What to do with CHS

[Don Bowen]

Tests have shown that CHS, with all the latest mods, has confused the AI. I am going to run a test today in Human-vs-AI but I expect to significant improvement.

I am going to release CHS in one of it's two current forms (see below) for Head-to-Head play only. I will then begin an extensive series of experiments trying to figure out what has so confused the AI.

In testing I made a second version of CHS that might have some value in Head-to-Head. The two versions are:

1. 18th Japanese Division moved from Taan to Canton to slow down Japanese invasion of Malaya.

2. 18th Japanese Division left at Taan with load orders for Kota Bharu, approximately 32 additional transports/cargo ships assigned to initial invasion task forces to compensate for carrying capacity reductions. This allows the entire 5th, 18th, 48th Divisions and 65th Brigade to be loaded.

Let me know what you think.

Don

[Bradley7735]

Hi Don,

How hard would it be to reverse any changes made to japanese units that start the game on 12/7? I think changes to the existing ships and/or LCU's may be what's causing the problems. Maybe by reversing some of the changes (ship capacity/ship slots/LCU slots/???) the AI will be able to take over the DEI by May?

Can you compare your scenario files with the Stock 15 files and see if there are major differences between the starting units?

Maybe the LCU's need to stay in the stock 15 slots (not sure if you've changed them around or not. Maybe the starting ships need to be in the same slots as well?)

bc

[Ron Saueracker]

Maybe the LCU's need to stay in the stock 15 slots (not sure if you've changed them around or not. Maybe the starting ships need to be in the same slots as well?)

If this is the case I think whomever is responsible for "not making this known" should be hauled into the town centre and pelted with rocks and garbage.[X(][:@]

[Don Bowen]

ORIGINAL: Bradley7735

Hi Don,

How hard would it be to reverse any changes made to japanese units that start the game on 12/7? I think changes to the existing ships and/or LCU's may be what's causing the problems. Maybe by reversing some of the changes (ship capacity/ship slots/LCU slots/???) the AI will be able to take over the DEI by May?

Can you compare your scenario files with the Stock 15 files and see if there are major differences between the starting units?

Maybe the LCU's need to stay in the stock 15 slots (not sure if you've changed them around or not. Maybe the starting ships need to be in the same slots as well?)

bc

Sorry - six months worth of changes to damn near everything. A diff between Scenario 15 and CHS would highlight 90% of the entries. I've a few tests in mind to try and isolate the problem area:

1. Reset Capacity (currently in progress as it is the easiest).
2. Run CHS with V1.4 (I think I have a copy somewhere).
3. Run Tests with Japanese AI vs Human but in continous mode
4. Beer, lots of Beer.

[Bradley7735]

4. Beer, lots of Beer.

I'd buy you one, but I'm too far away.

[Ron Saueracker]

ORIGINAL: Ron Saueracker

Maybe the LCU's need to stay in the stock 15 slots (not sure if you've changed them around or not. Maybe the starting ships need to be in the same slots as well?)

If this is the case I think whomever is responsible for "not making this known" should be hauled into the town centre and pelted with rocks and garbage.[X(][:@]

You know, this is the kind of stuff which has me concerned now. I really think this needs to be cleared up officially before some really good folks potentially waste more of their time and vast effort on the editor. I remember comments earlier by some official types which warned Don, Mike etc that some of the things being done might adversely affect the game AI but no elaboration came, possibly due to company policy.

Again...Perhaps work should be suspended until all ramifications are cleared up.

[bstarr]

ORIGINAL: Ron Saueracker

ORIGINAL: Ron Saueracker

Maybe the LCU's need to stay in the stock 15 slots (not sure if you've changed them around or not. Maybe the starting ships need to be in the same slots as well?)

If this is the case I think whomever is responsible for "not making this known" should be hauled into the town centre and pelted with rocks and garbage.[X(][:@]

You know, this is the kind of stuff which has me concerned now. I really think this needs to be cleared up officially before some really good folks potentially waste more of their time and vast effort on the editor. I remember comments earlier by some official types which warned Don, Mike etc that some of the things being done might adversely affect the game AI but no elaboration came, possibly due to company policy.

Again...Perhaps work should be suspended until all ramifications are cleared up.

Back up and punt, maybe, but quit, hell no. I played CHM almost to '43 before a problem was found and I believe that one was fixed with 1.5 (disappearing Zeroes). If the problems an AI issue, then work on a HtH only version until the AI can be sorted out. By the time the AI is figured out, the "playtesting" on the HtH only version will have worked out the other kinks.

Anyway, for what it's worth, that's my opinion.

[m10bob]

Silly-ass me for thinkin' this out loud, but since the CHS modders are all WITP proponents (in one way or another),just seems good business sense for the company to be a little more open to the modders,specifically.??

[Captain Cruft]

I would say release it for H2H only now. My guess would be that most people who will want to play it are PBEMers. I myself will always have a good look at the scenario before playing it anyway.

Personally I have two remaining concerns:

1) Are all aircraft types defined in an air group so as to facilitate them appearing in the Player-Defined Upgrade list (as per thread here and in main forum)?
2) Has any explanation of the strange lack of penetration results I reported earlier come forth? If you remember these were both with ships that had the "non-standard" 5 mm armor installed.

[jwilkerson]

Well I'll have to admit - "Pry warned us" at least in a sense - because he said messin' around too much will break it.

My "hope" it that it is not 99% of the changes ... but 1 or 2 specific changes that are doing the breaking ... question is ... how to find those ... looking one by one isn't feasible unless we're very lucky ( or unless - hey Don - I don't suppose you ran this same test on CHS prior to 1.5 and had it run "clean" did you ? If so then we only have to look at changes made for 1.5 and maybe we can back all those out ... and figure out minimal way to upgrade to 1.5 retest and then it still still go forward with more changes from that point ... we need a tighter baseline ).

I for one do not intend to play against the AI ... but for me the AI still has a very important purpose ... to enable testing ... to wit the tests we've been running ... we couldn't run those tests w/o the AI ... and in modern software practice, having an automated test harness is an absolute requirement ... so we need the AI to work. Now in a normal software project the people making the changes have access to the people who wrote the test harness ( in the AI in this case ) and they can work out why the "test harness" isn't working ... not sure we can do that in this case ... so question still remains ... how do we resolvle the issue with the AI ... I think it does need to be resolved. Would I hold up planned CHS release to wait on this ? No, because we haven't really released CHS yet anyway .. first release was "Alpha" and really this is "Alpha upgraded to 1.5" ... so I'd release it anyway with the caveat that it isn't ready for serious games yet and certainly not ready for games against the AI.

[Bradley7735]

Here's a thought....

Is there a difference between running a game as human vs AI with it set on continuous and human vs AI where the human enters orders?

bstar ran a game into 43 without problems a short time ago. maybe there is a significant difference when you have the human entering orders every turn?

just a thought.

[Don Bowen]

Ive been running a quick test this morning, with encouraging results.

I un-did the 25% reduction is AP/AK/TK capacity and ran for 1 month - AI vs AI. The AI seems to have woken up quite a bit. Still not completely right as the second group of attacks did not happen (Naga, Kuching). However, the AI is currently sending it's CVs down to Rabaul so things are moving. Also, only a single allied sub as been sunk by aircraft (thru Jan 17th, 1942). I also note that NO ship has been sunk during the Pearl Harbor raids under V1.5: 1 Scenario 15 run and perhaps half a dozen CHS.

This is my Plan:

1. Release CHS tomorrow for head-to-head play. I still would like feedback on which version to release: 18th Japanese Division in Canton or 18th Japanese Division in Taan (targeted to Kota Bharu) plus extra transports in 1st Day TF.
2. Begin work on a Human vs Japanese AI version, trying to isolate a set of circumstances that will make it an enjoyable game. This will result in a second version of CHS "just for me" but I would be willing to make it available at some time in the future.
3. Continue support on the "standard" CHS as required and continue to accept input from all the other contributors.
4. Tonight, when I lay me down to sleep, I'm going to run Pry's Scenario 33 (with reduced capacities) for comparison.

Other Notes: Not only did the AI wake up with larger Capacities, it also used a bold attack directly on Manila with the 48th Division, ignoring San Mercelino and Bataan. The Allied AI was too stupid to react and most of the Asiatic Fleet was lost in Harbor. Malaya attacks are proceeding somewhat slowly - only about 1/2 way down the Peninsula. The 55th Division is approaching Moulmein. And, the AI has not kept Pearl Harbor supplied and it is almost time to eat the rats there.

P.S. To Bradley. I think there might be some advantage given to the AI if the "Japanese Computer" setting is selected. I have notice the Japanese AI seems to do better in this setting than in Computer-Computer. Perhaps a "gee the Japanese are AI and the Allied are human so the AI needs improved odds" sort of thing. Something else to test.

Don

[jwilkerson]

ORIGINAL: Don Bowen


Ive been running a quick test this morning, with encouraging results.

I un-did the 25% reduction is AP/AK/TK capacity and ran for 1 month - AI vs AI. The AI seems to have woken up quite a bit. Still not completely right as the second group of attacks did not happen (Naga, Kuching). However, the AI is currently sending it's CVs down to Rabaul so things are moving. Also, only a single allied sub as been sunk by aircraft (thru Jan 17th, 1942). I also note that NO ship has been sunk during the Pearl Harbor raids under V1.5: 1 Scenario 15 run and perhaps half a dozen CHS.

This is my Plan:

1. Release CHS tomorrow for head-to-head play. I still would like feedback on which version to release: 18th Japanese Division in Canton or 18th Japanese Division in Taan (targeted to Kota Bharu) plus extra transports in 1st Day TF.
2. Begin work on a Human vs Japanese AI version, trying to isolate a set of circumstances that will make it an enjoyable game. This will result in a second version of CHS "just for me" but I would be willing to make it available at some time in the future.
3. Continue support on the "standard" CHS as required and continue to accept input from all the other contributors.
4. Tonight, when I lay me down to sleep, I'm going to run Pry's Scenario 33 (with reduced capacities) for comparison.

Other Notes: Not only did the AI wake up with larger Capacities, it also used a bold attack directly on Manila with the 48th Division, ignoring San Mercelino and Bataan. The Allied AI was too stupid to react and most of the Asiatic Fleet was lost in Harbor. Malaya attacks are proceeding somewhat slowly - only about 1/2 way down the Peninsula. The 55th Division is approaching Moulmein. And, the AI has not kept Pearl Harbor supplied and it is almost time to eat the rats there.

P.S. To Bradley. I think there might be some advantage given to the AI if the "Japanese Computer" setting is selected. I have notice the Japanese AI seems to do better in this setting than in Computer-Computer. Perhaps a "gee the Japanese are AI and the Allied are human so the AI needs improved odds" sort of thing. Something else to test.

Don

Well as always I want the units especially the large units - like divisions - in the places they were historically - but if we know moving this division to the historical location "breaks the AI" ... then at least for this release ( until we can determine a better fix ) we should put the 18th back in Taan ( targeted to Kota Bharu ).


BTW on the turn I shut my test down ( 5/28/42 ) there just happened to be a big carrier battle going on in the Solomons sea ... Saratoga got sunk ... and she was sunk by 2 KB task forces near green Island and some Betty's from Rabaul ... so the AI can get itself into carrier battles in the Solomons !

[Lemurs!]

Joe, you just want the 18th division in Canton because then it will take me two more days to crush your pathetic, feeble British forces in Malaya! [:D]

Then, after i take Malaya the glorious Imperial Japanese forces shall open Hooters!

Mike

[Tanaka]

ORIGINAL: Don Bowen


Ive been running a quick test this morning, with encouraging results.

I un-did the 25% reduction is AP/AK/TK capacity and ran for 1 month - AI vs AI. The AI seems to have woken up quite a bit. Still not completely right as the second group of attacks did not happen (Naga, Kuching). However, the AI is currently sending it's CVs down to Rabaul so things are moving. Also, only a single allied sub as been sunk by aircraft (thru Jan 17th, 1942). I also note that NO ship has been sunk during the Pearl Harbor raids under V1.5: 1 Scenario 15 run and perhaps half a dozen CHS.

This is my Plan:

1. Release CHS tomorrow for head-to-head play. I still would like feedback on which version to release: 18th Japanese Division in Canton or 18th Japanese Division in Taan (targeted to Kota Bharu) plus extra transports in 1st Day TF.
2. Begin work on a Human vs Japanese AI version, trying to isolate a set of circumstances that will make it an enjoyable game. This will result in a second version of CHS "just for me" but I would be willing to make it available at some time in the future.
3. Continue support on the "standard" CHS as required and continue to accept input from all the other contributors.
4. Tonight, when I lay me down to sleep, I'm going to run Pry's Scenario 33 (with reduced capacities) for comparison.

Other Notes: Not only did the AI wake up with larger Capacities, it also used a bold attack directly on Manila with the 48th Division, ignoring San Mercelino and Bataan. The Allied AI was too stupid to react and most of the Asiatic Fleet was lost in Harbor. Malaya attacks are proceeding somewhat slowly - only about 1/2 way down the Peninsula. The 55th Division is approaching Moulmein. And, the AI has not kept Pearl Harbor supplied and it is almost time to eat the rats there.

P.S. To Bradley. I think there might be some advantage given to the AI if the "Japanese Computer" setting is selected. I have notice the Japanese AI seems to do better in this setting than in Computer-Computer. Perhaps a "gee the Japanese are AI and the Allied are human so the AI needs improved odds" sort of thing. Something else to test.

Don

You guys need to make two versions of CHS...

1. CHS Head to Head: As historical as possible

2. CHS AI: Will have limitations....

[Andrew Brown]

Don,

Have you tried running the same AI vs AI test with one of Pry's latest scenarios? He has also reduced AP/AK capacities, so I think that this is not the cause of the problem (assuming that Pry's scenarios play out OK when using the AI). The cause may be the slot number of transport ships and/or groups, depending on how much these were changed around (I am at work so I can't look for myself). The AI MAY depend on such slot numbers, rather than load capacity, to calculate load values and for transport TFs. This is only a guess though - it would be extremely helpful to get a hint or two about this from one of the developers.

I would hate to have to increase base values again. And I would also not like to increase supply. In fact I think that it should be reduced further than we already have, for both the Allies and the Japanese.

[TheElf]

I think Baby steps. Release it as a H2H for now and advertise it as such. I think enought people understand the challenges here and will be patient. After all PBEM is where it is at. Think of it as Beta phase two. As more people strike up PBEM games the feedback will increase. In the interim continue testing/tweaking the AI side.

[pad152]

Human vs Japanese AI version

I always thought that Witp needed seperate scenarios to play against the AI. The AI needs help (more better pilots, resources, air groups, better aircraft upgrades, etc.)

[Don Bowen]

The new Human-vs-Human release of CHS is now available at Spooky's. Doesn't work well with the AI - poor thing seems overwhelmed.

Anyway, enjoy. Please post any problems and, as always, thanks to Spooky for his excellent support.

Don

[Tanaka]

ORIGINAL: Don Bowen


The new Human-vs-Human release of CHS is now available at Spooky's. Doesn't work well with the AI - poor thing seems overwhelmed.

Anyway, enjoy. Please post any problems and, as always, thanks to Spooky for his excellent support.

Don

Don I see you decided to leave the pilots pool as is???