Discussion Section of 1/9/13: 1.1 R Demo

Chapter 1
Discussion Section of 1/9/13

1.1
1
R demo
grad.desc = function(FUN = function(x, y) x^2 + 2 *
2
3
y^2, rg = c(-3, -3, 3, 3), init = c(-3, 3), gamma = 0.05,
4
5
tol = 0.001, gr = NULL, len = 50, interact = FALSE, col.contour = "red",
6
7
col.arrow = "blue", main) {
8
9
nmax = ani.options("nmax")
10
11
x = seq(rg[1], rg[3], length = len)
12
13
y = seq(rg[2], rg[4], length = len)
14
15
nms = names(formals(FUN))
16
17
grad = if (is.null(gr))
18
19
deriv(as.expression(body(FUN)), nms, function.arg = TRUE) else {
20
function(...) {
21
22
res = FUN(...)
23
24
attr(res, gradient) = matrix(gr(...), nrow = 1, ncol = 2)
25
26
res
27
28
29
30
31
32
33
z = outer(x, y, FUN)
CHAPTER 1. DISCUSSION SECTION OF 1/9/13
34
35
if (interact) {
36
contour(x, y, z, col = "red", xlab = nms[1], ylab = nms[2],
37
38
main = "Choose initial values by clicking on the graph")
39
40
xy = unlist(locator(1))
41
42
43
} else {
44
xy = init
45
46
47
48
49
newxy = xy - gamma * attr(grad(xy[1], xy[2]), "gradient")
50
51
gap = abs(FUN(newxy[1], newxy[2]) - FUN(xy[1], xy[2]))
52
53
if (missing(main))
54
main = eval(substitute(expression(z == x), list(x = body(FUN))))
55
56
57
i = 1
58
59
while (gap > tol && i <= nmax) {
60
#dev.hold()
61
62
contour(x, y, z, col = col.contour, xlab = nms[1], ylab = nms[2],
63
64
main = main)
65
66
xy = rbind(xy, newxy[i, ])
67
68
newxy = rbind(newxy, xy[i + 1, ] - gamma * attr(grad(xy[i +
69
70
1, 1], xy[i + 1, 2]), "gradient"))
71
72
arrows(xy[1:i, 1], xy[1:i, 2], newxy[1:i, 1], newxy[1:i,
73
74
2], length = par("din")[1] / 50, col = col.arrow)
75
76
gap = abs(FUN(newxy[i + 1, 1], newxy[i + 1, 2]) - FUN(xy[i +
77
78
1, 1], xy[i + 1, 2]))
79
80
ani.pause()
81
82
i = i + 1
83
84
if (i >= nmax) warning(Maximum number of iterations reached!)
85
86
87
88
89
invisible(list(par = newxy[i - 1, ], value = FUN(newxy[i -
90
91
1, 1], newxy[i - 1, 2]), iter = i - 1, gradient = grad,
1.1.
R DEMO
92
persp = function(...) persp(x, y, z, ...)))
93
94
95
96
97
grad.desc()
98
99
100
101
# source("z.r")
102
103
104
105
#install package mvtnorm
106
107
108
109
sigma <- matrix(c(4,2,2,3), ncol=2)
110
111
x <- rmvnorm(n=500, mean=c(1,2), sigma=sigma)
112
113
colMeans(x)
114
115
var(x)
116
117
118
119
#install.packages("mvtnorm","/a/b/c/")
120
121
#.libPaths("/a/b/c/")
122
123
library(mvtnorm)
124
125
126
127
sigma <- matrix(c(4,2,2,3), ncol=2)
128
129
x <- rmvnorm(n=500, mean=c(1,2), sigma=sigma)
130
131
colMeans(x)
132
133
var(x)
134
135
136
137
#vector
138
139
140
141
a = c(1,2)
142
143
144
145
#list
146
147
Nile
148
149
hn <- hist(Nile)
150
151
152
153
hn
154
155
156
157
hn[1]
158
159
hn[[1]]
160
161
hn[[1]][1]
162
163
164
165
# Matrix construction from dataframe
166
167
168
169
PSNR = data.frame(cbind(c(0.72761,0.70622,0.713415,0.707539),c(0.8625723,0.835956,0.924474,0.909785),c(0.9254554,0.901021,0.9521
170
171
172
173
PSNR
174
175
176
177
P = as.matrix(PSNR)
178
179
180
181
182
183
184
185
str(P)
186
187
188
189
class(P)
190
191
192
193
dim(P)
194
195
196
197
#matrix construction from scratch
198
199
200
201
P = matrix(nrow = 5 ,ncol= 5)
202
203
204
205
206
207
str(P) #structure
1.1.
R DEMO
208
209
class(P) #class
210
211
212
213
dim(P)
214
215
216
217
218
219
220
221
P[1,1] = 5
222
223
224
225
226
227
228
229
P = matrix(c(1,2,3,4,5,6,7,8,9,0,1,2,3,4,5,6,7,8,9,0,1,2,3,4,5),nrow = 5, ncol =5)
230
231
232
233
234
235
236
237
#Matrix construction using rbind / cbind
238
239
240
241
P = rbind(c(1,2,3,4,5),c(6,7,8,9,0))
242
243
244
245
246
247
248
249
P = cbind(c(1,2,3,4,5),c(6,7,8,9,0))
250
251
252
253
254
255
256
257
#Adding to a matrix
258
259
260
261
P = rbind(c(1,2,3,4,5),c(6,7,8,9,0))
262
263
264
265
P = rbind(P,P)
266
267
268
269
270
271
272
273
#matrix indexing
274
275
276
277
278
279
280
281
P[1,]
282
283
284
285
P[,1]
286
287
288
289
P[1:4,]
290
291
292
293
P[,1:4]
294
295
296
297
#Submatrices from a matrix
298
299
300
301
302
303
P[c(2,4),]
304
305
306
307
P[c(2,4),c(1,3)]
308
309
310
311
#Copying submatrices
312
313
314
315
P[2:3,] = rep(100,10)
316
317
318
319
320
321
322
323
nrow(P)
1.1.
R DEMO
324
325
326
327
#Deleting from a matrix
328
329
330
331
332
333
334
335
P = P[,1:4]
336
337
338
339
340
341
#Matrix addition / subtraction
342
343
344
345
P + P
346
347
348
349
#Scalar multiplication/div
350
351
352
353
5 * P
354
355
356
357
#component by component mult/div
358
359
360
361
P * P
362
363
364
365
V = c(1,2,3,4,5)
366
367
368
369
P * V #row i multiplied by component i
370
371
372
373
#multiply column i by component i
374
375
376
377
V = V[-5]
378
379
380
381
t(t(P) *V)
382
383
384
385
#matrix multiplication
386
387
388
389
P %*% t(P)
390
391
392
393
#matrix multiplication with a vector
394
395
396
397
V = c(1,2,3,4) #column vector
398
399
400
401
P %*% V
402
403
404
405
# Outer product
406
407
408
409
V %o% V
410
411
412
413
#inner product
414
415
416
417
crossprod(V,V) #actually returns t(A) %*% B
418
419
420
421
#Inverse of a matrix
422
423
P = cbind(P,c(1,2,0,1,0))
424
425
426
427
P[,3] = c(13,14,17,19,23)
428
429
430
431
P[,4] = c(29,31,37,43,47)
432
433
434
435
solve(P)
436
437
438
439
#Solving Linear systems
1.1.
R DEMO
440
441
442
443
444
445
V = c(1,2,3,4,5)
446
447
448
449
P[,3] = c(13,14,17,19,23)
450
451
452
453
P[,4] = c(29,31,37,43,47)
454
455
456
457
P[,5] = c(30,31,37,43,47)
458
459
460
461
solve(P,V)
462
463
464
465
#Eigen values and Eigen vector
466
467
468
469
470
471
472
473
y = eigen(P)
474
475
476
477
#obtaining stationary distributions using eigenvalues
478
479
480
481
p <- matrix ( rep ( 0 , 100 ) , nrow=10)
482
483
onesixth <- 1/6
484
485
for ( i in 1 : 10 ) {
486
487
for ( j in 1 : 6 ) {
488
489
k <- i + j
490
491
if ( k > 10) k <- k - 10
492
493
p[ i , k ] <- onesixth
494
495
496
497
10
498
499
500
501
findpi1 <- function(p) {
502
503
n <- nrow(p)
504
505
imp <- diag(n) - t(p) # I-P
506
507
imp[n,] <- rep(1,n) # insert row of 1s
508
509
rhs <- c(rep(0,n-1),1) # form right-hand side
510
511
pivec <- solve(imp,rhs) # solve the system
512
513
return(pivec)
514
515
516
517
518
519
n <- nrow(p)
520
521
522
523
pivec <- eigen(t(p))$vectors[,1] # find first eigenvector of P transpose
524
525
526
527
if (Re(pivec[1]) < 0) pivec <- -Re(pivec) #could be negative
528
529
530
531
532
533
pivec <- pivec / sum(pivec) # normalize to sum to 1

Discussion Section of 1/9/13: 1.1 R Demo

Caricato da

Informazioni sul documento

Descrizione originale:

Titolo originale

Copyright

Formati disponibili

Condividi questo documento

Condividi o incorpora il documento

Opzioni di condivisione

Hai trovato utile questo documento?

Questo contenuto è inappropriato?

Copyright:

Formati disponibili

Discussion Section of 1/9/13: 1.1 R Demo

Caricato da

Copyright:

Formati disponibili

Chapter 1

Discussion Section of 1/9/13

grad.desc = function(FUN = function(x, y) x^2 + 2 *

y^2, rg = c(-3, -3, 3, 3), init = c(-3, 3), gamma = 0.05,

tol = 0.001, gr = NULL, len = 50, interact = FALSE, col.contour = "red",

col.arrow = "blue", main) {

x = seq(rg[1], rg[3], length = len)

y = seq(rg[2], rg[4], length = len)

deriv(as.expression(body(FUN)), nms, function.arg = TRUE) else {

attr(res, gradient) = matrix(gr(...), nrow = 1, ncol = 2)

CHAPTER 1. DISCUSSION SECTION OF 1/9/13

contour(x, y, z, col = "red", xlab = nms[1], ylab = nms[2],

main = "Choose initial values by clicking on the graph")

newxy = xy - gamma * attr(grad(xy[1], xy[2]), "gradient")

gap = abs(FUN(newxy[1], newxy[2]) - FUN(xy[1], xy[2]))

main = eval(substitute(expression(z == x), list(x = body(FUN))))

while (gap > tol && i <= nmax) {

contour(x, y, z, col = col.contour, xlab = nms[1], ylab = nms[2],

newxy = rbind(newxy, xy[i + 1, ] - gamma * attr(grad(xy[i +

1, 1], xy[i + 1, 2]), "gradient"))

arrows(xy[1:i, 1], xy[1:i, 2], newxy[1:i, 1], newxy[1:i,

2], length = par("din")[1] / 50, col = col.arrow)

gap = abs(FUN(newxy[i + 1, 1], newxy[i + 1, 2]) - FUN(xy[i +

1, 1], xy[i + 1, 2]))

if (i >= nmax) warning(Maximum number of iterations reached!)

invisible(list(par = newxy[i - 1, ], value = FUN(newxy[i -

1, 1], newxy[i - 1, 2]), iter = i - 1, gradient = grad,

persp = function(...) persp(x, y, z, ...)))

#install package mvtnorm

sigma <- matrix(c(4,2,2,3), ncol=2)

x <- rmvnorm(n=500, mean=c(1,2), sigma=sigma)

sigma <- matrix(c(4,2,2,3), ncol=2)

x <- rmvnorm(n=500, mean=c(1,2), sigma=sigma)

CHAPTER 1. DISCUSSION SECTION OF 1/9/13

# Matrix construction from dataframe

#matrix construction from scratch

P = matrix(c(1,2,3,4,5,6,7,8,9,0,1,2,3,4,5,6,7,8,9,0,1,2,3,4,5),nrow = 5, ncol =5)

#Matrix construction using rbind / cbind

CHAPTER 1. DISCUSSION SECTION OF 1/9/13

P = matrix(c(1,2,3,4,5,6,7,8,9,0,1,2,3,4,5,6,7,8,9,0,1,2,3,4,5),nrow = 5, ncol =5)

#Submatrices from a matrix

P = matrix(c(1,2,3,4,5,6,7,8,9,0,1,2,3,4,5,6,7,8,9,0,1,2,3,4,5),nrow = 5, ncol =5)

#Deleting from a matrix

P = matrix(c(1,2,3,4,5,6,7,8,9,0,1,2,3,4,5,6,7,8,9,0,1,2,3,4,5),nrow = 5, ncol =5)

#Matrix addition / subtraction

#component by component mult/div

P * V #row i multiplied by component i

#multiply column i by component i

#matrix multiplication with a vector

V = c(1,2,3,4) #column vector

crossprod(V,V) #actually returns t(A) %*% B

#Solving Linear systems

CHAPTER 1. DISCUSSION SECTION OF 1/9/13

P = matrix(c(1,2,3,4,5,6,7,8,9,0,1,2,3,4,5,6,7,8,9,0,1,2,3,4,5),nrow = 5, ncol =5)

#Eigen values and Eigen vector

P = matrix(c(1,2,3,4,5,6,7,8,9,0,1,2,3,4,5,6,7,8,9,0,1,2,3,4,5),nrow = 5, ncol =5)

#obtaining stationary distributions using eigenvalues

p <- matrix ( rep ( 0 , 100 ) , nrow=10)

onesixth <- 1/6

if ( k > 10) k <- k - 10

CHAPTER 1. DISCUSSION SECTION OF 1/9/13

findpi1 <- function(p) {