Author Topic: Nylon Calculus: Who Should Boston Take With The Third Pick?  (Read 7712 times)

0 Members and 1 Guest are viewing this topic.

Re: Nylon Calculus: Who Should Boston Take With The Third Pick?
« Reply #15 on: May 27, 2016, 07:51:36 PM »

Offline Casperian

  • Ray Allen
  • ***
  • Posts: 3501
  • Tommy Points: 545
Here is last year's PAWS ranking



Tyus Jones 4th, Kevon Looney 5th, Porzingis 40th...

seems to me like a random dude with a random model with random results.

1) You don't assess the quality of this sort of model based on anecdotal hits and misses. You assess it on how well it predicts overall for a large body of players.
2) The model isn't trying to predict the draft.  The model is trying to project future NBA performance.  So looking at how it did after just one year is way too premature.

What I can see is that the three worst rated players on the list I posted are Porzingis, Hezonja and Trey Lyles. I think it's safe to say that there's no way these three guys will end up having the three worst NBA careers out of last year's draft.
« Last Edit: May 27, 2016, 07:57:17 PM by Casperian »
In the summer of 2017, I predicted this team would not win a championship for the next 10 years.

3 down, 7 to go.

Re: Nylon Calculus: Who Should Boston Take With The Third Pick?
« Reply #16 on: May 27, 2016, 07:55:16 PM »

Offline PickNRoll

  • Don Chaney
  • *
  • Posts: 1691
  • Tommy Points: 199
Interesting article.  I like Nylon Calculus, even if I disagree with the conclusion.  Last year's draft according to the PAWS-RSCI model:

1. Okafor
2. Towns
3. Tyus Jones
4. Russell
5. Kevon Looney
6. Kaminsky
7. Delon Wright
8. S. Johnson
9. Turner
10. Cliff Alexander
Portis, Winslow, Vezenkov, J. Grant, B. Dawson

Our own Terry Rozier and Jordan Mickey ranked 53rd and 56th respectively.  :)

Well, there is no way the model can account for some types of data, such as the impact that workouts have on GM's draft boards.

Models can't account for 1 important detail.  Is the guy a good basketball player?  This model, for instance, uses age somehow.  How exactly, I don't know, but here's a note from last year:

"Devin Booker falls despite being a well regarded recruit, simply because he is such a young prospect getting less boost from his [..] age."

So riddle me this.  How do age and basketball ability correlate?

The "impact of workouts on GM's draft boards" isn't relevant to the model.  As you said, it's trying to model NBA production, not GM behavior.  Is it a good predictor of the relative production of NBA prospects?  Based on last year, I would say not very good at all and worse than the big board I had as a random fan.

Re: Nylon Calculus: Who Should Boston Take With The Third Pick?
« Reply #17 on: May 27, 2016, 08:02:27 PM »

Offline dreamgreen

  • NCE
  • Ray Allen
  • ***
  • Posts: 3558
  • Tommy Points: 182
TP thanks for the info. Sometimes I think all these computer programs used to evaluate players is overrated and you should use logic with the eye test. But gathering all intel is good for sure.

Re: Nylon Calculus: Who Should Boston Take With The Third Pick?
« Reply #18 on: May 27, 2016, 08:32:43 PM »

Offline mmmmm

  • NCE
  • Rajon Rondo
  • *****
  • Posts: 5308
  • Tommy Points: 862
Here is last year's PAWS ranking



Tyus Jones 4th, Kevon Looney 5th, Porzingis 40th...

seems to me like a random dude with a random model with random results.

1) You don't assess the quality of this sort of model based on anecdotal hits and misses. You assess it on how well it predicts overall for a large body of players.
2) The model isn't trying to predict the draft.  The model is trying to project future NBA performance.  So looking at how it did after just one year is way too premature.

What I can see is that the three worst rated players on the list I posted are Porzingis, Hezonja and Trey Lyles. I think it's safe to say that there's no way these three guys will end up having the three worst NBA careers out of last year's draft.

So?  That doesn't make the model 'random'.

You should by no means ever, ever, ever expect to use a model like this to say, "This is how player A is going to be!" and make a decision on that player based on that.  That would be stupid.

The model is simply producing probabilities based on the inputs available to it.   It provides a probability that should lead you to look further into each player.   For example, the model doesn't look fondly on Jaylen Brown.  Why?  One should dive deep into his data and profile to try to understand why it differs with most mocks.   You could then choose to disagree with the model or the mocks.

But the model won't be right or wrong based on your decision and it won't be 'random' based on how Jaylen Brown individually turns out as an NBA player.
NBA Officiating - Corrupt?  Incompetent?  Which is worse?  Does it matter?  It sucks.

Re: Nylon Calculus: Who Should Boston Take With The Third Pick?
« Reply #19 on: May 27, 2016, 08:41:36 PM »

Offline Casperian

  • Ray Allen
  • ***
  • Posts: 3501
  • Tommy Points: 545

You should by no means ever, ever, ever expect to use a model like this to say, "This is how player A is going to be!" and make a decision on that player based on that.  That would be stupid.

I agree, except that's kinda what you claimed

Quote
2) The model isn't trying to predict the draft.  The model is trying to project future NBA performance.

...

Quote
The model is simply producing probabilities based on the inputs available to it.   It provides a probability that should lead you to look further into each player

The same is true for every model, as it is for the subjective value systems people apply when they make their own predictions for fun. Why should I pay attention to this one specifically, when there are myriads others out there who came to less questionable conclusions?

So, nothing has changed. A random dude with a random model (with questionable legitimacy).

Unless, of course, you picked this one specifically because it makes the point you want it to make.
In the summer of 2017, I predicted this team would not win a championship for the next 10 years.

3 down, 7 to go.

Re: Nylon Calculus: Who Should Boston Take With The Third Pick?
« Reply #20 on: May 27, 2016, 08:45:03 PM »

Offline loco_91

  • Bailey Howell
  • **
  • Posts: 2087
  • Tommy Points: 145
TP for the nice explanation of that rather crowded figure. While I do agree with the ultimate conclusion that Bender is narrowly the best choice, a couple things are puzzling to me:
-First, it's odd that Bender gets a plus in the "scoring" category, when ostensibly his pro scoring numbers were mediocre; perhaps 3p% is heavily weighted.
-Second, it's odd that Simmons (and Bender too, but to a lesser extent) gets such a small bump in the "distributing" category. I view this as his most important skill, but clearly the model does not agree; perhaps Simmons is hurt by being such an outlier in this regard, as the model doesn't know what to do with such Simmons's astronomical assist numbers as a PF.

And a couple things stood out as interesting:
-Overall, the model cares a lot about Blks+Stls and not very much about scoring.
-The model really likes Simmons's rebounding, and I feel that this is the most underappreciated aspect of his game. He's a stud on the boards.
-The model might be underrating Poeltl, as he gets a slight negative on Stls-Blks while being clearly a plus defender. He just tends to defend straight up instead of going for blocks, especially this year as he's had to play lots of minutes without fouling.
-The model is not very impressed with Murray's scoring, and it really dislikes his lack of steals. I'm not sure how discouraging this is exactly, but it's disappointing.
-I strongly agree with the model about Deyonta Davis. He claims that he can shoot 3's now; if that's true, he's a dark horse to be taken top five.

To answer your questions:

1) I don't know the exact formulae in the model off the top of my head, but I suspect that the scoring contribution isn't based just on scored point production.  Even though Bender scored modestly, he actually scored very efficiently.  His per-minute rate was pretty good (around 14/36) and even more important, his scoring efficiency was excellent, as his TS% is right around 60%.

2) Again, this probably has to do with both efficiency and volume.   Both Simmons and Bender are good passers, but in the big picture, they don't pass and generate as much playmaking volume and efficiency as a point guard like Dunn or Valentine.  Simmons' efficiency here is probably hurt by the poor shooters around him late in the year after injuries took out his better shooting teammates and Bender is probably hurt by raw volume being low.  Plus, some of Bender's best passing isn't necessarily in the form of assists, but rather in outlet passes.

Blocks and steals as defensive metrics have been shown in a few studies to have very strong correlations for success in the NBA.  So that's probably why they are key parts of the model.

I concur about Simmons' rebounding.  He is so explosive and fast at going after rebounds.  He doesn't rebound like a conventional 'big', blocking for position.  He tends to go after them more the way a rebounding wing like Turner or Rondo would, attacking from the edge, but because he's 6' 9" and super-athletic, he just gobbles them all up.

I concur also about Deyonta Davis.   If he really can be expected to shoot 3s ... he could be scary good.  I may nudge him up on my own board a notch or two!

Random aside -- this may not be the best 'comp' for him, but every time I look at Davis,


I keep thinking of Sleepy Sam Perkins:


Haha, yeah, Davis  is a really sleepy looking dude.

Re: Nylon Calculus: Who Should Boston Take With The Third Pick?
« Reply #21 on: May 27, 2016, 08:49:05 PM »

Offline mmmmm

  • NCE
  • Rajon Rondo
  • *****
  • Posts: 5308
  • Tommy Points: 862
Interesting article.  I like Nylon Calculus, even if I disagree with the conclusion.  Last year's draft according to the PAWS-RSCI model:

1. Okafor
2. Towns
3. Tyus Jones
4. Russell
5. Kevon Looney
6. Kaminsky
7. Delon Wright
8. S. Johnson
9. Turner
10. Cliff Alexander
Portis, Winslow, Vezenkov, J. Grant, B. Dawson

Our own Terry Rozier and Jordan Mickey ranked 53rd and 56th respectively.  :)

Well, there is no way the model can account for some types of data, such as the impact that workouts have on GM's draft boards.

Models can't account for 1 important detail.  Is the guy a good basketball player?  This model, for instance, uses age somehow.  How exactly, I don't know, but here's a note from last year:

"Devin Booker falls despite being a well regarded recruit, simply because he is such a young prospect getting less boost from his [..] age."

So riddle me this.  How do age and basketball ability correlate?

The age factor comes in in two ways:
1) When weighing the relevance of current season data, the age of the candidate relative to his competitors is important.  Seniors in the NCAA tend to produce better than freshman.   A 26 year old veteran in the Spanish ACB is going to produce better than an 18 year old rookie fresh up from Juniors.  Thus, the model gives more weight to the relevance of numbers when a younger player produces when playing older competition.

2) A young player is expected to have more upside potential.

Quote
The "impact of workouts on GM's draft boards" isn't relevant to the model.  As you said, it's trying to model NBA production, not GM behavior.  Is it a good predictor of the relative production of NBA prospects?  Based on last year, I would say not very good at all and worse than the big board I had as a random fan.
Again, assessing the quality of the model's predictions after one season would be stupid because one season -- the rookie season at that -- does not constitute an NBA career.   And looking at a handful of players anecdotally is not a statistical measure.

The quality can only be assessed over multiple seasons for multiple players over multiple drafts.

This is done with historical data in order to test the model.   Regressively running it over historical data allows you to tune the relevance (weights) of all factors to get the best statistical fit to what actually happened. 

The expectation is not about getting every single player projected 100% correct.  It's about getting _most_ players projected reasonably accurately.

Make your 'big board' as a random fan every year to predict the NBA career for every class and do it for a decade or two, see how your projections look for each draft after 8 years after each draft and get back to us about how well you did.   Otherwise, how well your random fan big board did last year is not really very helpful.

NBA Officiating - Corrupt?  Incompetent?  Which is worse?  Does it matter?  It sucks.

Re: Nylon Calculus: Who Should Boston Take With The Third Pick?
« Reply #22 on: May 27, 2016, 09:03:46 PM »

Offline mmmmm

  • NCE
  • Rajon Rondo
  • *****
  • Posts: 5308
  • Tommy Points: 862

You should by no means ever, ever, ever expect to use a model like this to say, "This is how player A is going to be!" and make a decision on that player based on that.  That would be stupid.

I agree, except that's kinda what you claimed

Quote
2) The model isn't trying to predict the draft.  The model is trying to project future NBA performance.

Except, no possible reading of the English language should take "trying to project" to mean "is going to be".   Except your reading, I guess.
Quote
...

Quote
The model is simply producing probabilities based on the inputs available to it.   It provides a probability that should lead you to look further into each player

The same is true for every model, as it is for the subjective value systems people apply when they make their own predictions for fun. Why should I pay attention to this one specifically, when there are myriads others out there who came to less questionable conclusions?

So, nothing has changed. A random dude with a random model (with questionable legitimacy).

Unless, of course, you picked this one specifically because it makes the point you want it to make.

The difference (and purpose) to attempting to build a computer model as opposed to relying on some individual human expertise is that the computer model can be tested by running it against hundreds and hundreds of players' data from the past and test it's overall quality.

The model wasn't written to produce the result that you (apparently) happen to not prefer and that I happen to concur with.  It was written to produce the results of previous NBA player careers based on their previous pre-NBA data.  Those results can be graded.

How well it's projections for more recent drafts and the future work will have to wait a few years to be graded.

NBA Officiating - Corrupt?  Incompetent?  Which is worse?  Does it matter?  It sucks.

Re: Nylon Calculus: Who Should Boston Take With The Third Pick?
« Reply #23 on: May 27, 2016, 10:02:05 PM »

Offline Casperian

  • Ray Allen
  • ***
  • Posts: 3501
  • Tommy Points: 545

You should by no means ever, ever, ever expect to use a model like this to say, "This is how player A is going to be!" and make a decision on that player based on that.  That would be stupid.

I agree, except that's kinda what you claimed

Quote
2) The model isn't trying to predict the draft.  The model is trying to project future NBA performance.

Except, no possible reading of the English language should take "trying to project" to mean "is going to be".   Except your reading, I guess.

Phew, for a second there I thought you were an arrogant wind bag, thank God that's not the case.

So, since it's apparently my limited command of the english language that's at the core of the problem, please explain to this giant dofus here the effective difference between "trying to project" and "is going to be". Because to me, it's tomato tomato, for all intents and purposes.

Sure, one talks in absolutes and the other in hypotheticals, but only the absolutes make sense when trying to gauge the predictive accuracy of such a model, anyway.
Otherwise, why "rank" them in the first place? That fact alone should tell you that there's an intent of trying to predict how a player's NBA career "is going to be" relative to his peers.


Quote
Quote

The same is true for every model, as it is for the subjective value systems people apply when they make their own predictions for fun. Why should I pay attention to this one specifically, when there are myriads others out there who came to less questionable conclusions?

So, nothing has changed. A random dude with a random model (with questionable legitimacy).

Unless, of course, you picked this one specifically because it makes the point you want it to make.

The difference (and purpose) to attempting to build a computer model as opposed to relying on some individual human expertise is that the computer model can be tested by running it against hundreds and hundreds of players' data from the past and test it's overall quality.

And there are many different computer models. You still haven't answered my question.

Quote
The model wasn't written to produce the result that you (apparently) happen to not prefer and that I happen to concur with.  It was written to produce the results of previous NBA player careers based on their previous pre-NBA data.  Those results can be graded.

This isn't a matter of preference in my case, but thanks for admitting it is in yours.


Quote
How well it's projections for more recent drafts and the future work will have to wait a few years to be graded.

So, as I said, questionable legitimacy.
« Last Edit: May 27, 2016, 10:08:26 PM by Casperian »
In the summer of 2017, I predicted this team would not win a championship for the next 10 years.

3 down, 7 to go.

Re: Nylon Calculus: Who Should Boston Take With The Third Pick?
« Reply #24 on: May 27, 2016, 11:03:47 PM »

Offline mmmmm

  • NCE
  • Rajon Rondo
  • *****
  • Posts: 5308
  • Tommy Points: 862

You should by no means ever, ever, ever expect to use a model like this to say, "This is how player A is going to be!" and make a decision on that player based on that.  That would be stupid.

I agree, except that's kinda what you claimed

Quote
2) The model isn't trying to predict the draft.  The model is trying to project future NBA performance.

Except, no possible reading of the English language should take "trying to project" to mean "is going to be".   Except your reading, I guess.

Phew, for a second there I thought you were an arrogant wind bag, thank God that's not the case.

So, since it's apparently my limited command of the english language that's at the core of the problem, please explain to this giant dofus here the effective difference between "trying to project" and "is going to be". Because to me, it's tomato tomato, for all intents and purposes.

Sure, one talks in absolutes and the other in hypotheticals, , but only the absolutes make sense when trying to gauge the predictive accuracy of such a model, anyway.  Otherwise, why "rank" them in the first place? That fact alone should tell you that there's an intent of trying to predict how a player's NBA career "is going to be" relative to his peers.
No where does Johnson's model assign a 100% probability to any of it's predictions.
Quote

Quote
Quote

The same is true for every model, as it is for the subjective value systems people apply when they make their own predictions for fun. Why should I pay attention to this one specifically, when there are myriads others out there who came to less questionable conclusions?

So, nothing has changed. A random dude with a random model (with questionable legitimacy).

Unless, of course, you picked this one specifically because it makes the point you want it to make.

The difference (and purpose) to attempting to build a computer model as opposed to relying on some individual human expertise is that the computer model can be tested by running it against hundreds and hundreds of players' data from the past and test it's overall quality.

And there are many different computer models. You still haven't answered my question.
If your question is, "Why should I pay attention to this one specifically...?", then you already paid attention to it only because you clicked on the link and read about it.  Beyond that, whether you pay it any more attention is entirely up to you. 

Quote
Quote
The model wasn't written to produce the result that you (apparently) happen to not prefer and that I happen to concur with.  It was written to produce the results of previous NBA player careers based on their previous pre-NBA data.  Those results can be graded.

This isn't a matter of preference in my case, but thanks for admitting it is in yours.
Given that I had already stated in my OP that I concurred with Johnson's conclusion, you could have thanked me much earlier.
Quote
Quote
How well it's projections for more recent drafts and the future work will have to wait a few years to be graded.

So, as I said, questionable legitimacy.

Yes.  As you said.

(I'm not sure what "legitimacy" is supposed to mean in this context so I'll just agree, hoping that makes you feel better about things.)

G'night.
NBA Officiating - Corrupt?  Incompetent?  Which is worse?  Does it matter?  It sucks.