v1.23.08 Playtest - 1812 Scenario

[pzgndr]

For the past couple of weeks, I have been fine-tuning and adjusting the new AI functions I've recently implemented. Retreats, advances, redeployments, and movements of special units (Gd, Arty) are tricky to get "just right." Things like force ratios for the strategic situation, the operational situation within an enemy nation, etc. The 1812 Scenario is the testbed I'm using, playing as Russia against France and other MPs as computer opponents on hard difficulty. The following screenshots show the development of the 1812 campaign...

[pzgndr]

June 1812 opening moves. As Russia, my strategy is to run. Northern armies retreating to the forest between Minsk and Vitebsk, the southern army retreating to Kiev. France advances to Kovno/Grodno/Brest-Litovsk.

[pzgndr]

July 1812. France moves first! So it's the dreaded double-move, and Davout leads the attack against Russian forces in the north. Russia chooses Defend, France chooses Echelon. The combat wasn't so terrible, but the French pursuit was brutal. Barclay retreats with what's left.

[pzgndr]

August 1812. Russia continues its retreat to Moscow, not wanting to take further chances at this point. France advances to siege Minsk, Vitebsk, and Smolensk. Napoleon lags behind with the Guard and Artillery.

[pzgndr]

September 1812. Russia holds firm in Moscow, but southern army under Tormasov advances west. My thought is to potentially threaten Frances rear flank depending on what they do. France continues a slow advance but does not proceed past Smolensk.

[pzgndr]

October 1812. Russia decides to go for Brest-Litovsk, easily taking the French garrison. France reacts violently and dispatches large forces under Napoleon himself back to Brest-Litovsk to stabilize the situation. Russia has a choice, to either defend and risk certain heavy losses or attempt a withdrawal and hope for the best. So, Russia chooses Withdraw but it fails. France chooses Assault, plus its artillery bombardments, and chews up the Russian defenders. Stragglers are totally eliminated by the French pursuit.

This was an army lost that didn't have to happen. However, it made France pull forces back to the frontier far away from Moscow, and the weather is starting to get cold...

[pzgndr]

November 1812. A quiet month. The Russian army holds in Moscow. French minor forces cautiously advance beyond Smolensk.

[pzgndr]

December 1812. Another quiet month. Russian infantry reinforcements in September and again in December are starting to beef up the army again. Looking at some of the French corps strengths, reduced more because of foraging losses than battle losses, I suspect the force ratio has shifted to the point where the French AI isn't making bold advances. Russia holds, and can start 1813 in fairly good shape.

[pzgndr]

A few closing comments. I have run and re-run this campaign probably a dozen times in the past week. Each time I'll see something that warrants an adjustment to the code, so with each "fix" it gets better.

This was a pretty good run, but I'm still not happy with how the Gd/Arty special units advance in the campaign. I'm also questioning the retreats, since those French minor corps close to Moscow should be close to being outnumbered and pulling back. So I will relook these code sections.

If nothing else, folks should recognize that the computer opponent is now something to be reckoned with. The AI is still not brilliant, but is much more capable than before and that makes it a bit more challenging. I will keep poking at v1.23.08 for the next couple of weeks, running playtests with both the 1805 and 1812 campaigns. I do hope to finalize an official release later this month. Keeping fingers crossed!

[DonCzirr]

Nice work Bill - thanks.

I'll upgrade once you do the official release and give my Brits a run on the full campaign.

[pzgndr]

ORIGINAL: DonCzirr
Nice work Bill - thanks.

Thank you. I share this stuff to show what I'm thinking about. The 'old' AI did not advance, retreat, or redeploy forces over long distances. Now it does. Sometimes good, sometimes not so good, so I playtest to see what the AI is doing and why, think about what it should be doing better and how to code that, and then make adjustments.

By the way, I created a new advance function for special units (Gd/Arty) so that's working quite well now. Quite well. I changed the Russian retreats to be 2-3 areas now, and also looking to apply that to other nations at war with France outside their home nation areas. Quick playtest as Turkey to watch AI France and AI Russia had the French steamrolling to Smolensk and beyond, and Russia went for a conditional surrender in October. Wow. I think I have unleashed the computer opponent! Now I'm pondering how the AI can garrison home cities during land move retreats, to help slow down invaders.

[pzgndr]

Round #2. It's been about 10 days and I've been busy trying to tackle several AI issues. Firstly, as discussed previously, I added a function for retreaters to garrison home cities. I did change the Russian retreats to be 2-3 areas in Russia. I also added a feature to un-assign leaders in overwhelming numbers situations, to help preserve AI nations' leadership from being all killed or captured. Another nagging issue was how the AI was placing and removing depots, and that's taken me several days to figure out and resolve to my satisfaction.

So, what follows next is another playtest of the 1812 campaign. This time I'm playing Turkey so I can observe a hands-off AI-vs-AI game. I'm pleasantly surprised at the results. Here we go...

[pzgndr]

June 1812. During Russia Reinforcement, leaders Bagration and Tormasov are un-assigned. Odds are 7:1 and 6:1, so they take a break. Their armies retreat back past Minsk and Kiev, leaving a garrison at Brest-Litovsk. Forward depots are removed, and rear depots established at Vitebsk and Kiev. France advances forward, assaulting Brest-Litovsk but the fortress holds.

[pzgndr]

July 1812. Russia continues its retreat. Barclay and Wittgenstein retreat 2 areas, leaving a garrison at Kovno. The other army also retreats 2 areas, to Smolensk. Oddly, Russian VI and XI Corps stayed at Kovno. France moves to attack both Barclay and Kovno. Barclay wins the battle and Murat retreats. Napoleon wins at Kovno, forcing the two corps to retreat with about 3 factors left.

Comment. Usually, Barclay would retreat 3 areas with all corps and be north of the Dvina River. This time he left himself vulnerable but held his own. The variability of retreats (1-2 areas or 2-3 areas, depending on odds) cause some uncertainty, and sometimes creates opportunity. The 2 corps at Kovno should have retreated toward the nearest depot, toward the northeast. There is a glitch somewhere that I have been trying to track down and resolve for good. Usually the retreats work correctly, but occasionally I still see things like this.

[pzgndr]

August 1812. France moves first, and gets a double move! Murat goes right back after Barclay, 73:33, but gets handed another defeat. Jerome goes after the Kovno stragglers with a trivial combat, 45:3, but Russia still manages to limp away with 1 militia. Barclay retreats back past Vitebsk. Russia assigns Kutnesov to the army at Smolensk.

[pzgndr]

September 1812. A quiet month. Russia holds its positions around Smolensk. Napoleon advances past Vilna towards Vitebsk. Davout is dispatched to deal with the last of the Kovno stragglers back in Konigsberg; 2 corps and 20 factors going the wrong way...

[pzgndr]

October 1812. The race is on. Barclay retreats to Moscow, and Kutnesov retreats to Tula. France advances to Vitebsk, but Napoleon lags behind with the Guard.

Comment. The AI was doing pretty good placing depots, but here Russia has not placed depots at either Moscow or Tula for some reason. An issue I need to look at, tomorrow...

[pzgndr]

November 1812. I didn't expect the AI to go for it, but it did. Murat charges ahead to Moscow and attacks, 54:36. L'audace, l'audace, toujours l'audace. However, to charge ahead and then select the Probe chit, and Barclay responds with Esc Counter Attack, the battle does not end well for the French. Russia holds.

Edit. Also notice how the southern army is now starting to retreat, now that the capital Moscow is threatened.

[pzgndr]

December 1812. Russian forces are consolidating at Moscow and standing firm. Still no depots being placed at Moscow or Tula. French do not attack again, but oddly move north to the forest west of Moscow. This ended up being fairly historical, in that the French made it to Moscow and the Russians are still holding on. I've seen the Russian AI sue for peace, a conditional surrender, to the French AI, but not this time. Well played.

Considering the several battles that Russia won and France lost, a quick look at the VPs shows Great Britain ahead at 80%, France in second place at 77%, and Russia a close third place at 74%. Not too bad.

[pzgndr]

Final comments. The advances and retreats appear to be working well. I'm still concerned about how Napoleon lags behind with the Guard, but the Artillery moved forward with Murat and was used against Barclay and the Russian artillery so that's good. The retreats leaving behind a garrison helps balance things out by causing some delays, at least for AI-vs-AI. A human player probably would not leave behind so many forces to besiege and would press ahead.

Another comment. I noticed after all these months that Bessarabia was not ceded to Russian control. I fixed that. The two Russian corps that start there did not retreat, either initially or once Moscow was threatened. So that's another issue I want to resolve.

Final comment. These playtests have been VERY productive. The computer opponent is now behaving like a more challenging adversary. I want to resolve these minor issues I mentioned and then move on to some final playtests. I want to play out 1812 through 1814 as Austria and/or Great Britain to see how the end-game looks. And then back to 1805 to see how the first couple of years look with these AI improvements. After that, maybe a few more minor adjustments and then wrap it up for official release. A couple more weeks hopefully. Fingers crossed...