Let HAL do lineup?

[voovits]

Just remember something.

The topic of discussion kinda shifted to HALs choice of starters, which is not always reliable.
What I rely on, and swear by, is the batting order he uses out of those starters. That will reliably tell me how good/bad the players in the lineup are.

[ROBERTLATORRE]

THREE times I was lucky enough to get George Brett in the AD. Three times HAL batted him 8th/9th against RHP/LHP. I cut him each time, worst card each time.

I know there is a lot of room for interpretation, but sometimes HAL will shoot up the red flare on good players, particularly if they have 4 good and 1 bad card to choose from.

So always with a grain of salt, but if it's obvious i usually go with it.

[gbrookes]

Jimmy C:

I left some key openings in my "vs. righty" lineup the other night. Hal put Huskey in RF instead of Salmon. I know Salmon has his worst card ('99) but it is still .862 OPS. Plus he is "E" in that year. Huskey's best OPS card is .830 and all his cards lean left. Salmon is a better defensive player in RF also. Yet...Hal started Huskey against a righty over Salmon.

I think Hal uses the Eenie-Meenie-Miney-Moe method way too often for lineup selection...[quote:86399a39a4]

Jimmy C, I agree with you. Using my run production calcs, I estimate that Husky's 97 card and Salmon's 99 card are about even offensively vs. RHP, BUT Salmon's defense is better. For what it's worth, Salmon also has fewer GB A's, and he is a slightly better base runner, but the defense and arm should be enough for HAL to start Salmon instead of Huskey, by my calculations.

However, I HAVE noticed something similar to this in my experiences with HAL:

I sometimes find that HAL starts players where I don't expect him to start them, even when I KNOW what the cards are. This, despite the fact that my model does a pretty darn good job of predicting HAL's choices.

My conclusion from this is that HAL uses some weighting toward the run production for each player for facing the OTHER handedness of pitching. So, in this example, I have noticed that HAL might start Huskey (97) over Salmon (99), because of Huskey's superior hitting against LHP! You may laugh, groan, or say I am really off my rocker, but I HAVE noticed this. My thinking is that HAL provides some weighting to the other handed portion of the card, based on the possibility of a relief pitcher coming in with the opposite throwing arm.

The other very distinct possibility is that my own run production model does NOT properly model HAL's run production model. So, if HAL assigns a lower run production factor to walks than I do, then he might pick Husky instead of Salmon. But, my own calculations say that Husky's inferior defense should have him sitting on the bench - not to mention Salmon's great arm in RF!

Here are my calcs, rounded to nearest whole number product, (using 1-10, or .5, for the ballpark homeruns):
Huskey - 97

walks - 0
singles - 14 * 2 = 28 points
doubles - 5.85 *3.25 = 19
triples - .6 *4.44 = 3
homeruns - 5.4 * 5.75 = 31
ballpark HRs - 4 * 5.75 = 23

total - 104

Salmon - 99

walks - 21* 1.4 = 29
singles - 3 * 2 = 6
doubles - 7.45 * 3.25 = 24
triples - .8 *4.44 = 3
homeruns 4 * 5.75 = 23
ballpark HRs - 3 * 5.75 = 17

total - 102

By contrast, Huskey vs LHP is 127, while Salmon is 98.

I'll see if I can find another example, but I can honestly say that I have seen HAL do this kind of thing in other situations.

Geoff[/quote:86399a39a4]

[gbrookes]

i got the quote marks mixed up.
geoff

[voovits]

How accurate do you say your formula is Geoff? I've been trying to come up with some sort of formula myself to help figure out HAL, but haven't been able to come up with anything that works consistently.

I know there has to be some sort of hard mathematical formula that HAL uses, he is a computer after all.

[gbrookes]

I'd say it would duplicate HAL's choices over 90% of the time. Maybe over 95% of the time. Left or Right, it works consistently well.

I find that HAL factors in the defense using the same factors, with the strat probabilities for the X chart. To convert the defense factor to a batting card factor, I multiply the defense results times the number of die rolls per 216 for each X chart position, and then multiply that by 10 (for a batter lower in the order) or by 8 (for a batter higher in the order). The 8/10 factors are derived as 40 potential pitcher/defense rolls per game, divided by 5 offensive plate appearances (higher in the order) or 4 offensive plate appearances (lower in the order).

I have simplified the approach to defense, at the cost of accuracy. I just do a simple addition of the hits from the range results, and the errors from the e ratings. I do, however, treat 2 baggers differently from 1 baggers, etc, as though they were doubles and singles.

I'll do another post later with an example of the defense adjustment that I use.

I know some players (like spider) actually include adjustments for arms, base stealing and even GB A's and strikeouts. I don't do mathematical adjustments for those items. Spider's formula is different than mine, but overall it is the same general idea and produces similar results. For the other factors, I just use my own judgement to over-ride the math results (say for arm strength or base stealing). Spiders model attempts to estimate actual run production/prevention, whereas I just wanted something that I felt was easy to use and understand. The per-player results for me work out around 100, which I kind of like - like an index. My index doesn't really mean anything in itself - it only means something in relation to other player's indexes.

We both use the ratings disk, and then just add a bunch of calculations to it, with formulas copied and pasted to each player line. The main nuisance and time consumer with that is just converting some of the alpha cells to their proper numeric format, where they have asterisks in them. I added a couple of fields for things like N versus W power (1 or 0, respectively).

I want to clearly acknowledge my debt to DeanTSC for his NERP work, and the offense/defense articles posted in the strategy forum, and to Marcus Wilby, who referred me (and others on these boards) to some really interesting empirical data on run production factors, that was developed using some kind of regression analysis. I am a shameless borrower, but i will acknowledge my debt! I found the empirical data that Marcus accessed to be the best predictor of HAL's decisions. I took the most recent year from Marcus's data and averaged it for that most recent year. I think it was the year 2000 data that I used.

:)
Geoff

:)

G

[gbrookes]

And, by the way, my record is only just barely over .500!

So take everything I say with a grain (or box) of salt!
G

[gbrookes]

Well, I was right on most of my assessments, mainly on the batters. On the pitchers, I was maybe just slightly better than 50/50.

But I WAS right on most of the batters, including Molitor, who had his 4th card, as I guessed when HAL refused to put him in the batters box at all in the team's first series of the season!

It's just plain harder to assess the pitchers, especially the starting pitchers. I realize that, although sometimes the performance and statistics lead to an accurate conclusion, almost as often the statistics are skewed by the randomness of die rolls. There is a frequent occurrence of what turns out to be good pitcher cards getting "hammered" by the overall good quality of hitting in most TSN strat leagues, as the managers simply pick the better hitters to play - even in the mystery leagues, managers tend to get the good hitters out. Next time, I will be more patient with my pitchers!!! :) G