Cup of Coffee

June 24, 2015: Small Mistakes

70 Comments

My managers missed one of my two submissions for time off for my vacation, and now I'm essentially being forced to do their job and get all of it covered. It'll all be worth it when I'm gone.

70 thoughts on “June 24, 2015: Small Mistakes”

  1. If any of our Buffalonians are going to hear the Buffalo Philharmonic on the Fourth of July, you'll hear my nephew playing the timpani with them.

      1. No, he's filling in for someone (maybe one of those three) who can't be there. He's played with a lot of major orchestras doing that. He hopes to get a permanent gig, but he's 25, still in graduate school, and is making pretty good money as a timpani temp, so he's fine with it for now.

          1. Aye it is. My brother got a music degree then ended up joining the Air Force so he could play. Of course, he's also only 23, so he had time to find something.

  2. I see that local boy Brett Graves got a two-out stint (1 K) from the bullpen in the Midwest League all-star game. I texted his mom this morning and it sounds like they had a good time at the game.

  3. You want evidence that the Wolves will not draft Towns? It's nearly irrefutable.

    In a case of dueling freshman big men, the overriding debate of the 2015 draft is probably whether the Timberwolves should take Towns or Okafor. Our model says the answer is pretty clearly Towns, who has more than twice as great a chance of becoming a superstar as Okafor — and about a 20 percent lower probability of becoming a mere role player or bust. Okafor is much heavier for his height — a no-no according to the model — while Towns is a superior defender and passer, a good combination for a big man in today’s NBA.

    1. even after he's been told personally by the wolves he will be drafted?

      I think the joke is more likely Okafor becomes a major superstar and Towns is out of the league in four years

      1. the model lacks parameters for "odds of becoming a coke/meth addict" or "odds of becoming crippled by a self-inflicted gunshot wound at a stripper bar" or something. So, yea. Who knows?

        1. FWIW, one of my friends (ugh I'm sorry to start yet another story like this) just started a job at the agency that represents Towns, and I guess they're not even pretending like they don't know where he's going.

          1. That is comforting.

            Do we need to start referring to him as Karl Anthony [redacted]?

    2. As with all data fitting models, this one should be read with a salt lick at your side. It's a model for forecasting Statistical Plus/Minus over a player's years 2-5 in the NBA. Forecasting is hard, especially about the future. And where the hell is the variance estimate for those point estimates?

      I admit that I don't fully understand the model type that underlies these predictions (Random Forest model, a variant of so-called CART models).

      Basically, they are using a regression-tree type model to fit observed year 2-5 SPM values for past Chad Ford top-100 player draft candidates (presumably there is a selection step that allows for censoring, which, of course, is a "bust" outcome). They don't tell us squat about the criteria used to avoid over-fitting of the data. I'm not sure how relevant that is for generating point-estimates of parameter values for subsequent prediction, except to the degree that I simply don't believe that these estimated parameter values are the same as the "true" parameter values, nor that the prediction model should be expressed without a random error overall.

      Blah blah blah, tl;dr. These forecasts are fun to look at, but I can't really evaluate how much credence to give them.

            1. Prediction is Bohring.
              from Bohr's wikiquote:

              Prediction is very difficult, especially about the future.
              As quoted in Teaching and Learning Elementary Social Studies (1970) by Arthur K. Ellis, p. 431
              The above quote is also attributed to various humourists and the Danish poet Piet Hein: "det er svært at spå - især om fremtiden"
              It is also attributed to danish cartoonist Storm P (Robert Storm Petersen).
              The Danish source, used by Bohr and Petersen, has been traced back to Markus M. Ronner in 1918 by lundskovdk-citater.
              Variant: It's hard to make predictions, especially about the future.

      2. I want Towns because I'm gonna call him KAT-man. And when he makes a good play, I'm gonna yell, KAT-man DO!

      3. I believe that one of the reasons people like random forests is that they are typically pretty robust against overfitting. My experience with a lot of these machine learning models, though, is that you can't at all judge by the method in general, because there are enough levers and dials that there is a big opportunity for user error.

        I agree that it would be nice if more forecasts focused on the range and probability of outcomes. That said, I feel like they do a pretty good job with this by providing the percentage forecast for star/starter/role player/bust. If you look at someone like Frank Kaminsky, the model basically says "well, he's not going to be a star, but other than that, I have no idea" given that there is essentially an equal probability in the model of starter/role player/bust. The only thing I might add is something like a 10th/90th percentile SPM, since that's their metric of choice.

        I think the most important thing about doing forecasts like this is to do them consistently over time and try to learn from your mistakes. As you say, forecasting is difficult, but actually paying attention to how your forecasts perform over time is the only good way I know to get better at it.

        1. Russell's numbers are interesting, too. Looking at his probabilities, his project SPM is clearly brought down by his huge bust potential. The model says he has at least as good a chance as KAT of becoming a star, but also that he has a huge chance of becoming a total bust. Based on all the numbers in the table, I think you could make a good case for putting him 3rd on the board.

        2. Yea, I guess the percentages are supposed to be shares of the probability distribution above/below each cut point (which are exogenous and somewhat arbitrary; i.e., the authors chose the cut points for this classification scheme). I hadn't quite grokked that. Would be nice for them to have explained it in a sentence.

        3. do a pretty good job with this by providing the percentage forecast for star/starter/role player/bust
          Yeah, that's basically the model's StDev projection presented in a different form. If we knew the average SPM of each category, we could back into a StDev estimate for each player.
          I didn't find it in a quick scan of the article, so I jumped to the table.
          I solved for the average values for each class that perfectly fit the top four, and fit the top ten relatively well.
          Star 5.95 SPM
          Starter 1.29 SPM
          Role -.21 SPM
          Bust -1.05 SPM
          (The worst fit in the top ten is for #5, who seems to be most all-or-none, so his average value if star is probably different than most of the others.)

          Using this, the estimated Standard Deviation of Value for the top ten are:
          Rank. Initials SPM, StDev. (Coeff of Var)
          1. KAT 1.03, 2.18 (212%)
          2. JW 0.88, 1.82 (207%)
          3. SJ 0.65, 1.72 (265%)
          4. JO 0.52, 1.68 (323%)
          5. DR 0.51, 2.37 (465%)
          6. DJ 0.49, 1.22 (249%)
          7. DB 0.47, 1.75 (372%)
          8. WCS 0.35, 1.58 (451%)
          9. RHJ 0.31, 1.16 (374%)
          10. TL 0.27, 1.25 (463%)

          1. I'd have liked seeing the plots of the probability distributions for a few of the players being forecasted. Again, I haven't thought real hard about this, but those distributions ain't gonna be normal or truncated normal, are they?

            IIRC, they also set the cut points for the categories to restrict the numbers in each bin to pre-defined counts (1, 10, 25, everyone else).

  4. Morning game alert: If you can't wait for the Twins game to start, Fort Myers, with Stephen Gonsalves pitching, is hosting Palm Beach in a game starting right about now.

    1. 1b is overrated anyway. DH is almost as valuable defensively, so why not try an overshift?

    2. I'm not sure If I correctly remember the rules, but I think that if Molly submits that, and Vargas plays 1b in the top of the first, Hughes has to bat for Mauer (who can't return to the game).
      That'd be really different. Pitcher batting 8th is yesterday's news. Pitcher batting third!

  6. So today, I got a facebook friend request from Fritz Peterson. You can never be one hundred percent sure of these things on facebook, but it certainly appears to actually be former Yankees pitcher Fritz Peterson. I find that to be pretty cool.

  7. Spooky's subject line put a Dan Deacon song in my head...
    https://www.youtube.com/watch?v=h-uOH_9TwUc
    Why won't these bees leave me alone?
    I hate them bees.

    My dad is so cool, he is the coolest dad in dad school SelectShow

    1. So much mocking! I don't think I'd seen a few of those before- some are truly ridiculous.

        1. Virginia is explicit while Massachusetts merely implies, which is much more insane.

    4. Yes, there's a lot to mock, but taken as a whole, that criticism sucked. They were against simple and they were against complicated. That leaves basically nothing.

      Personally, I hate all the flags that try to put any kind of small writing or seal on the flag. (Washington's flag is particularly bad here--the flag has writing that says "The Seal"on it? What is it, the seal or the flag? What next, "This Is Not A Pipe?" Though I'm not a big fan of all these states, I feel like the flags in Colorado, Texas, New Mexico, Alaska, Arizona, South Carolina, and Tennessee are all pretty reasonable as flags go.

      1. Well, that's kindof the point. It is juvenile.

        Add Arkansas to your list.
        Among those with text and/or seals, Louisiana and Oklahoma are pretty good.

        Minnesota could be more-mocked, "Oh, look, the Indian's leaving!"

        Virginia's flag actually inspired a Spookymilk game.

      2. Yeah, these were pretty much my thoughts too. My acceptable flags include Arizona, Indiana, New Mexico, Ohio (weird, but I kind of dig, but only if actually non-rectangular like Nepal), Oklahoma, Tennessee, and Texas*.

        I also agree that pretty much everything with a seal should be right out. Blech. Flags should be simple and dynamic, as I am on record saying.

        * (I like the idea of the Alaska flag, but the execution is lacking; speaking of non-contiguous states, wtf Hawaii?)

        1. You convinced me back then, and I still subscribe to your thinking. Please produce more of it for me to consume.

          1. Also, seriously, I keep expecting the Washington one to start speaking with crudely animated jaw cutouts in the style Monty Python

        2. The Arizonan flag is as well-known around here as Old Glory. It's the only state flag I've ever seen tattooed on an arm, as far as I can remember.

          I have my issues with this place, but that flag isn't one of them. It's pretty cool.

        3. Oh sure, you're able to find one of the other links I'd been looking for.

          Anyway, I think New Mexico is my favorite of all the state flags. And Chicagah does have an excellent city flag.

      3. As criticism, it's pretty much rubbish. But as snark, I enjoyed it.

        That said, previous vexillological links shared here have been far more substantive. (Naturally, I can't find the discussion I most wanted to link to . . . )

          1. That is a quality reference, and I find myself consuming this new thinking of yours.

        1. Oh, I thought your "now he's resting. And now so am I." implied that you had been there.
          Maybe surgery was in Missouri because you were around and could help him recuperate.

          I had to find a way to rhyme and I'd already claimed my dad was the coolest in dad school.

  10. So, I cook dinner most nights, and recently Dr. Chop brought home America's Test Kitchen's Cooking for Two cookbook. I'm dubious about most cookbooks other than How To Cook Everything, and The Better Homes, but this book has provided two excellent meals in the first week. The first, Singapore Noodles with shrimp and red pepper, will be an immediate mainstay at the slaughterhouse, but the second was O.M.G level crazy good. I'll just say that there'll be a pork post in the future.

  11. Excitement in bSville. Next door neighboor's pickup truck burst into flames on the street in front of our house, flames shooting about a dozen feet into the air. Engine compartment completely burned out (small explosion and more flames -- the hood is completely gone). Tree in front of our house is singed, but luckily did not go up like a Christmas tree.

    Luckily, we live about a quarter mile from a fire station. They were here within about 5 minutes and doused the fire before anything else was engulfed. Like, say, my house.

    Wife took out our kitchen fire extinguisher for neighbor to use early on. Like pissing on a forest fire. And then neighbor says "I have a can of gas in the truck" and everybody backed way the eff up.

    1. That's horrifying! Very glad you're all right. What caused the truck to burst into flames?

        1. Pretty much. The truck was parked (not running) when it burst into flames. I don't know how long it had been sitting there. Presumably, the engine was still hot and there was a gas leak and...electrical short?

          1. In our previous apartment, one of our neighbors had his car burst into flames in the parking lot. The fire department said they respond to a half dozen or so a year (in a town of 50,000+). It apparently just happens.

    2. The Good Ol' Postseason Nemesis Yanquis.
      I doubt either team plays in the WC game though.

Comments are closed.