Bioavailability/FAO_micro_RF.R at master · willywutang/Bioavailability · GitHub

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
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
#' Random Forest regression bioavailability predictions of wheat tissue micronutrient contents from soil data
#' Soil and wheat plant wet chemistry data courtesy of FAO (doc @ https://www.dropbox.com/s/gwk07tanhu86tqj/Silanpaa%20Report.pdf?dl=0)
#' MIR soil data courtesy of ICRAF (2016)
#' M. Walsh, May 2016

# Required packages
# install.packages(c("devtools","caret","doParallel","randomForest")), dependencies=TRUE)
suppressPackageStartupMessages({
  require(devtools)
  require(caret)
  require(doParallel)
  require(randomForest)
})

# Data setup --------------------------------------------------------------
# Run this first: https://github.com/mgwalsh/Bioavailability/blob/master/FAO_micro_setup.R
# or run ...
# SourceURL <- "https://raw.githubusercontent.com/mgwalsh/Bioavailability/master/FAO_micro_setup.R"
# source_url(SourceURL)

# Labels: Wheat plant micro-nutrient concentrations (ppm)
pB  <- fao_cal$pB  ## Boron
pCu <- fao_cal$pCu ## Copper
pMn <- fao_cal$pMn ## Manganese
pMo <- fao_cal$pMo ## Molybdenum
pZn <- fao_cal$pZn ## Zinc
pFe <- fao_cal$pFe ## Iron

# Covariates
wetc <- fao_cal[c(4:24)] ## Wet chemistry calibration data
wetv <- fao_val[c(4:24)] ## Wet chemistry validation data
mirc <- fao_cal[c(32:1795)] ## MIR calibration data
mirv <- fao_val[c(32:1795)] ## MIR validation data

# RF models ---------------------------------------------------------------
# Start doParallel to parallelize model fitting
mc <- makeCluster(detectCores())
registerDoParallel(mc)

# Control setup
set.seed(1385321)
tc <- trainControl(method = "oob", allowParallel = TRUE)
tg <- expand.grid(mtry=seq(5, 20, by=1))

# Plant Boron content (ppm) -----------------------------------------------
# Wet chemistry covariates
pB_wet.rfo <- train(wetc, pB,
                    preProc = c("center", "scale"),
                    method = "rf",
                    ntree = 501,
                    tuneGrid = tg,
                    trControl = tc)
print(pB_wet.rfo)
pB_wet.imp <- varImp(pB_wet.rfo, useModel = FALSE)
plot(pB_wet.imp, cex=1.2, cex.axis=2, col="black", top=20)
plot(pB~fitted(pB_wet.rfo), xlab="Soil wet chemistry model fit (ppm)", ylab="Boron content of wheat plants (ppm)",
     xlim=c(-0.1,100.1), ylim=c(-0.1,100.1), cex=0.8, wetc)
abline(c(0,1), col="red", lwd=2)

# MIR covariates
pB_mir.rfo <- train(mirc, pB,
                    preProc = c("center", "scale"),
                    method = "rf",
                    ntree = 501,
                    tuneGrid = tg,
                    trControl = tc)
print(pB_mir.rfo)
pB_mir.imp <- varImp(pB_mir.rfo, useModel = FALSE)
plot(pB_mir.imp, top=20)

# Plant Copper content (ppm) ----------------------------------------------
# Wet chemistry covariates
pCu_wet.rfo <- train(wetc, pCu,
                     preProc = c("center", "scale"),
                     method = "rf",
                     ntree = 501,
                     tuneGrid = tg,
                     trControl = tc)
print(pCu_wet.rfo)
pCu_wet.imp <- varImp(pCu_wet.rfo, useModel = FALSE)
plot(pCu_wet.imp, top=20)

# MIR covariates
pCu_mir.rfo <- train(mirc, pCu,
                     preProc = c("center", "scale"),
                     method = "rf",
                     ntree = 501,
                     tuneGrid = tg,
                     trControl = tc)
print(pCu_mir.rfo)
pCu_mir.imp <- varImp(pCu_mir.rfo, useModel = FALSE)
plot(pCu_mir.imp, top=20)

# Plant Manganese content (ppm) -------------------------------------------
# Wet chemistry covariates
pMn_wet.rfo <- train(wetc, pMn,
                     preProc = c("center", "scale"),
                     method = "rf",
                     ntree = 501,
                     tuneGrid = tg,
                     trControl = tc)
print(pMn_wet.rfo)
pMn_wet.imp <- varImp(pMn_wet.rfo, useModel = FALSE)
plot(pMn_wet.imp, top=20)

# MIR covariates
pMn_mir.rfo <- train(mirc, pMn,
                     preProc = c("center", "scale"),
                     method = "rf",
                     ntree = 501,
                     tuneGrid = tg,
                     trControl = tc)
print(pMn_mir.rfo)
pMn_mir.imp <- varImp(pMn_mir.rfo, useModel = FALSE)
plot(pMn_mir.imp, top=20)

# Plant Zinc content (ppm) ------------------------------------------------
# Wet chemistry covariates
pZn_wet.rfo <- train(wetc, pZn,
                     preProc = c("center", "scale"),
                     method = "rf",
                     ntree = 501,
                     tuneGrid = tg,
                     trControl = tc)
print(pZn_wet.rfo)
pZn_wet.imp <- varImp(pZn_wet.rfo, useModel = FALSE)
plot(pZn_wet.imp, cex=1.2, col="black", top=20)
plot(pZn~fitted(pZn_wet.rfo), xlab="Soil wet chemistry model fit (ppm)", ylab="Zinc content of wheat plants (ppm)",
     xlim=c(-0.1,100.1), ylim=c(-0.1,100.1), cex=0.8, wetc)
abline(c(0,1), col="red", lwd=2)

# MIR covariates
pZn_mir.rfo <- train(mirc, pZn,
                     preProc = c("center", "scale"),
                     method = "rf",
                     ntree = 501,
                     tuneGrid = tg,
                     trControl = tc)
print(pZn_mir.rfo)
pZn_mir.imp <- varImp(pZn_mir.rfo, useModel = FALSE)
plot(pZn_mir.imp, top=20)

# Plant Iron contents (ppm) -----------------------------------------------
# Wet chemistry covariates
pFe_wet.rfo <- train(wetc, pFe,
                     preProc = c("center", "scale"),
                     method = "rf",
                     ntree = 501,
                     tuneGrid = tg,
                     trControl = tc)
print(pFe_wet.rfo)
pFe_wet.imp <- varImp(pFe_wet.rfo, useModel = FALSE)
plot(pZn_wet.imp, top=20)

# MIR covariates
pFe_mir.rfo <- train(mirc, pFe,
                     preProc = c("center", "scale"),
                     method = "rf",
                     ntree = 501,
                     tuneGrid = tg,
                     trControl = tc)
print(pFe_mir.rfo)
pFe_mir.imp <- varImp(pFe_mir.rfo, useModel = FALSE)
plot(pFe_mir.imp, top=20)

# Stop doParallel
stopCluster(mc)

# Test set predictions ----------------------------------------------------
# Wet chemistry covariates
pB_wet  <- predict(pB_wet.rfo, wetv)
pCu_wet <- predict(pCu_wet.rfo, wetv)
pMn_wet <- predict(pMn_wet.rfo, wetv)
pMo_wet <- predict(pMo_wet.rfo, wetv)
pZn_wet <- predict(pZn_wet.rfo, wetv)
pFe_wet <- predict(pFe_wet.rfo, wetv)
wet_pred <- cbind.data.frame(pB_wet,pCu_wet,pMn_wet,pMo_wet,pZn_wet,pFe_wet)
wet_test <- fao_val[c("SSID","pB","pCu","pMn","pMo","pZn","pFe")]
wet_eval <- cbind(wet_test, wet_pred)

# MIR covariates
pB_mir  <- predict(pB_mir.rfo, mirv)
pCu_mir <- predict(pCu_mir.rfo, mirv)
pMn_mir <- predict(pMn_mir.rfo, mirv)
pMo_mir <- predict(pMo_mir.rfo, mirv)
pZn_mir <- predict(pZn_mir.rfo, mirv)
pFe_mir <- predict(pFe_mir.rfo, mirv)
mir_pred <- cbind.data.frame(pB_mir,pCu_mir,pMn_mir,pMo_mir,pZn_mir,pFe_mir)
mir_test <- fao_val[c("SSID","pB","pCu","pMn","pMo","pZn","pFe")]
mir_eval <- cbind(mir_test, mir_pred)

# Write data files --------------------------------------------------------
write.csv(wet_eval, "wet_eval.csv", row.names=F)
write.csv(mir_eval, "mir_eval.csv", row.names=F)