-
Notifications
You must be signed in to change notification settings - Fork 0
/
measure_umps.R
150 lines (133 loc) · 6.93 KB
/
measure_umps.R
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
# This script will produce a data frame containing several consistency and accuracy measurements
# on each MLB umpire.
library(dplyr)
library(tibble)
library(sp)
library(rgeos)
library(pracma)
library(MASS)
pitches <- as_data_frame(readRDS("pitches2017.Rda"))
games17inc <- as_data_frame(readRDS("games17inc.Rda"))
games17inc <- subset(games17inc, npitch>=50) # throw out games with less than 50 pitches
games17inc <- subset(games17inc, npitch<=300) # throw out games with more than 300 pitches
umpid <- unique(pitches$umpID)
numumps <- length(umpid)
ngames <- numeric(numumps)
npitch <- numeric(numumps)
umpname <- character(numumps)
# Average inconsistency metrics
aiR1 <- numeric(numumps)
aiR10 <- numeric(numumps)
aiIDX7 <- numeric(numumps)
aiCH <- numeric(numumps)
aiACH7 <- numeric(numumps)
accRB <- numeric(numumps) # season accuracy using rule-book rectangle
accCZ <- numeric(numumps) # season accuracy using 2017 consensus zone
accS <- numeric(numumps) # season accuracy using 2017 umpire's zone (self-accuracy)
errHD <- numeric(numumps) # Hausdorff distance from consensus KDE contours
errSD <- numeric(numumps) # Area of symmetric difference with consensus KDE contours
czMAD <- numeric(numumps) # mean absolute difference of sample of CDF estimates
zsize <- numeric(numumps) # season zone size (area of KDE contour)
rBB <- numeric(numumps) # walk rate
rK <- numeric(numumps) # strikeout rate
# Rule book zone: up/down pz's have been normalized to go from
# 1.5 to 3.5. Width of baseball is 0.245 feet, so we add 1/2 of
# a baseball's width to each edge. Width of plate is 17 inches.
# (17/12)/2+0.245/2 = 0.8308333
rbzoneX <- c(-0.8308333, 0.8308333, 0.8308333, -0.8308333)
rbzoneY <- c(1.3775, 1.3775, 3.6225, 3.6225)
# Consensus zone: Pitches that are called strikes
# 50% or more of the time. Computed in consensus_zones.R.
czonepoly <- readRDS("conzonepoly.Rda")
conczKDE <- readRDS("conczKDE.Rda")
for(i in 1:numumps) {
#for(i in 1:2) { # for testing
uid <- umpid[i]
pitchdata <- subset(pitches, umpID == uid)
calledPitches <- pitchdata[pitchdata$des=="Ball" |
pitchdata$des=="Ball In Dirt" |
pitchdata$des=="Called Strike", c("px","pz","des","stand")]
calledPitches <- calledPitches[!is.na(calledPitches[,1]),]
npitch[i] <- nrow(calledPitches)
ngames[i] <- length(unique(pitchdata$gameday_link))
umpname[i] <- as.character(pitchdata$umpName[1])
incMetrics <- subset(games17inc, umpid == uid)
aiR1[i] <- mean(incMetrics$incR1)
aiR10[i] <- mean(incMetrics$incR10)
aiIDX7[i] <- mean(incMetrics$incIDX7)
aiCH[i] <- mean(incMetrics$incCH)
aiACH7[i] <- mean(incMetrics$incACH7)
balls <- calledPitches[calledPitches$des=="Ball" | calledPitches$des=="Ball In Dirt",
c("px", "pz", "stand")]
strikes <- calledPitches[calledPitches$des=="Called Strike", c("px", "pz", "stand")]
accRB[i] <- (sum(point.in.polygon(strikes$px, strikes$pz, rbzoneX, rbzoneY)) +
nrow(balls) - sum(point.in.polygon(balls$px, balls$pz, rbzoneX, rbzoneY))) /
npitch[i]
accCZ[i] <- (sum(point.in.polygon(subset(strikes, stand == "L")$px,
subset(strikes, stand == "L")$pz,
czonepoly$L$px, czonepoly$L$pz)) +
sum(point.in.polygon(subset(strikes, stand == "R")$px,
subset(strikes, stand == "R")$pz,
czonepoly$R$px, czonepoly$R$pz)) +
nrow(balls) -
sum(point.in.polygon(subset(balls, stand == "L")$px,
subset(balls, stand == "L")$pz,
czonepoly$L$px, czonepoly$L$pz)) -
sum(point.in.polygon(subset(balls, stand == "R")$px,
subset(balls, stand == "R")$pz,
czonepoly$R$px, czonepoly$R$pz))) /
npitch[i]
stk <- list(L=data_frame(), R=data_frame())
bll <- list(L=data_frame(), R=data_frame())
cp <- list(L=data_frame(), R=data_frame())
stkKDE <- list(L=list(), R=list())
cpKDE <- list(L=list(), R=list())
czKDE <- list(L=list(), R=list())
szcontour <- list(L=list(), R=list())
szcontourdf <- list(L=data.frame(), R=data.frame())
for(s in c("L", "R")) {
stk[[s]] <- strikes[strikes$stand==s,c("px","pz")]
bll[[s]] <- balls[balls$stand==s,c("px","pz")]
cp[[s]] <- calledPitches[calledPitches$stand==s,c("px","pz")]
stkKDE[[s]] <- kde2d(stk[[s]]$px, stk[[s]]$pz, n=200, lims = c(-2,2,0,5))
cpKDE[[s]] <- kde2d(cp[[s]]$px, cp[[s]]$pz, n=200, lims = c(-2,2,0,5))
czKDE[[s]] <- stkKDE[[s]]
czKDE[[s]]$z <- czKDE[[s]]$z/cpKDE[[s]]$z*nrow(stk[[s]])/nrow(cp[[s]])
szcontour[[s]] <- contourLines(czKDE[[s]], levels=0.5)
szcontourdf[[s]] <- data.frame(px = szcontour[[s]][[1]]$x, pz = szcontour[[s]][[1]]$y)
}
accS[i] <- (sum(point.in.polygon(subset(strikes, stand == "L")$px,
subset(strikes, stand == "L")$pz,
szcontourdf$L$px, szcontourdf$L$pz)) +
sum(point.in.polygon(subset(strikes, stand == "R")$px,
subset(strikes, stand == "R")$pz,
szcontourdf$R$px, szcontourdf$R$pz)) +
nrow(balls) -
sum(point.in.polygon(subset(balls, stand == "L")$px,
subset(balls, stand == "L")$pz,
szcontourdf$L$px, szcontourdf$L$pz)) -
sum(point.in.polygon(subset(balls, stand == "R")$px,
subset(balls, stand == "R")$pz,
szcontourdf$R$px, szcontourdf$R$pz))) /
npitch[i]
cpL <- SpatialPolygons(list(Polygons(list(Polygon(as.matrix(czonepoly$L))),ID="cpL")))
cpR <- SpatialPolygons(list(Polygons(list(Polygon(as.matrix(czonepoly$R))),ID="cpR")))
upL <- SpatialPolygons(list(Polygons(list(Polygon(as.matrix(szcontourdf$L))),ID="upL")))
upR <- SpatialPolygons(list(Polygons(list(Polygon(as.matrix(szcontourdf$R))),ID="upR")))
# Using rgeos:
zsize[i] <- (gArea(upL) + gArea(upR))/2
errHD[i] <- gDistance(cpL,upL, hausdorff = TRUE) + gDistance(cpR, upR, hausdorff = TRUE)
errSD[i] <- gArea(gSymdifference(cpL, upL)) + gArea(gSymdifference(cpR, upR))
czMAD[i] <- mean(abs(czKDE$L$z-conczKDE$L$z)) + mean(abs(czKDE$R$z-conczKDE$R$z))
playdata <- pitchdata[!duplicated(pitchdata$play_guid.1),]
# playdata <- pitchdata[!duplicated(pitchdata$ab_id),] # TODO: add play_guid.1 column
num_walks <- sum(playdata$event == "Walk")
rBB[i] <- num_walks/nrow(playdata)
num_ks <- sum(playdata$event == "Strikeout")
rK[i] <- num_ks/nrow(playdata)
if((i %% 10) == 0) cat(".")
}
umps17 <- data.frame(umpid, umpname, ngames, npitch, aiR1, aiR10, aiIDX7, aiCH, aiACH7,
accRB, accCZ, accS, errHD, errSD, czMAD, zsize, rBB, rK)
saveRDS(umps17, file="umps17.Rda")
regularUmps17 <- umps17[umps17$ngames >= 20,]