-
Notifications
You must be signed in to change notification settings - Fork 3
/
smooth-curve_07.Rmd
336 lines (266 loc) · 10.9 KB
/
smooth-curve_07.Rmd
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
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
---
title: "Composite curves and bootstrap C.I.'s using locfit() <br> (smooth-curve.R)"
output:
html_document:
css: SI-md-08.css
fig_caption: yes
highlight: haddock
number_sections: yes
theme: united
toc: yes
toc_float: false
collapsed: no
---
```{r smooth-curve, echo=FALSE}
options(width = 105)
knitr::opts_chunk$set(dev='png', dpi=300, cache=TRUE)
pdf.options(useDingbats = TRUE)
```
# Introduction #
This script (`smooth-curve.R`) fits composite curves, with bootstrapping by site confidence intervals, using the `locfit()` function. This version implements a fixed window (half) width approach (as opposed to the "span", or a constant-proportion, variable-width window). As is the case with other analysis scripts, the script has two parts, a set-up portion that changes from analysis to analysis, and a computation part that does not change. This version reads "presampled" data.
# Set up #
The variables `queryname`, `basename` and `binname` are used to compose the path and filenames for the presampled data.
```{r pathnames}
# composite curve via the locfit package
# bootstrap-by-site confidence intervals
# names
queryname <- "v3i"
basename <- "zt21k"
binname <- "bw20"
# paths for input and output .csv files -- modify as appropriate
datapath <- "/Projects/GPWG/GPWGv3/GCDv3Data/v3i_Rscripts/"
sitelistpath <- "/Projects/GPWG/GPWGv3/GCDv3Data/v3i_RScripts/v3i_sitelists/"
sitelist <- "v3i_nsa_globe"
outpath <- "e:/Projects/GPWG/GPWGv3/data/v3i_Rscripts/v3i_curves/"
# presampled/binned files
csvpath <- "/Projects/GPWG/GPWGv3/GCDv3Data/v3i_Rscripts/v3i_presamp_csv/"
csvname <- paste("_presamp_influx_",basename,"_",binname,".csv", sep="")
```
Load the `locfit` library and set parameters. The window width `hw` is, following convention, the half-window width (of the tricube weight function). The number of bootstrap resampling replications is given by `nreps`.
```{r locfitLibrary}
library(locfit)
# locfit (half) window-width parameter
hw <- 500 # bandwidth (smoothing parameter)
# number of bootstrap samples/replications
nreps <- 200
```
Set the target ages for fitted values.
```{r targage}
# target ages for fitted values
targbeg <- -60
targend <- 22000
targstep <- 20
```
Set array sizes for saving bootstrap results.
```{r otherSetup}
# array sizes
maxrecs <- 2000
maxreps <- 1000
```
The composite curve and individual bootstrap curves can take a long time to plot. The output can be redirected to a .pdf file by setting `plotout` to `pdf`. (Set to `screen` to plot the usual way.)
```{r pdf}
# plot output
plotout <- "screen" # "screen" # "pdf"
```
# Calculation preliminaries#
## Initial steps ##
Set various output paths and filenames.
```{r pathnames2}
# no changes below here
# site list file
sitelistfile <- paste(sitelistpath, sitelist, ".csv", sep="")
sitelistfile
# curve (output) path and file
curvecsvpath <- paste(datapath,queryname,"_curves/",sep="")
# if output folder does not exist, create it
dir.create(file.path(datapath, paste(queryname,"_curves/",sep="")), showWarnings=FALSE)
curvefile <- paste(sitelist,"_locfit_",basename,"_",binname,"_",as.character(hw),"_",
as.character(nreps), ".csv", sep="")
print(curvecsvpath)
print(curvefile)
```
Code block to implement writing .pdf to file:
```{r pdf setup}
# .pdf plot of bootstrap iterations
if (plotout == "pdf") {
pdffile <- paste(sitelist,"_locfit_",basename,"_",binname,"_",as.character(hw),"_",
as.character(nreps),".pdf", sep="")
print(pdffile)
}
```
Read the list of sites to be processed. Note that this is the site list may contain sites that after transforming and presampling may have no useful data. Those sites are ignored when the data are read in.
```{r readsites}
# read the list of sites
sites <- read.csv(sitelistfile)
head(sites)
ns <- length(sites[,1]) #length(sites$ID_SITE)
ns
```
Define arrays for data, fitted values and statistics.
```{r arrayDefinition}
# arrays for data and fitted values
age <- matrix(NA, ncol=ns, nrow=maxrecs)
influx <- matrix(NA, ncol=ns, nrow=maxrecs)
nsamples <- rep(0, maxrecs)
targage <- seq(targbeg,targend,targstep)
targage.df <- data.frame(x=targage)
lowage <- targage - hw; highage <- targage + hw
ntarg <- length(targage)
yfit <- matrix(NA, nrow=length(targage.df$x), ncol=maxreps)
# arrays for sample number and effective window span tracking
ndec <- matrix(0, ncol=ntarg, nrow=ns)
ndec_tot <- rep(0, ntarg)
xspan <- rep(0, ntarg)
ninwin <- matrix(0, ncol=ntarg, nrow=ns)
ninwin_tot <- rep(0, ntarg)
```
## Read data ##
Read and store data. Note that sites without data, even if in the site list, are skipped using the `file.exists()` function.
```{r readData, results='hide'}
# read and store the presample (binned) files as matrices of ages and influx values
ii <- 0
for (i in 1:ns) {
#i <- 1
sitenum <- sites[i,1] # sites$ID_SITE[i]
print(sitenum)
siteidchar <- as.character(sitenum)
if (sitenum >= 1) siteid <- paste("000", siteidchar, sep="")
if (sitenum >= 10) siteid <- paste("00", siteidchar, sep="")
if (sitenum >= 100) siteid <- paste("0", siteidchar, sep="")
if (sitenum >= 1000) siteid <- paste( siteidchar, sep="")
inputfile <- paste(csvpath, siteid, csvname, sep="")
print(inputfile)
if (file.exists(inputfile)) {
indata <- read.csv(inputfile)
nsamp <- length(indata$age) #
if (nsamp > 0) {
ii <- ii+1
age[1:nsamp,ii] <- indata$age #
influx[1:nsamp,ii] <- indata$zt #
nsamples[ii] <- nsamp
}
}
}
nsites <- ii
```
Number of sites with data.
```{r nsites}
# number of sites with data
nsites
```
As the presample files are read, individual files will be listed:
```{r printExample, echo=FALSE}
cat(" [1] 1 \n [1] \"/Projects/GPWG/GPWGv3/GCDv3Data/v3plus/v3plus_presamp_csv/0001_presamp_influx_zt-lme_bw10.csv\" \n ... ")
```
Trim the input data to an appropriate range given the target ages, and censor (set to `NA`) any tranformed influx values greater than 10.
```{r trim}
# trim samples to age range
influx[age >= targend+hw] <- NA
age[age >= targend+hw] <- NA
# censor abs(influx) values > 10
influx[abs(influx) >= 10] <- NA
age[abs(influx) >= 10] <- NA
```
## Find number of sites and samples contributing to fitted values ##
The number of sites with samples (`ndec_tot`) and the number of samples (`ninwin_tot`) that contribute to each fitted value are calculated, along with the effective window width or "span". This code is clunky, but parallels that in the Fortran versions.
```{r ninwin}
# count number of sites that contributed to each fitted value
ptm <- proc.time()
for (i in 1:ntarg) {
agemax <- -1e32; agemin <- 1e32
for (j in 1:nsites) {
for (k in 1:nsamples[j]) {
if (!is.na(age[k,j])) {
ii <- (age[k,j]-targage[1])/targstep + 1
#print (c(i,j,k,ii))
if (ii > 0 && ii <= ntarg) {ndec[j,ii] = 1}
if (age[k,j] >= targage[i]-hw && age[k,j] <= targage[i]+hw) {
ninwin[j,i] = 1
if (agemax < age[k,j]) {agemax <- age[k,j]}
if (agemin > age[k,j]) {agemin <- age[k,j]}
}
}
}
}
ndec_tot[i] <- sum(ndec[,i])
ninwin_tot[i] <- sum(ninwin[,i])
xspan[i] <- agemax - agemin
}
proc.time() - ptm
#head(cbind(targage,ndec_tot,xspan,ninwin_tot))
```
# Curve-fitting and bootstrapping #
First, the overall composite curve, i.e. without bootstrapping, determined and saved for plotting over the individual bootstrap curves later. Second, the `nreps` individual bootstrap samples are selected, and a composit curve fitted to each and saved.
## Composite curve ##
The steps in getting this curve include:
1. Reshaping the data matrices (`age` and `influx`) into vectors
2. Running `locfit()`, and
3. Getting fitted values at the target ages
The composite curve using all data is calculated as follows:
```{r locfit}
ptm <- proc.time()
# 1. reshape matrices into vectors
x <- as.vector(age)
y <- as.vector(influx)
lfdata <- data.frame(x,y)
lfdata <- na.omit(lfdata)
x <- lfdata$x; y <- lfdata$y
# 2. locfit
# initial fit, unresampled (i.e. all) data
loc01 <- locfit(y ~ lp(x, deg=1, h=hw), maxk=800, family="qrgauss")
summary(loc01)
# 3. get fitted values
pred01 <- predict(loc01, newdata=targage.df, se.fit=TRUE)
loc01_fit <- data.frame(targage.df$x, pred01$fit)
fitname <- paste("locfit_",as.character(hw), sep="")
colnames(loc01_fit) <- c("age", fitname)
head(loc01_fit)
proc.time() - ptm
```
## Bootstrap-by-site confidence intervals ##
The bootstrap confidence intervals are obtained by sampling with replacement sites (as opposed to individual samples), fitting a composite curve using `locfit()`, and saving these. The 95-percent confidence intervals are then determined for each target age using the `quantile()` function.
```{r boostrapping, results='hide'}
# bootstrap samples
ptm <- proc.time()
# Step 1 -- Set up to plot individual replications
if (plotout == "pdf") {pdf(file=paste(curvecsvpath,pdffile,sep=""))}
plot(x, y, xlab="Age (BP 1950)", ylab=fitname, xlim=c(22000,-100), xaxp = c(22000, 0, 22), ylim=c(-1.5,0.5), type="n")
# Step 2 -- Do the bootstrap iterations, and plot each composite curve
set.seed(10) # do this to get the same sequence of random samples for each run
for (i in 1:nreps) {
print(i)
randsitenum <- sample(seq(1:nsites), nsites, replace=TRUE)
# print(head(randsitenum))
x <- as.vector(age[,randsitenum])
y <- as.vector(influx[,randsitenum])
lfdata <- data.frame(x,y)
lfdata <- na.omit(lfdata)
x <- lfdata$x; y <- lfdata$y
locboot <- locfit(y ~ lp(x, deg=1, h=hw), maxk=800, maxit=20, family="qrgauss")
predboot <- predict(locboot, newdata=targage.df, se.fit=TRUE)
yfit[,i] <- predboot$fit
# note plotting lines is slowww
lines(targage.df$x, yfit[,i], lwd=2, col=rgb(0.5,0.5,0.5,0.10))
if (i %% 10 == 0) {print(i)}
}
# Step 3 -- Plot the unresampled (initial) fit
fitname <- paste("locfit_",as.character(hw), sep="")
colnames(loc01_fit) <- c("age", fitname)
lines(loc01_fit[,1], loc01_fit[,2], lwd=2, col="red")
# Step 4 -- Find and add bootstrap CIs
yfit95 <- apply(yfit, 1, function(x) quantile(x,prob=0.975, na.rm=T))
yfit05 <- apply(yfit, 1, function(x) quantile(x,prob=0.025, na.rm=T))
lines(targage.df$x, yfit95, lwd=1, col="red")
lines(targage.df$x, yfit05, lwd=1, col="red")
if (plotout == "pdf") {dev.off()}
```
## Output ##
The fitted curves are written out in the usual way.
```{r output}
curveout <- data.frame(cbind(targage.df$x, pred01$fit, yfit95, yfit05, ndec_tot, xspan, ninwin_tot))
colnames(curveout) <- c("age", "locfit", "cu95", "cl95", "nsites", "window", "ninwin")
outputfile <- paste(curvecsvpath, curvefile, sep="")
write.table(curveout, outputfile, col.names=TRUE, row.names=FALSE, sep=",")
proc.time() - ptm
```
After fitting, it would be good to check for `locfit()` warnings (i.e. `warnings()`).