CPU Baserunning AI Cheese

[fistsoffaith]

I've begun to notice a pattern in games against the CPU where the decision to send a runner seems to be determined by the eventual outcome of the play. For example, let's say Player X is on first when Player Y hits a single in the gap. The throw comes in to the cutoff man, who then relays the ball home. 9 out of 10 times, Player X stops at third as the cutoff man gets the ball. In these 9 times, the relay throw from the cutoff man to home ranges from fairly accurate to spot on. However, 1 out of the 10 times, the runner will decide to test the defense and sprint home, despite the cutoff man already having the ball. When this happens, the throw home, despite getting there egregiously ahead of the runner, seems to always miss. The catcher will not be able to dig it, or it will be so askew that he'll be put into a "running off the base" animation.

I suppose this could just be bad luck, but since I began tracking it, the last five times a CPU runner took off in a situation that I judged as making little sense (in other words, a situation that should have resulted in a clear out and the runner being showcased on a "what was he thinking?" ESPN special), he was safe because of an errant throw/bad dig four out of the five times. In the same period of time, the amount of times I've had an errant throw/bad dig in a situation where the runner decided to hold? Twice, despite the far larger quantity of opportunities. I will note that I'm only using plays at home for this data, although I've observed the same thing happen at other bases.

I thought I'd check with others about this; any data I produce represents such a small sample size that it could just be really bad luck. That being said, if it's not bad luck, it has to be one of two things - 1) that the presence of a runner testing the defense reduces fielding attributes significantly or 2) that the CPU calculates the outcome before making the decision. #2 seems unlikely, as I don't think that's how the game's algorithms work. That leaves the theory of fielding attributes being decimated by runners testing the defense, which doesn't make a whole lot of sense either. Anyone else have any input?

[nomo17k]

When you make these plays, are you actually making them smoothly?

I think in fielding, if you are not making smooth plays (like preloading sufficiently so that fielding animations make smooth transitions, making throws with right amount of strength, not under/overthrowing), you may be increasing the chance of making errors/errant throws.

For example, I often notice that the throw after a fielder makes a fielding mistake tends to be harder to make and less accurate (and often errant). It's a sort of mechanism to simulate what tends to happen in real life (and one miscue often leads to another in real life as well).

[fistsoffaith]

Quote: Originally Posted by nomo17k

When you make these plays, are you actually making them smoothly? I think in fielding, if you are not making smooth plays (like preloading sufficiently so that fielding animations make smooth transitions, making throws with right amount of strength, not under/overthrowing), you may be increasing the chance of making errors/errant throws. For example, I often notice that the throw after a fielder makes a fielding mistake tends to be harder to make and less accurate (and often errant). It's a sort of mechanism to simulate what tends to happen in real life (and one miscue often leads to another in real life as well).

Yeah, that was something I considered when I first started noticing the issue - I thought, at first, that maybe the CPU was seeing my preload was off and then going because of it. However, I didn't observe a correlation between preload and aggressiveness after I started keeping track of the data. None of the four data points in my "four out of five" stat were preloaded poorly, while multiple of my "runner didn't go" data points were.

[nomo17k]

Quote: Originally Posted by fistsoffaith

Yeah, that was something I considered when I first started noticing the issue - I thought, at first, that maybe the CPU was seeing my preload was off and then going because of it. However, I didn't observe a correlation between preload and aggressiveness after I started keeping track of the data. None of the four data points in my "four out of five" stat were preloaded poorly, while multiple of my "runner didn't go" data points were.

Honestly you will be disappointed if you try to look for evidence that CPU actually "cheats" based on your input... because it does not. And a sample of n = 5 means really nothing in any statistical senses.

I think long throws have been made less accurate this year, that I can agree but what you are seeing is that CPU is just in general aggressive this year.

If long throws are much more accurate, then you will be complaining that CPU baserunners are too dumb that they go for extra bases when they clearly would be out.

[fistsoffaith]

Quote: Originally Posted by nomo17k

Honestly you will be disappointed if you try to look for evidence that CPU actually "cheats" based on your input... because it does not. And a sample of n = 5 means really nothing in any statistical senses. I think long throws have been made less accurate this year, that I can agree but what you are seeing is that CPU is just in general aggressive this year. If long throws are much more accurate, then you will be complaining that CPU baserunners are too dumb that they go for extra bases when they clearly would be out.

Oh, I absolutely concede that the sample size is far too small. We're talking about maybe a dozen games played since I started paying attention to this.

That being said, I wouldn't really consider a throw from the edge of the infield to home to be a "long throw." And if the problem truly is the degradation of throw accuracies, I'm wondering why I don't see it on plays where the runner doesn't go. It could just be me having terrible luck, but I'd like to hear from others to see if there have been any similar experiences.

And regarding the "the computer doesn't cheat" comment, like I said, I absolutely don't think it's feasible that the computer is cheating and predicting how the play is going to play out - that's just not how games work. That being said, I think it is possible that, in a situation where the runner is actively pursuing an additional base, fielding attributes are affected. Why this would be the case, I'm not sure, but it's just what I've observed so far.

[Musty_Elbow]

I'm glad somebody else has noticed this. This is one of my few gripes of the game. It seems the CPU will not advance unless it will be a bad throw. It just takes some of the realism out as it's very hard to nab a guy at the plate on a single up the middle from second. It can look like I will beat him by 3 steps but the throw will end up 20 feet down the line and he'll slide in easily. It's frustrating.

[Hustle1man]

Amen.

The craziest thing ive seen is the CPU scoring from second on a sac fly to RF! It wasnt even to the track and the guy had like 75 spd. The one thing that sucks is as a player you cant see the 3b coach to get the stop sign. And sure you can see the play but if the computer presets bad animation conceivably it could do the same for you.

[raneman85]

Quote: Originally Posted by Hustle1man

Amen. The craziest thing ive seen is the CPU scoring from second on a sac fly to RF! It wasnt even to the track and the guy had like 75 spd. The one thing that sucks is as a player you cant see the 3b coach to get the stop sign. And sure you can see the play but if the computer presets bad animation conceivably it could do the same for you.

This is my major complaint that has been showing up too frequently. Most of the other glitches I can live with. I've slowed the replay down and half the time the runner at the plate should have been called out, but there is no replay at home.