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
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
|
### Now, the question is, (1) whether to have the method for data
### frames (which is what's of immediate interest) here or in lme4.
### It doesn't really make sense to have it in lme4, since we won't be
### defining a new class. But most of the features would be borrowed
### from nlme (2) whether to have methods for any other class (like
### 'lm' maybe)
### data frame method: gplotArgs.data.frame(x, <special args>, ...)
## 4 levels of information:
## o attr(x, "ginfo")
## o <special args>
## o attr(x, "gplot.args")
## o ...
## the first 2 determine a reasonable list, overridden by the 3rd and
## then the 4th
## should we have a non-generic "gplotArgs<-" for setting the
## "gplot.args" attribute?
## First, we need some good default panel functions. These need to
## depend on
## (1) the type of display formula -
## o factor ~ numeric [ model: numeric ~ 1 | factor ]
## o numeric ~ numeric [ model: numeric ~ numeric | factor ]
## o ~ numeric [ model: factor ~ numeric | factor ] NEW
panel.df.fn <- # factor ~ numeric
function(x, y, groups = NULL, ...)
{
panel.dotplot(x=x, y=y, groups = groups, ...)
}
panel.df.nn <- # numeric ~ numeric
function(x, y, lines = all(type == "p"), type = "p", ...)
{
panel.xyplot(x, y, type = type, ...)
if (lines)
{
y.avg <- tapply(y, x, mean) # lines through average y
y.avg <- y.avg[!is.na(y.avg)]
if (length(y.avg) > 0)
{
xvals <- as.numeric(names(y.avg))
ord <- order(xvals)
panel.xyplot(xvals[ord], y.avg[ord], type = "l", ...)
}
}
}
panel.df <- # combines above 2, should be called
function(x, y = NULL, groups = NULL, grid = TRUE, ...)
{
yNull <- is.null(y)
groupsNull <- is.null(groups)
xFactor <- is.factor(x)
yFactor <- is.factor(y)
if (yNull) ## for factor response, may not be in use yet
{
if (groupsNull) panel.densityplot(x = x, ...)
else panel.superpose(x = x, groups = groups,
panel.groups = "panel.densityplot", ...)
}
else if (!xFactor && !yFactor) ## numeric ~ numeric
{
if (grid) panel.grid()
if (groupsNull) panel.df.nn(x = x, y = y, ...)
else panel.superpose(x = x, y = y, groups = groups,
panel.groups = panel.df.nn, ...)
}
else if (yFactor && !xFactor)
{
panel.df.fn(x = x, y = y, groups = groups, ...)
}
else stop("can't handle both x and y being factors yet")
invisible()
}
## display formula
## generally, the display is determined by the arguments 'formula' and
## 'groups'. In the nlme scheme, the formula associated with the data
## is a model formula, not a display formula. The display is further
## determined by the 'outer' and 'inner' arguments.
## Let's try to outline a strategy for this.
## attr(x, "ginfo") can contain things used in nlme's groupedData.
## formula can be of the form 'y ~ x | id' (only one variable in 'id',
## although it can be an interaction). inner=~a becomes default
## grouping factor. outer=~b+c used for ordering levels of id by
## values of y. outer can be also an argument to
## gplotArgs.data.frame. If TRUE, it's equivalent to outer =
## ginfo$outer, or it could be =~e+f. In either case, it becomes the
## conditioning variables, id becomes groups (inner would then be
## ignored)
## In the formula itself, if x is a factor, the display formula
## becomes id ~ y, with groups = x (unless x = 1, then no groups)
## .typeInDF <- function(x, data)
## {
## if (x == "1") "one"
## else if (is.factor(data[[x]])) "factor"
## else if (is.numeric(data[[x]])) "numeric"
## else stop(paste("don't recognize", class(data[[x]])))
## }
## should work with expressions like log(height) as well
.typeInDF <- function(x, data)
{
if (all(is.na(x))) return(NA)
x <- eval(parse(text = x), data, parent.frame())
if (length(x) == 1 && x == 1) "one"
else if (is.factor(x)) "factor"
else if (is.numeric(x)) "numeric"
else stop(paste("don't recognize", class(x)))
}
## S3 method: underlies groupedData() in lme4
## FIXME: the following doesn't work in the docs. What's the recommended way?
## \method{"gplotArgs<-"}{data.frame}(x, value)
"gplotArgs<-.data.frame" <-
function(x, value)
{
if (!is.list(value)) stop("assigned value must be list")
process.args <-
function(formula,
order.groups = TRUE,
FUN = function(x, ...) max(x, na.rm = TRUE),
outer = NULL, inner = NULL,
labels = list(), units = list(), ...)
{
list(ginfo =
list(formula = formula,
order.groups = order.groups,
FUN = FUN,
outer = outer,
inner = inner,
labels = labels,
units = units),
dots = list(...))
}
pargs <- do.call("process.args", value)
attr(x, "ginfo") <- pargs$ginfo
if (length(pargs$dots) > 0)
attr(x, "gplot.args") <- pargs$dots
x
}
## S3 method
## basic idea: get defaults from "ginfo" attribute, then overwrite by
## "gplot.args" attribute, followed by ...
gplotArgs.data.frame <-
function(x, display.formula, outer = FALSE, inner = FALSE,
groups = NULL,
...,
subset = TRUE)
{
## The final result should contain only standard trellis args,
## with a special component plotFun, and optionally a
## display.formula, which overrides formula. However, for data
## frames, there are some other issues.
## The "ginfo" attribute can only contain info traditionally in
## nlme groupedData objects. This, along with explicit arguments
## here, will be used to create a list of trellis args. These can
## be overridden by the "gplot.args" attribute, and then by
## ... here.
ginfo <- attr(x, "ginfo")
gplot.args <- attr(x, "gplot.args")
## equivalent to default method if ginfo is NULL
if (is.null(ginfo) || !is.list(ginfo))
return(updateList(gplot.args, list(...)))
## First (longish) task: determine display formula and groups
## groups is by far the most irritating thing to handle. The only
## options I can think of: (1) evaluate groups here and pass it on
## and (2) change lattice to allow groups to be a formula. Use
## (1) for now
groups <- eval(substitute(groups), x, parent.frame()) # typically NULL
## subset poses a similar problem
subset <- eval(substitute(subset), x, parent.frame())
if (missing(display.formula))
display.formula <- gplot.args$display.formula
## major step: get the display formula, but only if it's NULL
model.formula <- ginfo$formula
if (!is.null(display.formula))
## no point in jumping through hoops
{
ans <- list(display.formula = display.formula)
}
else if (!is.null(model.formula)) ## the interesting stuff
{
vars <-
list(resp = .responseName(model.formula),
cov = .covariateName(model.formula),
grp = .groupsName(model.formula))
if (ginfo$order.groups)
{
## reorder grp based on values of resp
if (is.null(ginfo$FUN)) ginfo$FUN <- function(x, ...) max(x, na.rm = TRUE)
respVar <- vars$resp
grpVar <- vars$grp
scores <- tapply(x[[respVar]], x[[grpVar]], ginfo$FUN)
if (inherits(ginfo$outer, "formula"))
{
outerVar <- .covariateName(ginfo$outer)
outer.unique <- tapply(x[[outerVar]], x[[grpVar]], unique)
ord <- order(outer.unique, scores)
}
else
ord <- order(scores)
x[[grpVar]] <- factor(x[[grpVar]], levels = names(scores)[ord])
}
## display formula may be further modified by inner and outer.
## How does that affect rest of the calculations?
if (is.logical(outer) && outer) outer <- ginfo$outer
if (is.logical(inner) && inner) inner <- ginfo$inner
## both cannot happen. outer makes outer the conditioning
## variables, normal grp becomes groups. inner behaves as
## groups. outer takes precedence.
if (inherits(outer, "formula"))
{
if (is.null(groups)) groups <- as.formula(paste("~", vars$grp))
vars$grp <- .covariateName(outer)
}
else if (inherits(inner, "formula"))
{
## FIXME: this may not be the right thing to do
if (is.null(groups)) groups <- inner
}
varTypes <-
lapply(vars, .typeInDF, data = x)
## Next step depends on varTypes
## case 1: cov = "one" - grp ~ resp
## case 2: cov = "factor" - grp ~ resp, groups = cov
## case 1: cov = "numeric" - resp ~ cov | grp
fc <- switch(varTypes$cov,
one = paste(vars$grp, "~", vars$resp),
factor = paste(vars$grp, "~", vars$resp),
numeric =
paste(vars$resp, "~", vars$cov, "|", vars$grp))
ans <-
list(display.formula = as.formula(fc))
if (varTypes$cov == "factor" && is.null(groups))
groups <-
eval(parse(text = vars$cov), x, parent.frame())
}
else
{
stop("no formula!")
}
## determine default plot function based on display.formula
dvars <-
list(resp = .responseName(ans$display.formula),
cov = .covariateName(ans$display.formula),
grp = .groupsName(ans$display.formula)) ## NA is none
dvarTypes <-
lapply(dvars, .typeInDF, data = x)
if (dvarTypes$resp == "numeric" && dvarTypes$cov == "numeric")
plotFun.constructed <- "xyplot"
else if (dvarTypes$resp == "factor" && dvarTypes$cov == "numeric")
plotFun.constructed <- "dotplot"
else {
str(dvarTypes)
stop("unsupported combination")
}
## other stuff in ginfo?
ylab.constructed <-
if ("labels" %in% names(ginfo) && dvars$resp %in% names(ginfo$labels))
ginfo$labels[[dvars$resp]]
else dvars$resp
xlab.constructed <-
if ("labels" %in% names(ginfo) && dvars$cov %in% names(ginfo$labels))
ginfo$labels[[dvars$cov]]
else dvars$cov
if ("units" %in% names(ginfo) && dvars$resp %in% names(ginfo$units))
ylab.constructed <- paste(ylab.constructed, ginfo$units[[dvars$resp]])
if ("units" %in% names(ginfo) && dvars$cov %in% names(ginfo$units))
xlab.constructed <- paste(xlab.constructed, ginfo$units[[dvars$cov]])
ans <-
updateList(ans,
list(plotFun = plotFun.constructed,
data = x,
panel = panel.df,
groups = groups,
subset = subset,
xlab = xlab.constructed, ylab = ylab.constructed,
aspect = if (plotFun.constructed == "xyplot") "xy" else "fill",
auto.key =
switch(plotFun.constructed,
xyplot = list(points = FALSE, lines = TRUE, space = "right"),
dotplot = list(points = TRUE, space = "right"))))
if (!is.null(gplot.args))
ans <- updateList(ans, gplot.args) ## leave this out?
updateList(ans, list(...))
}
|