Probably not too much... That's why I think Zatkof is the odd man out; Murray and/ or Jarry should play back-up (i.e. given their potential ability to return assets). It might not happen this year, but next?? One stays; one gets moved. Proving abilities @ the NHL level gives management much greater flexibility.
- out_of_market

Murray and Jarry need to get as many reps as possible. Murray should get a few NHL games this year but I don't want him as the backup.

Perron sid kessel
Plot geno horn
Kuni Bonino bennett
Sund Cullen dupuis

Letang mattaa
Cole cleds
Dumo lovejoy
Gonchar

MAF
Murray


Seems like a solid grouping. Id like to see pouliot up but if he needs one more half season in wbs then so be it. As long as cleds doesn't get lost in waivers...

Perron sid kessel
Plot geno horn
Kuni Bonino bennett
Sund Cullen dupuis

Letang mattaa
Cole cleds
Dumo lovejoy
Gonchar

MAF
Murray


Seems like a solid grouping. Id like to see pouliot up but if he needs one more half season in wbs then so be it. As long as cleds doesn't get lost in waivers...
- nh4442

'Cleds'? How about Dinger.

I think Kuni has done enough to date;

Kuni, Sid, Philbo (brother of Bilbo Baggins)
BB, Geno, Horny
Perron, Boner, Sundy
Plot, Culldog, Duper

Tanger, Maatta
Cole, Dinger
Dumo, Lovejob

Scuds.

BB is a skill player that has played ok with Geno in the past, & is also (by the looks of video) a bit of a character which we all know Geno is. Good mix.

The best scenario is trading fleury and have Murray work out as a number one.

Doubtful we go down that route. Fleury seems well liked in the room and has been with the core for a long time.

Cap and asset management wise having a young elc goalie like Murray would be a bf Benefit. Fleury will be a pen almost certainly for two more years. Murray will be in wbs most of this year and next year I bet they make him the back up and give him 20-30 starts to prove himself and boost his value as log as he is performing.

1 PP goal for the winner & a SHG for the losing team. Just goes to show!


This is what bugs me, you break it down to all sorts of different events when you are presented with an anomaly to your stats but I see so many times on here - possession wins games, with no further information. You guys give generalised statements in support of your Analytics but when presented with 'facts' that dispute them, you go all .......well, there's this & there's that & then if this happens & that happens.........

Generalised statements & averages = current Analytics.
- Aussiepenguin

I'm not disagreeing w/ your assessment. By profession, I have a statistics, machine learning/ artificial intelligence background. The main gripe I have w/ analytics (used in media/ blogs) is correlation and averages. The game is constrained by time, and space, and transitions to various states throughout; it cannot easily be explained by either. Anyone who tells you differently is a ______.

Don't substitute correlation between shots (Corsi/ Fen) and winning with causation. Statistically, there is a major difference. Correlations can be found anywhere; determining causation requires controlled experimentation. Being able to predict an outcome is another story, as-is conditional probabilities (and odds of winning). By example, I've read studies on first-goal winning percentages that show a probability > 65% even 70% when period and time left in the game is considered. Also, don't flip probability for certainty... they're two different concepts.

Just for poops and giggles:
http://www.tylervigen.com/spurious-correlations

1. in those photos, I bet Plotnikov had no idea what any of this was about (someone probably gave him the spark notes version, but still)

2. I think Clendening has looked very good. good player who makes good decisions with the puck and has that offensive flare. based on his play and bonino last night, that trade was an absolute steal.

3. Plotnikov has looked a lot better. he's been playing better hockey with and without the puck.

4. Johnston has A LOT of options at forward. multiple ways to build the top 9. important for him to experiment to see what works best.

5. A bit under the radar, but I thought Cullen played well tonight also.

6. Murray looks like a 5 year veteran in net. to see a goalie at that age play in the NHL with that kind of poise and positioning is rare. also, he sees the play really well. it will be hard to get full value for him in a trade because a lot of teams undervalue goalie prospects until they prove themselves in the league. i think the penguins might be better off to hold on to him unless a deal comes up to swap him for a position player prospect with similar pedigree.

7. looked like a fun atmosphere tonight. thought the penguins looked good, especially given the talent TB was playing with tonight.

I'm not disagreeing w/ your assessment. By profession, I have a statistics, machine learning/ artificial intelligence background. The main gripe I have w/ analytics (used in media/ blogs) is correlation and averages. The game is constrained by time, and space, and transitions to various states throughout; it cannot easily be explained by either. Anyone who tells you differently is a ______.

Don't substitute correlation between shots (Corsi/ Fen) and winning with causation. Statistically, there is a major difference. Correlations can be found anywhere; determining causation requires controlled experimentation. Being able to predict an outcome is another story, as-is conditional probabilities (and odds of winning). By example, I've read studies on first-goal winning percentages that show a probability > 65% even 70% when period and time left in the game is considered. Also, don't flip probability for certainty... they're two different concepts.

Just for poops and giggles:
http://www.tylervigen.com/spurious-correlations
- out_of_market

So the machine learning & artificial intelligence experiments are completed under what conditions? Are they a static procedure initially then dynamic - not sure how a dynamic environment would go for a machine? The simple fact that learning & experimenting with different technologies are completed under scientific conditions. Sport itself is not a science & cannot be studied as such. These stats guys are saying sample size is important, well the fact that 1 single game is the only sample size under the same conditions or as close to as you will get tells you so much. No game is the same & no player is the same, conditions, opposition, ice, physical condition of the players, mentality, health etc etc, so how can anybody come out & say player X is 'this'? They are generalising the data they have & averaging out the results of past events. They cannot predict because they don't have the data to do so. It would be like creating a robot to walk on a flat surface & take his results - let's say time taken over 10m. Then take him outside & make him do the same on grass, then dirt, then mud, the uphill, then down hill then on gravel in water etc. Then not take into account wind rain etc & take all the results & say he will walk 10m in x minutes. It's an average, you are not breaking down the robots strengths & weaknesses in the conditions the tests were completed in.

In the same way Corsi is calculated by the players on the ice strength, if a player is on the ice with good players his Corsi will be usually better than if he's on the ice with lesser players - & the fact there has to be zero input to achieve a positive result is corrupt. If the robot walks with a wind he will usually do it better than against the wind.

There are so many unknowns in hockey these stats & the way they are correlated are corrupt. If you want to use Corsi you have to use it as a game by game number (& very recent), & take into consideration ALL the 'conditions' of that game. Hero charts are worthless unless you want a history lesson on averages for that player. Some people may like averages as a truth, I don't.

Kris Russell, Dennis Wideman, Jared Spurgeon, Carl Gunnarson, Dan Hamhuis

Which guy would you want and what kind of deal would it take?

Kris Russell, Dennis Wideman, Jared Spurgeon, Carl Gunnarson, Dan Hamhuis

Which guy would you want and what kind of deal would it take?
- YouMeAndDupuis9

If Sprong really is NHL level, Horny may get you 1 of those +.

HCMJ talking about Clendening....

“He’s jumped ahead of a lot of other defensemen,” Johnston said. “He’s caught our attention every game he’s played. He’s played with poise.”

If Sprong really is NHL level, Horny may get you 1 of those +.
- Aussiepenguin

0 reasons to trade Hornqvist.

Kris Russell, Dennis Wideman, Jared Spurgeon, Carl Gunnarson, Dan Hamhuis

Which guy would you want and what kind of deal would it take?
- YouMeAndDupuis9

Russell isn't very good. Wideman is terrible defensively. Spurgeon would cost a fortune as he's excellent. Know very little about Gunnarson but doesn't seem like an upgrade. Hamhuis is probably closest to realistic and I'm guessing at least a 2nd and 3rd would probably be the going rate for him. Might be a 1st or two 2nds.

HCMJ talking about Clendening....

“He’s jumped ahead of a lot of other defensemen,” Johnston said. “He’s caught our attention every game he’s played. He’s played with poise.”
- madmike71

That's exactly how I have seen it. Dumoulin has shown me little. I don't see him do anything noticeable one way or the other. Pouliot looks like a mixed bag with some nice plays along with some poor ones. I'm seeing consistently solid play from Clendening. He's a right shot too which the team needs.

0 reasons to trade Hornqvist.
- SuperHenderson13

Just saying. It would be suicide to do, but you never know how good Sprong will be???

That's exactly how I have seen it. Dumoulin has shown me little. I don't see him do anything noticeable one way or the other. Pouliot looks like a mixed bag with some nice plays along with some poor ones. I'm seeing consistently solid play from Clendening. He's a right shot too which the team needs.
- jfkst1

It would be nice to see the defense playing with our full team infront of them. Not to defend Pouliot too much because I think he hasnt been good in the pre season, but he may be looking to make plays that arent materializing because the WBS crew just arent able to see the opportunities that he does.
Id like to give Pouliot an opportunity to showcase with the top six is playing before sending him to the AHL.

As for Clendening, I think he is a nice compliment to Cole, Id like to see that pairing to start the year.

It would be nice to see the defense playing with our full team infront of them. Not to defend Pouliot too much because I think he hasnt been good in the pre season, but he may be looking to make plays that arent materializing because the WBS crew just arent able to see the opportunities that he does.
Id like to give Pouliot an opportunity to showcase with the top six is playing before sending him to the AHL.

As for Clendening, I think he is a nice compliment to Cole, Id like to see that pairing to start the year.
- stackthepads

Pouliot's mistakes look largely individual though. He just doesn't look any different to me than how he was last season. I'd still rather have him on the roster than Scuderi or Gonchar. How about:
Maatta-Letang
Cole-Lovejoy
Pouliot-Clendening
Scuderi

Pouliot's mistakes look largely individual though. He just doesn't look any different to me than how he was last season. I'd still rather have him on the roster than Scuderi or Gonchar. How about:
Maatta-Letang
Cole-Lovejoy
Pouliot-Clendening
Scuderi
- jfkst1

I agree with you that Pouliots game has been spotty, I just wonder if he can do more with the puck when he has better options. But that his work with the puck, what he really needs is alot of work on his game away from the puck. He is always just a step out of position, I think he was able to get away with that in junior because he has such good speed, but in the NHL he needs to be on that puck faster.

For my d pairs i'd start with:
Maatta Letang
Cole Clendening
Pouliot Lovejoy <--- Lovejoy did work great with Fowler, maybe he can help Pouliot bring his game up to speed.
Dumo Scuderi

So the machine learning & artificial intelligence experiments are completed under what conditions? Are they a static procedure initially then dynamic - not sure how a dynamic environment would go for a machine? The simple fact that learning & experimenting with different technologies are completed under scientific conditions. Sport itself is not a science & cannot be studied as such. These stats guys are saying sample size is important, well the fact that 1 single game is the only sample size under the same conditions or as close to as you will get tells you so much. No game is the same & no player is the same, conditions, opposition, ice, physical condition of the players, mentality, health etc etc, so how can anybody come out & say player X is 'this'? They are generalising the data they have & averaging out the results of past events. They cannot predict because they don't have the data to do so. It would be like creating a robot to walk on a flat surface & take his results - let's say time taken over 10m. Then take him outside & make him do the same on grass, then dirt, then mud, the uphill, then down hill then on gravel in water etc. Then not take into account wind rain etc & take all the results & say he will walk 10m in x minutes. It's an average, you are not breaking down the robots strengths & weaknesses in the conditions the tests were completed in.

In the same way Corsi is calculated by the players on the ice strength, if a player is on the ice with good players his Corsi will be usually better than if he's on the ice with lesser players - & the fact there has to be zero input to achieve a positive result is corrupt. If the robot walks with a wind he will usually do it better than against the wind.

There are so many unknowns in hockey these stats & the way they are correlated are corrupt. If you want to use Corsi you have to use it as a game by game number (& very recent), & take into consideration ALL the 'conditions' of that game. Hero charts are worthless unless you want a history lesson on averages for that player. Some people may like averages as a truth, I don't.
- Aussiepenguin

Thats a lot Aussie! You raise valid concerns. I'll pick my spots...

My comments about causation weren't related to sports, but rather generalized to things like design of experiments (DOE), AB testing, discrete choice studies. You picked a good example with the robot in more ways than one. How about AI for self-driving cars?! I'm not going to tackle ML/ AI in this post, but if your interested send a PM.

As far as stats in hockey, team evaluations are okay'ish. On the other hand, player evaluations usually don't pass muster, especially given their co-dependence rather than independence. The Corsi example you site is spot-on. I took RW to task the other day about the Scuderi post. Theoretically, a player could stand in the ref's half circle his entire shift and not even touch the puck, and he would accrue positive or negative Corsi numbers. Possession metrics are good for evaluating teams - not players, and even worse for players across teams. Even goals and assists and the Glass-to-Crosby or HERO analysis are technically flawed because the events are not mutually independent.

Sample size can be overcome through Bayesian methods, and are more of a concern w/ Frequentists. Analytics, especially player evaluations needs a boost in this regard (IMO). Hierarchical (multi-level) models are better suited for measuring at the player level. Hierarchical models can have varying slopes and intercepts (e.g. team and position) instead of regressing to a singular fitted line (i.e. average). Many of the conditions that you referred to would be considered exogenous variables in a predictive model. No system or model can capture every known (or unknown) variable and subsequently the error term (residual) absorbs the effect(s) thereof. Anyone can build a predictive model for team points, game outcomes, player points, etc... the key point being that not a single model will be 100% accurate. There will always be a level of error.

There are more patterns to hockey games than you might believe. Think of the game a series of transitions from one state to another, or what happens between faceoffs. Faceoffs (and location) reset the state of the game. The object of constructing a model is to determine probabilities of scoring, or allowing goals under particular condition (or reset). As an example, you could calculate the probability of scoring a goal off of the faceoff because of icing. Well, if the data has been collected over time. Which happens to be a good place to end..

Yes.

I agree with you that Pouliots game has been spotty, I just wonder if he can do more with the puck when he has better options. But that his work with the puck, what he really needs is alot of work on his game away from the puck. He is always just a step out of position, I think he was able to get away with that in junior because he has such good speed, but in the NHL he needs to be on that puck faster.

For my d pairs i'd start with:
Maatta Letang
Cole Clendening
Pouliot Lovejoy <--- Lovejoy did work great with Fowler, maybe he can help Pouliot bring his game up to speed.
Dumo Scuderi
- stackthepads

Those are good pairings, there is no law that says you must play the two best defensemen together. So you can mix and match the top four.

As far as Pouliott, you do one of three things with him. Second pairing, which he has not earned, send him to WBS to play top minutes or play him third paring to let him develop at the speed of the NHL game. I would like to see him play at the NHL level as I believe that will be better for him and the team, much like Letang did when he was Pouliott's age.

Thats a lot Aussie! You raise valid concerns. I'll pick my spots...

My comments about causation weren't related to sports, but rather generalized to things like design of experiments (DOE), AB testing, discrete choice studies. You picked a good example with the robot in more ways than one. How about AI for self-driving cars?! I'm not going to tackle ML/ AI in this post, but if your interested send a PM.

As far as stats in hockey, team evaluations are okay'ish. On the other hand, player evaluations usually don't pass muster, especially given their co-dependence rather than independence. The Corsi example you site is spot-on. I took RW to task the other day about the Scuderi post. Theoretically, a player could stand in the ref's half circle his entire shift and not even touch the puck, and he would accrue positive or negative Corsi numbers. Possession metrics are good for evaluating teams - not players, and even worse for players across teams. Even goals and assists and the Glass-to-Crosby or HERO analysis are technically flawed because the events are not mutually independent.

Sample size can be overcome through Bayesian methods, and are more of a concern w/ Frequentists. Analytics, especially player evaluations needs a boost in this regard (IMO). Hierarchical (multi-level) models are better suited for measuring at the player level. Hierarchical models can have varying slopes and intercepts (e.g. team and position) instead of regressing to a singular fitted line (i.e. average). Many of the conditions that you referred to would be considered exogenous variables in a predictive model. No system or model can capture every known (or unknown) variable and subsequently the error term (residual) absorbs the effect(s) thereof. Anyone can build a predictive model for team points, game outcomes, player points, etc... the key point being that not a single model will be 100% accurate. There will always be a level of error.

There are more patterns to hockey games than you might believe. Think of the game a series of transitions from one state to another, or what happens between faceoffs. Faceoffs (and location) reset the state of the game. The object of constructing a model is to determine probabilities of scoring, or allowing goals under particular condition (or reset). As an example, you could calculate the probability of scoring a goal off of the faceoff because of icing. Well, if the data has been collected over time. Which happens to be a good place to end..
- out_of_market

I'll reply brief, scoring is 99.9999999999999% goalie. There is no guaranteed move/tactic/play that will get you a goal whenever you do it. 1 day you may score the next 20 you won't due to the goalie. That is the unknown in scoring that you cannot explain. It doesn't matter if you have all the possession & a sog of 1000000 if the goalie stops them all, you don't score. You could be Sidney Crosby or Rob Scuderi, but if the goalie either stops it or let's it in depends if it's a goal or not.

As of right now it looks like the top three performing defensemen are:

Letang Maata Clendening

So the question becomes who do you pair with each if you spread them to all three lines?

Letang-Pouliot
Maata-Cole
Clendening-Dumo
Lovejoy-Scuds

I know it seems odd to put Pouliot with Letang but Letang has the speed and skill to complement him. He is very experienced and could teach him a lot. Cole gives a bigger body to help out and complement Maata. Dumo needs someone with agility and skating and offensive upside so Clendening would complement him that way.
Scuds and Lovejoy are the bottom dwellers based on their current level of play.

And again, can't say enough about the goaltending and PK last night. Excellent job, for sure.

As of right now it looks like the top three performing defensemen are:

Letang Maata Clendening

So the question becomes who do you pair with each if you spread them to all three lines?

Letang-Pouliot
Maata-Cole
Clendening-Dumo
Lovejoy-Scuds

I know it seems odd to put Pouliot with Letang but Letang has the speed and skill to complement him. He is very experienced and could teach him a lot. Cole gives a bigger body to help out and complement Maata. Dumo needs someone with agility and skating and offensive upside so Clendening would complement him that way.
Scuds and Lovejoy are the bottom dwellers based on their current level of play.

And again, can't say enough about the goaltending and PK last night. Excellent job, for sure.
- powerhouse

You have to be very careful talking about the PK around here. Some say it's a skill that can't be taught, some say it's a skill anyone can be taught, some, well they say it's all the goalie no matter if it's traffic cones in front of him, others say it's all the Stig!!

You have to be very careful talking about the PK around here. Some say it's a skill that can't be taught, some say it's a skill anyone can be taught, some, well they say it's all the goalie no matter if it's traffic cones in front of him, others say it's all the Stig!!
- Aussiepenguin

Whatever works, I say. And the goal tending especially and PK secondly were just outstanding. And they had lots of penalties to kill!

Whatever works, I say. And the goal tending especially and PK secondly were just outstanding. And they had lots of penalties to kill!
- powerhouse

You don't watch Top Gear?

The penalties are a huge worry at the moment pour moi. That was a pre season objective that is just as consistent or even worse than before.