - 3/12/2016 11:01:58 PM   
PaxMondo


Posts: 7878
Joined: 6/6/2008
Status: offline
25Feb42 - Bataan seige starts
1 Mar Schwebo falls.



_____________________________

Pax

(in reply to PaxMondo)
Post #: 59
RE: I need Help - AI Testing - 3/14/2016 12:07:46 PM   
Andy Mac

 

Posts: 14121
Joined: 5/12/2004
From: Alexandria, Scotland
Status: offline
Thanks much appreciated

Yaab re Burma I wanted to force the Allied player to some unpalatable choices insufficinet units to cover every base I think is a good challenge to face !!!

(in reply to PaxMondo)
Post #: 60
Page:   <<   < prev  1 [2] 3 4 5   next >   >>
All Forums >> [New Releases from Matrix Games] >> War in the Pacific: Admiral's Edition >> Scenario Design and Modding >> RE: I need Help - AI Testing Page: <<   < prev  1 [2] 3 4 5   next >   >>
Jump to:





New Messages No New Messages
Hot Topic w/ New Messages Hot Topic w/o New Messages
Locked w/ New Messages Locked w/o New Messages
 Post New Thread
 Reply to Message
 Post New Poll
 Submit Vote
 Delete My Own Post
 Delete My Own Thread
 Rate Posts


Forum Software © ASPPlayground.NET Advanced Edition 2.4.5 ANSI

0.137