pFX

[SABR Matt]

I will be in Cleveland as well, Alan, and had a particular interest in the talks in PITCH f/x. I'm there to give a talk on my current research (Saturday 5:30 PM), but many other talks have caught my eye...looking forward to seeing what you have to say.

[alannathan]

PITCHf/x

Hi...Alan Nathan here. While babysitting my experiment in Lund Sweden, I surfed around and found this thread. Nice to see Dan Brooks and Mike Fast are alive and well. Just checking in to say hi. I'm afraid I don't know the rest of you yet. For those of you going to the annual SABR convention at the end of the month in Cleveland, I will be giving an oral presentation on PITCHf/x (Friday, 10:30). Later that day, the Science and Baseball Committee will meet (4-5 PM) and devote the entire session to discussions/presentations on PITCHf/x.

[Tink]

man, this PITCHf/x stuff is deep .. the more i explore it the deeper it gets .. thanks for the links to the PITCHf/x Summit, Mike - very interesting stuff .. my principle interest has been in trying to find some simple words to describe all this complexity in meaningful ways for me and readers .. the diagrams in this thread are the best i've seen .. i certainly don't have better simpler words to describe it yet

i now see that the "no spin" pitch is a red herring - fascinating and necessary for the creation of the metrics, but otherwise just added confusion

it is now clear that the BRK and PFX measurements alone are somewhat misleading, to say the least .. i don't feel bad at all about being confused .. in a few years we'll all likely be citing very exotic metrics as we all did back in the day when the radar gun made us all so excited about fastball pitchers .. but only if the simplest of verbal expressions can be meaningfully understood as easily as mph .. i remember standing near a car racing by at 90+ mph just to see what a Nolan Ryan fastball felt like .. it was so fast i didn't notice when it passed the 60' 6" point i had stepped off 20 yards away - a fairly accurate sensation

so far, it seems as if we're a long way from sensing pFx in a similarly meaningful fashion .. what i'd love to "see" today are comparative 3D trajectories in slow motion and at full speed to at least virtually "feel" and "sense" PITCHf/x metrics .. surely someone has done that .. if so, does anyone have any links?

now, i've learned from the Summit, on the horizon is HITf/x where bloop singles and searing line drives will be differentiated .. HR's will no longer be measured only by feet travelled, but have metrics all their own - way, way cool

anyway, very interesting stuff, guys .. thanks again for the introduction and education .. i've got a lot more to learn, a lot, lot more

[mikefast]

Yep, I love this stuff, too. The physics of baseball really interests me, as does the strategy that goes on between pitcher and batter.

Dr. Alan Nathan Physics of Baseball site is a good one, too, if you are interested in that sort of thing. His research has been very influential to me. http://webusers.npl.uiuc.edu/~a-nathan/pob/

By the way, it's worth noting that all the spin parameters we are talking about are being inferred/calculated from the movement. PITCHf/x is not actually measuring the spin of the ball. You can do that in a laboratory setting using much higher-speed cameras and reflective dots on the baseball, but in a game setting, cameras can't measure the spin.

The Trackman radar system that is used in golf can also be used to measure the spin of a baseball, but it's not in use yet in major league baseball as far as I know, and it's probably nowhere near as reliable a system as PITCHf/x.

[Tink]

thanks again, Mike .. yes, i am really interested .. i'm clueless what i'll do with it, but years ago my favortite toy was a very expensive VCR that enabled watching a Nolan Ryan curveball and fastball in 1/10 of a second increments .. a tenth of a second was much too slow, but the best available at the time for home use .. somewhere in that analogy is what i'm seeking in trying to understand PITCHf/x

as i'm coming to understand the analysis possible with PITCHf/x, i'm sensing that what once fascinated me was simply watching trajectories .. it appears with these many other pitch dynamics we're getting inside the trajectory down to the very spin on the ball .. that's way cool and pretty exciting for an old hack like me .. now, i'm itching to understand the relationship between the pitch dynamics and the batted ball - but many baby steps are needed first

[Tink]

thanks, Mike, i think i'm over the hump of trying to see the data in 3-space co-ordinates, for the moment .. your replies have been most helpful .. i appreciate you taking the time to tutor a natural tinker, especially with numbers

[mikefast]

If you're really interested in the details of how the PITCHf/x system works, you might take a look at the presentation that Sportvision CTO Marv White made at the PITCHf/x summit last month. It might be helpful for you to see how this data is collected.

You can download the whole ZIP file with all the presentations under the Presentations link here: http://sportvision.com/events/pfx.html. Then uncompress the file and look for the PDF file entitled "Marv White - PFX System Overview.pdf".

[Tink]

GOT IT, finally - jeez .. i've consistently been hung up trying to put 3-space co-ordinates on the values in the data fields which has caused a huge amout of my misunderstanding .. so, rather than thinking of the pitch clusters as pitch location (duh), i suspect different cluster locations are good indicators of something (i'm not even prepared to guess on that one at the moment)

time for a PitchFx reboot with this new perspective .. thanks for the link .. i'm sure i'll have more questions soon, but let me digest this for a while

[mikefast]

The origin of the PITCHf/x coordinate system is the point of home plate. The positive x-direction is toward the umpire/catcher's right. The positive y-direction is toward the pitcher's mound. The positive z-direction is up.

Sportvision assumes constant forces on the ball (drag, spin/lift, and gravity) and makes the best (Runga-Kutta) fit they can from the measured trajectory to this constant acceleration model. The constant acceleration fit gives them nine parameters: the initial position, initial velocity, and acceleration, all in three dimensions.

The pfx_x and pfx_z parameters that are reported in the Gameday XML data are calculated using the equations of motion:

pfx_x = 1/2 * x-acceleration * time^2, where time is the time it took the pitch to travel from y=40 feet to y=1.4 feet (the front of home plate).

pfx_z = 1/2 * (z-acceleration - gravity) * time^2

[BrooksBaseball]

First, a link to Bradford's stuff:

http://www.brooksbaseball.net/pfx/index.php?month=5&day=30&game=gid_2008_05_30_bosmlb_balmlb_1%2F&pitchSel=136268.xml&prevGame=gid_2008_05_30_bosmlb_balmlb_1%2F&prevDate=530

[*]8 actual pitches are charted[*]3 B=Balls (green)[*]2 S=Strikes (red) - swinging and/or foul, and [*]3 X=contact made (blue) - hit fair either an out, base hit of any type, or resulting in an error[*]for each pitch the co-ordindates plotted are pfx_x (horizontal break) and pfx_z (vertical break)[*] the 0, 0 point can be considered to be the "center" of the strike zone[*]these co-ordinates correspond to a physical location where an actual pitch "crosses" the plate[*]the cluster of pitches "down and in to RHH" are likely of the same type (curveball, slider, etc.) but not necessarily

First, just to make sure we're clear, the chart is showing the pfx_x by pfx_z, not the location of the ball. Location is something different than the path the ball took.

Here are the corresponding locations:


About how we get the Red Dot - the Red Dot is computed by the system, from which the rest of the PitchFX and Break parameters are derived.

[Tink]

Mike, thanks for the great ehancement to John Walsh's diagram.. i had completely forgotten that was the diagram for BRK .. that's a great visiualization of the break fields break_y, break_angle, and break_length .. but you added the label pfx_x to the blue line for Movement .. i've been thinking of pfx_x as the x-value in an (x, z) co-ordinate pair .. how does that translate into a line? does the line correspond to the base of the PFX triangle in Dan's?

WAIT A MINUTE, maybe i answered my own questions there .. Dan's illustration and yours are clearly labelled pfx_x .. maybe it is NOT a co-ordinate, but rather the base of what i'll call the PFX triangle

help

[Tink]

first, a Biblical reference: you guys have the patience of Job and i have more questions than Thomas thank you, again, for trying to answer them .. clearly, to me, there is something i am not yet understanding

let me start with checking a few basic assumptions referencing Dan's first chart of Bradford's pitches above titled Horizontal Break vs. Vertical Break

1. 8 actual pitches are charted
2. 3 B=Balls (green)
3. 2 S=Strikes (red) - swinging and/or foul, and
4. 3 X=contact made (blue) - hit fair either an out, base hit of any type, or resulting in an error
5. for each pitch the co-ordindates plotted are pfx_x (horizontal break) and pfx_z (vertical break)
6. the 0, 0 point can be considered to be the "center" of the strike zone
7. these co-ordinates correspond to a physical location where an actual pitch "crosses" the plate
8. the cluster of pitches "down and in to RHH" are likely of the same type (curveball, slider, etc.) but not necessarily

if those assumptions are correct, here is my immediate confusion .. the pfx_x (horizontal break) and pfx_z (vertical break) co-ordinates are for an actual pitch which in Dan's "transcendent" diagram corresponds to the blue dot

if that is correct and i understand your "transcendent" diagram, Dan, the value we call PFX is the diagonal distance between the blue dot and red dot (the co-ordinates of the "Computed location with NO spin" .. where then are the co-ordinates of the red dot .. how can we compute PFX without it?

i'm going to stop there for feed back in case i'm way off base, which would not surprise me, but i have several other questions .. btw, Dan, when you reply, if it's not too much trouble could you provide a link to your tabled data for Bradford's pitches .. i'd realy like to see the corresponding data fields for each pitch .. again, a huge thank you for your patience

[mikefast]

if so, can you provide a "transcendent" diagram of what the break fields describe .. i think i have a clue, but my diagramming skills are imaginary

Adapting John Walsh's diagram that I linked at the top of the page, here are the break elements from the Gameday XML data illustrated:

[BrooksBaseball]

very nice, StillFlash .. Dan's illustration is worth a 1000 words

Dan, my only quibble with your "transcendent" diagram is that the actual pitch location is generally higher than the computed NO SPIN location due to the effect of backspin .. isn't it?

Only for Fastballs, and only when they're thrown by guys with fairly standard deliveries. The diagram shows something that could be a slider thrown by a lefty (hence the filename! ), a sinker/2seamer thrown by a RHP with a sidearm grip, a splitter thrown by a RHP, a curve thrown by a lefty, etc.

For example, here is a plot of Chad Bradford's stuff from a few days ago:



The X axis here is pfx_x, the Y axis is pfx_z.

He has a heavily sidearm delivery, but you can see that the spin he throws on the ball generally causes it to move down and in to RHH.

Similarly, a guy like Manny Delcarmen gets a lot of downward motion due to spin on his curveball:



The reason Curveballs curve is because they are thrown with the opposite spin as fastballs (forward spin).

i haven't had a chance to look at the detailed data yet, but i'd guess from all that has been said that the same pythagorean calculation might also define the BRK value .. is that anywhere near correct?

answering my own question: i'd say NO .. i see in Dan's Tabled Data only these pitch variables or data fields relevant to BRK - break_y, break_angle, and break_length

based on the pitch_type field nearby, it appears the BRK value we see in Gameday is simply break_length .. is that correct, Dan?

I'm pretty sure that's correct. But I don't think it has anything to do with the placement of the Pitch Type field, the positions in the data don't really have anything to do with the similarity of the fields.

if so, can you provide a "transcendent" diagram of what the break fields describe .. i think i have a clue, but my diagramming skills are imaginary

It's already posted, at the top of this page, by Mike Fast.

The "BRK" variable is the red line on the graph, showing the "bendyness" of the pitch.


--Dan

[Tink]

very nice, StillFlash .. Dan's illustration is worth a 1000 words

Dan, my only quibble with your "transcendent" diagram is that the actual pitch location is generally higher than the computed NO SPIN location due to the effect of backspin .. isn't it?

i haven't had a chance to look at the detailed data yet, but i'd guess from all that has been said that the same pythagorean calculation might also define the BRK value .. is that anywhere near correct?

answering my own question: i'd say NO .. i see in Dan's Tabled Data only these pitch variables or data fields relevant to BRK - break_y, break_angle, and break_length

based on the pitch_type field nearby, it appears the BRK value we see in Gameday is simply break_length .. is that correct, Dan?

if so, can you provide a "transcendent" diagram of what the break fields describe .. i think i have a clue, but my diagramming skills are imaginary