Draft AI Redux for 1.85

[rsatlfan]

Shaun,

A couple of weeks back I posted a message regarding the version RC 1.84 RC2 and the computer AI while conducting a draft. I noted then that the computer made many weak picks when selecting its team(s).

Downloaded 1.85 last night and set up a two league eight team
association. I controlled one team and the computer had the other
fifteen. Financials weren't used and I used PureSim Classic with the
beginning year being 1905. Roster size was limited to 35. Set up the
franchises and conducted a draft.

My team picked 14th and I advanced the round pick-by-pick to see how the
computer did in drafting its teams. While the logic was a little better
than in 1.8.4.2 it still left a lot to be desired, especially in
choosing pitchers.

Specifically, the computer made several weak choices regarding pitchers
in the first two rounds. Ratings were low in many areas and potential
was simply not there as well. Rube Waddell was not picked until the end
of the 8th round and by the end of the draft the following pitchers were
not selected by any team: Christy Mathewson, Red Ames, Orval Overall,
Andy Coakley, George Mullin, and Ed Reulbach. All had great ratings and
performed well in real life.

Shaun, is making the computer Draft AI stronger something that you are working on currently? If the AI is poor it will be much easier for the human player to have immediate success rather than have to build a strong team over years of time. I think it lessens the challenge of the game to some extent.

Has anyone else noticed the poor computer logic when conducting a draft?

[eric517]

Oh, yes, I'm seeing it, too. See my comments in the 1.85 thread. Granted I am using fictional players and my issue has been with the amateur draft between seasons, so the situation may be a little different, but I do see the problem. In some amateur drafts I have seen as much as 20% of the draft picks be worthless scrubs. It's bad enough that I have considered a house rule to not select players in the amateur draft . . . I can only sign rookies during the free agent phase. I haven't gone that far yet because I love the drafting process. Yes, I agree, this probably does lessen the challenge of the game. It would probably be good (for fictional leagues) to eliminate scrubs from the draft pool. I'm not sure what purpose they serve, although I do remember a discussion a few years ago regarding scrubs and why they are there . . . but, maybe that was for historical leagues.

[Orcin]

I have posted about this at length in my 1946 St. Louis Cardinals replay thread in The Locker. I have conducted two drafts with the "improved" AI. In one, the AI was truly horrible. In the other, the AI did a pretty decent job and didn't leave much except scraps, but the order was questionable. I would like to see the AI further improved in this area, and it certainly should not leave HOF affinity players undrafted in favor of taking a scrub.

The AI should also be more aggressive about upgrading its roster in free agency. Most of the AI teams have a lot of unspent money, and they could sign a top player or two. Instead they wait until April and sign free agents after the player (unless he resists the urge) has cherry-picked the pool.

[Wrathchild]

ORIGINAL: eric517

I have considered a house rule to not select players in the amateur draft . . . I can only sign rookies during the free agent phase.

How do you not take a pick during the draft?

And is there a reason the phase has been renamed Offseason Draft from Amateur Draft? Does it have to do with financials being on/off?

And were free agents always in the Offseason/Amateur Draft? I thought previously I had seen only rookies.

[Orcin]

There are free agents in the draft? That's new.

[rjolley]

You can skip the draft by selecting Skip the Rest (I think that's the test on the button).

I usually skip the rest if I either have enough to fill my team or think the rest of the players aren't any good.

[Wrathchild]

ORIGINAL: Orcin

There are free agents in the draft? That's new.

They were in my last draft. Or, let me put it this way, at least those long-time players who had been drafted in the association's initial season and were mass released at the end of last season (Cy Young, Mike Donlin, etc.).

[CrashDavis]

ORIGINAL: Wrathchild

ORIGINAL: Orcin

There are free agents in the draft? That's new.

They were in my last draft. Or, let me put it this way, at least those long-time players who had been drafted in the association's initial season and were mass released at the end of last season (Cy Young, Mike Donlin, etc.).

Have you done a draft since v1.85 rolled out? Should now contain only true rookies.

[Wrathchild]

ORIGINAL: CrashDavis

ORIGINAL: Wrathchild

ORIGINAL: Orcin

There are free agents in the draft? That's new.

They were in my last draft. Or, let me put it this way, at least those long-time players who had been drafted in the association's initial season and were mass released at the end of last season (Cy Young, Mike Donlin, etc.).

Have you done a draft since v1.85 rolled out? Should now contain only true rookies.

No, I haven't. I'm in slow motion with this game right now. No freakin' time to devote to it! The last draft I did was with 1.84.4. Glad to hear that those vets won't be there in the next draft. Out of curiosity, any idea why the computer teams in 1.84.3 or 4 mass released so many of their long-term quality veterans at the end of the season? That's really been bothering me.

[rsatlfan]

Crash,

As someone who is very knowledgeable about PureSim and its development, what do you think about the strength of the Draft AI?

Do you see it being worked on to make it better?

[puresimmer]

ORIGINAL: rsatlfan

Shaun,

A couple of weeks back I posted a message regarding the version RC 1.84 RC2 and the computer AI while conducting a draft. I noted then that the computer made many weak picks when selecting its team(s).

Downloaded 1.85 last night and set up a two league eight team
association. I controlled one team and the computer had the other
fifteen. Financials weren't used and I used PureSim Classic with the
beginning year being 1905. Roster size was limited to 35. Set up the
franchises and conducted a draft.

My team picked 14th and I advanced the round pick-by-pick to see how the
computer did in drafting its teams. While the logic was a little better
than in 1.8.4.2 it still left a lot to be desired, especially in
choosing pitchers.

Specifically, the computer made several weak choices regarding pitchers
in the first two rounds. Ratings were low in many areas and potential
was simply not there as well. Rube Waddell was not picked until the end
of the 8th round and by the end of the draft the following pitchers were
not selected by any team: Christy Mathewson, Red Ames, Orval Overall,
Andy Coakley, George Mullin, and Ed Reulbach. All had great ratings and
performed well in real life.

Shaun, is making the computer Draft AI stronger something that you are working on currently? If the AI is poor it will be much easier for the human player to have immediate success rather than have to build a strong team over years of time. I think it lessens the challenge of the game to some extent.

Has anyone else noticed the poor computer logic when conducting a draft?

So, was this some sort of sandbox association? I have to say I have worked on the AI tirelessly lately, so maybe there is a nuance regarding how you created this assn.

[rjolley]

Shaun, I've seen the same type of thing with an association I started in 1970. I've been able to pluck stars towards the bottom of the first round that should've been top 5, players in the 3rd that should've gone in the first, and I see lowly rated players selected in the first and second round before these players. Now, on the one hand, I know who these players are. When I see Jim Rice available with the #17 pick, that's a steal for me. On the other hand, looking only at the ratings, I don't see how Jim Rice goes any lower than #10, especially with 3 or 4 teams picking 24 year old RF Joe Anderson who's rated  5 CH 5 PH 5 EY 56 POT or 28 year old SP Joe Johnson rated 5 ST 5 VE 5 CO 40 POT. I guess there's always a Matt Millen or Isiah Thomas running the draft, but there seems to be too many complete whiffs.

By the way, do those 5 CH 5 PH 5 EY players ever develop into anything useful?

[rsatlfan]

Shaun,

No, this was not a sandbox association. I used the options as mentioned
previously. That is, using the following setup: An association with
two leagues with eight teams in each. I did not use financials and
limited the teams to a 35 man roster. I used PureSim classic with the
start year being 1905.

Once the draft was completed I realized that I could have drafted all of
my position players first and wait until the last six rounds to pick six
of the nine pitchers because so many excellent ones were not even chosen by the teams controlled by the Draft AI. My rotation using those pitchers that were not picked by any computer controlled team would have turned out to be the six mentioned in my initial post in this thread. Having a rotation of Mathewson, Ames, Overall, Coakley, Mullin, and Reulbach would be tough to beat.

To confirm what I had found about a half hour ago I set up a new association with the same parameters to see if the Draft AI performed similarly. It did not choose the following pitchers in the 35 man draft: Rube Waddell, Christy Mathewson, Eddie Plank, Orval Overall, Ed Reulbach, George Mullin, Addie Joss, Jack Chesbro, Bob Ewing, Vic Willis and many other with better ratings than those chosen.

I am sure that you have been working hard on the AI, Shaun, and I think
that it is probably the toughest part of the simulation to get right.
However, I think that it is the most critical part of the game (rivaling in game computer manager decisions). In the case of the Draft AI it means the human player will be able to assemble an extremely strong (maybe unbeatable) team from the get go with minimal
effort.

Playing the game, I am very impressed with the amount of detail and
options available. Your hard work is very much appreciated by me and, I
know, everyone who plays PureSim.

Shaun, would you be able to run some tests to confirm what I posted?
Alternately, if there is a way to print out the draft results I would be
more than happy to run a series of drafts and send them to you so you
could see the drafting patterns and identify weaknesses.

Bob

[puresimmer]

ORIGINAL: rsatlfan

Shaun,

No, this was not a sandbox association. I used the options as mentioned
previously. That is, using the following setup: An association with
two leagues with eight teams in each. I did not use financials and
limited the teams to a 35 man roster. I used PureSim classic with the
start year being 1905.

Once the draft was completed I realized that I could have drafted all of
my position players first and wait until the last six rounds to pick six
of the nine pitchers because so many excellent ones were not even chosen by the teams controlled by the Draft AI. My rotation using those pitchers that were not picked by any computer controlled team would have turned out to be the six mentioned in my initial post in this thread. Having a rotation of Mathewson, Ames, Overall, Coakley, Mullin, and Reulbach would be tough to beat.

To confirm what I had found about a half hour ago I set up a new association with the same parameters to see if the Draft AI performed similarly. It did not choose the following pitchers in the 35 man draft: Rube Waddell, Christy Mathewson, Eddie Plank, Orval Overall, Ed Reulbach, George Mullin, Addie Joss, Jack Chesbro, Bob Ewing, Vic Willis and many other with better ratings than those chosen.

I am sure that you have been working hard on the AI, Shaun, and I think
that it is probably the toughest part of the simulation to get right.
However, I think that it is the most critical part of the game (rivaling in game computer manager decisions). In the case of the Draft AI it means the human player will be able to assemble an extremely strong (maybe unbeatable) team from the get go with minimal
effort.

Playing the game, I am very impressed with the amount of detail and
options available. Your hard work is very much appreciated by me and, I
know, everyone who plays PureSim.

Shaun, would you be able to run some tests to confirm what I posted?
Alternately, if there is a way to print out the draft results I would be
more than happy to run a series of drafts and send them to you so you
could see the drafting patterns and identify weaknesses.

Bob

So you do a 2 league 8 team assn and draft from a pool of 1905 players, with no finances, correct?

[puresimmer]

ORIGINAL: rsatlfan

Shaun,

No, this was not a sandbox association. I used the options as mentioned
previously. That is, using the following setup: An association with
two leagues with eight teams in each. I did not use financials and
limited the teams to a 35 man roster. I used PureSim classic with the
start year being 1905.

Once the draft was completed I realized that I could have drafted all of
my position players first and wait until the last six rounds to pick six
of the nine pitchers because so many excellent ones were not even chosen by the teams controlled by the Draft AI. My rotation using those pitchers that were not picked by any computer controlled team would have turned out to be the six mentioned in my initial post in this thread. Having a rotation of Mathewson, Ames, Overall, Coakley, Mullin, and Reulbach would be tough to beat.

To confirm what I had found about a half hour ago I set up a new association with the same parameters to see if the Draft AI performed similarly. It did not choose the following pitchers in the 35 man draft: Rube Waddell, Christy Mathewson, Eddie Plank, Orval Overall, Ed Reulbach, George Mullin, Addie Joss, Jack Chesbro, Bob Ewing, Vic Willis and many other with better ratings than those chosen.

I am sure that you have been working hard on the AI, Shaun, and I think
that it is probably the toughest part of the simulation to get right.
However, I think that it is the most critical part of the game (rivaling in game computer manager decisions). In the case of the Draft AI it means the human player will be able to assemble an extremely strong (maybe unbeatable) team from the get go with minimal
effort.

Playing the game, I am very impressed with the amount of detail and
options available. Your hard work is very much appreciated by me and, I
know, everyone who plays PureSim.

Shaun, would you be able to run some tests to confirm what I posted?
Alternately, if there is a way to print out the draft results I would be
more than happy to run a series of drafts and send them to you so you
could see the drafting patterns and identify weaknesses.

Bob

Thank you thank you!

It turns out there was just one silly bug with no finances in this case and once I fixed it all was well!

Matthewson and Waddell went 1,2 in the draft I just tested.