forked from AdamWilsonLabEDU/SpatialDataScience
-
Notifications
You must be signed in to change notification settings - Fork 0
Expand file tree
/
Copy pathParallelProcessing.html
More file actions
959 lines (883 loc) · 36.6 KB
/
ParallelProcessing.html
File metadata and controls
959 lines (883 loc) · 36.6 KB
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
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715
716
717
718
719
720
721
722
723
724
725
726
727
728
729
730
731
732
733
734
735
736
737
738
739
740
741
742
743
744
745
746
747
748
749
750
751
752
753
754
755
756
757
758
759
760
761
762
763
764
765
766
767
768
769
770
771
772
773
774
775
776
777
778
779
780
781
782
783
784
785
786
787
788
789
790
791
792
793
794
795
796
797
798
799
800
801
802
803
804
805
806
807
808
809
810
811
812
813
814
815
816
817
818
819
820
821
822
823
824
825
826
827
828
829
830
831
832
833
834
835
836
837
838
839
840
841
842
843
844
845
846
847
848
849
850
851
852
853
854
855
856
857
858
859
860
861
862
863
864
865
866
867
868
869
870
871
872
873
874
875
876
877
878
879
880
881
882
883
884
885
886
887
888
889
890
891
892
893
894
895
896
897
898
899
900
901
902
903
904
905
906
907
908
909
910
911
912
913
914
915
916
917
918
919
920
921
922
923
924
925
926
927
928
929
930
931
932
933
934
935
936
937
938
939
940
941
942
943
944
945
946
947
948
949
950
951
952
953
954
955
956
957
958
959
<!DOCTYPE html>
<html xmlns="http://www.w3.org/1999/xhtml">
<head>
<meta charset="utf-8">
<meta http-equiv="Content-Type" content="text/html; charset=utf-8" />
<meta name="generator" content="pandoc" />
<title>Parallel Computing with R</title>
<script src="site_libs/jquery-1.11.3/jquery.min.js"></script>
<meta name="viewport" content="width=device-width, initial-scale=1" />
<link href="site_libs/bootstrap-3.3.5/css/cosmo.min.css" rel="stylesheet" />
<script src="site_libs/bootstrap-3.3.5/js/bootstrap.min.js"></script>
<script src="site_libs/bootstrap-3.3.5/shim/html5shiv.min.js"></script>
<script src="site_libs/bootstrap-3.3.5/shim/respond.min.js"></script>
<link href="site_libs/font-awesome-4.5.0/css/font-awesome.min.css" rel="stylesheet" />
<style type="text/css">code{white-space: pre;}</style>
<link rel="stylesheet"
href="site_libs/highlight/textmate.css"
type="text/css" />
<script src="site_libs/highlight/highlight.js"></script>
<style type="text/css">
pre:not([class]) {
background-color: white;
}
</style>
<script type="text/javascript">
if (window.hljs && document.readyState && document.readyState === "complete") {
window.setTimeout(function() {
hljs.initHighlighting();
}, 0);
}
</script>
<style type="text/css">
h1 {
font-size: 34px;
}
h1.title {
font-size: 38px;
}
h2 {
font-size: 30px;
}
h3 {
font-size: 24px;
}
h4 {
font-size: 18px;
}
h5 {
font-size: 16px;
}
h6 {
font-size: 12px;
}
.table th:not([align]) {
text-align: left;
}
</style>
<link rel="stylesheet" href="styles.css" type="text/css" />
</head>
<body>
<style type = "text/css">
.main-container {
max-width: 940px;
margin-left: auto;
margin-right: auto;
}
code {
color: inherit;
background-color: rgba(0, 0, 0, 0.04);
}
img {
max-width:100%;
height: auto;
}
.tabbed-pane {
padding-top: 12px;
}
button.code-folding-btn:focus {
outline: none;
}
</style>
<style type="text/css">
/* padding for bootstrap navbar */
body {
padding-top: 51px;
padding-bottom: 40px;
}
/* offset scroll position for anchor links (for fixed navbar) */
.section h1 {
padding-top: 56px;
margin-top: -56px;
}
.section h2 {
padding-top: 56px;
margin-top: -56px;
}
.section h3 {
padding-top: 56px;
margin-top: -56px;
}
.section h4 {
padding-top: 56px;
margin-top: -56px;
}
.section h5 {
padding-top: 56px;
margin-top: -56px;
}
.section h6 {
padding-top: 56px;
margin-top: -56px;
}
</style>
<script>
// manage active state of menu based on current page
$(document).ready(function () {
// active menu anchor
href = window.location.pathname
href = href.substr(href.lastIndexOf('/') + 1)
if (href === "")
href = "index.html";
var menuAnchor = $('a[href="' + href + '"]');
// mark it active
menuAnchor.parent().addClass('active');
// if it's got a parent navbar menu mark it active as well
menuAnchor.closest('li.dropdown').addClass('active');
});
</script>
<div class="container-fluid main-container">
<!-- tabsets -->
<script src="site_libs/navigation-1.1/tabsets.js"></script>
<script>
$(document).ready(function () {
window.buildTabsets("TOC");
});
</script>
<!-- code folding -->
<div class="navbar navbar-inverse navbar-fixed-top" role="navigation">
<div class="container">
<div class="navbar-header">
<button type="button" class="navbar-toggle collapsed" data-toggle="collapse" data-target="#navbar">
<span class="icon-bar"></span>
<span class="icon-bar"></span>
<span class="icon-bar"></span>
</button>
<a class="navbar-brand" href="index.html">GEO 503: R Spatial Data Science</a>
</div>
<div id="navbar" class="navbar-collapse collapse">
<ul class="nav navbar-nav">
<li>
<a href="index.html">Home</a>
</li>
<li>
<a href="Syllabus.html">Syllabus</a>
</li>
<li>
<a href="Schedule.html">Schedule</a>
</li>
<li class="dropdown">
<a href="#" class="dropdown-toggle" data-toggle="dropdown" role="button" aria-expanded="false">
Content
<span class="caret"></span>
</a>
<ul class="dropdown-menu" role="menu">
<li>
<a href="CourseContent.html">About Course Content</a>
</li>
<li class="dropdown-header">Module 1: Introduction to R</li>
<li>
<a href="00_CourseIntroductionFrame.html">00 Course Introduction</a>
</li>
<li>
<a href="01_Rintro.html">01 First Steps</a>
</li>
<li>
<a href="02_DataWrangling.html">02 Data Wrangling</a>
</li>
<li>
<a href="03_Plotting.html">03 Plotting</a>
</li>
<li class="divider"></li>
<li class="dropdown-header">Module 2: Spatial Analyses</li>
<li>
<a href="04_Spatial.html">04 Spatial Data</a>
</li>
<li>
<a href="05_Raster.html">05 Spatial Raster Data</a>
</li>
<li>
<a href="06_RasterTwo.html">06 More Raster Data</a>
</li>
<li class="divider"></li>
<li class="dropdown-header">Module 2: Advanced Programming</li>
<li>
<a href="07_Reproducibile.html">07 Reproducible Research</a>
</li>
<li>
<a href="08_WeatherData.html">08 Weather Data Processing</a>
</li>
<li>
<a href="09_RemoteSensing.html">09 Satellite Data Processing</a>
</li>
<li>
<a href="10_ParallelProcessing.html">10 Parallel Processing</a>
</li>
<li>
<a href="11_Git.html">11 Version Control with Git</a>
</li>
<li class="divider"></li>
</ul>
</li>
<li class="dropdown">
<a href="#" class="dropdown-toggle" data-toggle="dropdown" role="button" aria-expanded="false">
Assignments
<span class="caret"></span>
</a>
<ul class="dropdown-menu" role="menu">
<li>
<a href="Homework.html">Homeworks</a>
</li>
<li>
<a href="PackageIntro.html">Package Introduction</a>
</li>
<li>
<a href="Project.html">Final Project</a>
</li>
</ul>
</li>
<li class="dropdown">
<a href="#" class="dropdown-toggle" data-toggle="dropdown" role="button" aria-expanded="false">
Resources
<span class="caret"></span>
</a>
<ul class="dropdown-menu" role="menu">
<li>
<a href="Resources.html">Resources</a>
</li>
<li>
<a href="GitSSHNotes.html">Setting up Github</a>
</li>
</ul>
</li>
</ul>
<ul class="nav navbar-nav navbar-right">
<li>
<a href="https://github.com/adammwilson/RDataScience">
<span class="fa fa-github fa-lg"></span>
</a>
</li>
</ul>
</div><!--/.nav-collapse -->
</div><!--/.container -->
</div><!--/.navbar -->
<div class="fluid-row" id="header">
<h1 class="title toc-ignore">Parallel Computing with R</h1>
</div>
<pre class="r"><code>library(knitr)
library(raster)
library(rasterVis)
library(dplyr)
library(ggplot2)
## New Packages
library(foreach)
library(doParallel)
library(arm)
library(coda)
library(ggmcmc)
library(fields)
library(snow)</code></pre>
<p>If you don’t have the packages above, install them in the package manager or by running <code>install.packages("doParallel")</code>.</p>
<div id="introduction" class="section level1">
<h1>Introduction</h1>
<div id="serial-computing" class="section level2">
<h2>Serial Computing</h2>
<p>Most (legacy) software is written for serial computation:</p>
<ul>
<li>Problem broken into discrete set of instructions</li>
<li>Instructions executed sequentially on a single processor</li>
</ul>
<p><img src="assets/serialProblem.gif" alt="https://computing.llnl.gov/tutorials/parallel_comp/" /> <span style="color:grey; font-size:1em;">Figure from <a href="https://computing.llnl.gov/tutorials/parallel_comp/">here</a> </span></p>
</div>
<div id="parallel-computation" class="section level2">
<h2>Parallel computation</h2>
<p>Parallel computing is the simultaneous use of multiple compute resources:</p>
<ul>
<li>Problem divided into discrete parts that can be solved concurrently</li>
<li>Instructions from each part execute simultaneously on different processors</li>
<li>An overall control/coordination mechanism is employed</li>
</ul>
<p><img src="assets/parallelProblem.gif" alt="https://computing.llnl.gov/tutorials/parallel_comp/" /> <span style="color:grey; font-size:1em;">Figure from <a href="https://computing.llnl.gov/tutorials/parallel_comp/">here</a> </span></p>
</div>
<div id="flynns-taxonomy" class="section level2">
<h2>Flynn’s taxonomy</h2>
<p>A classification of computer architectures (<a href="http://dx.doi.org/10.1109/TC.1972.5009071">Flynn, 1972</a>)</p>
<ul>
<li><em>Single Instruction, Single Data (SISD)</em>
<ul>
<li>No parallelization</li>
</ul></li>
<li><em>Single Instruction, Multiple Data (SIMD)</em>
<ul>
<li>Run the same code/analysis on different datasets</li>
<li>Examples:
<ul>
<li>different species in species distribution model</li>
<li>same species under different climates</li>
<li>different MCMC chains from a Bayesian Model</li>
</ul></li>
</ul></li>
<li><em>Multiple Instruction, Single Data (MISD)</em>
<ul>
<li>Run different code/analyses on the same data</li>
<li>Examples:
<ul>
<li>One species, multiple models</li>
</ul></li>
</ul></li>
<li><em>Multiple Instruction, Multiple Data streams (MIMD)</em>
<ul>
<li>Run different code/analyses on different data</li>
<li>Examples:
<ul>
<li>Different species & different models</li>
</ul></li>
</ul></li>
</ul>
<p><img src="assets/SISD.png" alt="http://en.wikipedia.org/wiki/Flynn%27s_taxonomy" /> <span style="color:grey; font-size:1em;">Figure from <a href="http://en.wikipedia.org/wiki/Flynn%27s_taxonomy">here</a></span></p>
</div>
<div id="our-focus-single-instruction-multiple-data-simd" class="section level2">
<h2>Our focus: <em>Single Instruction, Multiple Data (SIMD)</em></h2>
<ol style="list-style-type: decimal">
<li>Parallel functions within an R script
<ul>
<li>starts on single processor</li>
<li>runs looped elements on multiple ‘slave’ processors</li>
<li>returns results of all iterations to the original instance</li>
<li>foreach, multicore, plyr, raster</li>
</ul></li>
<li>Alternative: run many separate instances of R in parallel with <code>Rscript</code>
<ul>
<li>need another operation to combine the results</li>
<li>preferable for long, complex jobs</li>
<li>NOT planning to discuss in this session</li>
</ul></li>
</ol>
<div id="r-packages" class="section level3">
<h3>R Packages</h3>
<p>There are many R packages for parallelization, check out the CRAN Task View on <a href="http://cran.r-project.org/web/views/HighPerformanceComputing.html">High-Performance and Parallel Computing</a> for an overview. For example:</p>
<ul>
<li><a href="http://cran.r-project.org/web/packages/Rmpi/index.html">Rmpi</a>: Built on MPI (Message Passing Interface), a de facto standard in parallel computing.</li>
<li><a href="http://cran.r-project.org/web/packages/snow/index.html">snow</a>: Simple Network of Workstations can use several standards (PVM, MPI, NWS)</li>
<li><a href="https://stat.ethz.ch/R-manual/R-devel/library/parallel/doc/parallel.pdf">parallel</a> Built in R package (since v2.14.0).</li>
</ul>
<hr />
</div>
</div>
<div id="foreach-package" class="section level2">
<h2>Foreach Package</h2>
<p>In this session we’ll focus on the foreach package, which has numerous advantages including:</p>
<ul>
<li>intuitive <code>for()</code> loop-like syntax</li>
<li>flexibility of choosing a parallel ‘backend’ for laptops through to supercomputers (using multicore, parallel, snow, Rmpi, etc.)</li>
<li>nice options for combining output from parallelized jobs</li>
</ul>
<div id="documentation-for-foreach" class="section level3">
<h3>Documentation for foreach:</h3>
<ul>
<li><a href="http://cran.r-project.org/web/packages/foreach/foreach.pdf">foreach manual</a></li>
<li><a href="http://cran.r-project.org/web/packages/foreach/vignettes/foreach.pdf">foreach vignette</a></li>
<li><a href="http://cran.r-project.org/web/packages/foreach/vignettes/nested.pdf">Nested Loops</a></li>
</ul>
</div>
<div id="foreach-backends" class="section level3">
<h3>Foreach <em>backends</em></h3>
<ul>
<li><a href="http://cran.r-project.org/web/packages/doParallel/index.html">doParallel</a> best for use on multicore machines (uses <code>fork</code> on linux/mac and <code>snow</code> on windows).</li>
<li><a href="http://cran.r-project.org/web/packages/doMPI/vignettes/doMPI.pdf">doMPI</a>: Interface to MPI (Message-Passing Interface)</li>
<li><a href="http://cran.r-project.org/web/packages/doSNOW/doSNOW.pdf">doSNOW</a>: Simple Network of Workstations</li>
</ul>
</div>
</div>
</div>
<div id="simple-examples" class="section level1">
<h1>Simple examples</h1>
<div id="sequential-for-loop" class="section level2">
<h2><em>Sequential</em> <code>for()</code> loop</h2>
<pre class="r"><code>x=vector()
for(i in 1:3) x[i]=i^2
x</code></pre>
</div>
<div id="sequential-foreach-loop" class="section level2">
<h2><em>Sequential</em> <code>foreach()</code> loop</h2>
<pre class="r"><code>x <- foreach(i=1:3) %do% i^2
x</code></pre>
<p>Note that <code>x</code> is a list with one element for each iterator variable (<code>i</code>). You can also specify a function to use to combine the outputs with <code>.combine</code>. Let’s concatenate the results into a vector with <code>c</code>.</p>
</div>
<div id="sequential-foreach-loop-with-.combine" class="section level2">
<h2><em>Sequential</em> <code>foreach()</code> loop with <code>.combine</code></h2>
<pre class="r"><code>x <- foreach(i=1:3,.combine='c') %do% i^2
x</code></pre>
<p>Tells <code>foreach()</code> to first calculate each iteration, then <code>.combine</code> them with a <code>c(...)</code></p>
</div>
<div id="sequential-foreach-loop-with-.combine-1" class="section level2">
<h2><em>Sequential</em> <code>foreach()</code> loop with <code>.combine</code></h2>
<pre class="r"><code>x <- foreach(i=1:3,.combine='rbind') %do% i^2
x</code></pre>
</div>
<div id="your-turn" class="section level2">
<h2>Your Turn</h2>
<p>Write a <code>foreach()</code> loop that:</p>
<ul>
<li>generates 10 sets of 100 random values from a normal distribution (<code>rnorm()</code>)</li>
<li>column-binds (<code>cbind</code>) them.</li>
</ul>
<p><br><br><br><br><br><br><br><br><br><br><br><br><br><br></p>
<pre class="r"><code>x <- foreach(i=1:10,.combine='cbind') %do% rnorm(100)
head(x)%>%kable()
dim(x)</code></pre>
</div>
<div id="parallel-foreach-loop" class="section level2">
<h2><em>Parallel</em> <code>foreach()</code> loop</h2>
<p>So far we’ve only used <code>%do%</code> which only uses a single processor.</p>
<p>Before running <code>foreach()</code> in parallel, you have to register a <em>parallel backend</em> with one of the <code>do</code> functions such as <code>doParallel()</code>. On most multicore systems, the easiest backend is typically <code>doParallel()</code>. On linux and mac, it uses <code>fork</code> system call and on Windows machines it uses <code>snow</code> backend. The nice thing is it chooses automatically for the system.</p>
<pre class="r"><code># register specified number of workers
registerDoParallel(3)
# or, reserve all all available cores
#registerDoParallel()
# check how many cores (workers) are registered
getDoParWorkers() </code></pre>
<blockquote>
<p><em>NOTE</em> It may be a good idea to use n-1 cores for processing (so you can still use your computer to do other things while the analysis is running)</p>
</blockquote>
<p>To run in parallel, simply change the <code>%do%</code> to <code>%dopar%</code>. Wasn’t that easy?</p>
<pre class="r"><code>## run the loop
x <- foreach(i=1:3, .combine='c') %dopar% i^2
x</code></pre>
</div>
<div id="a-slightly-more-complicated-example" class="section level2">
<h2>A slightly more complicated example</h2>
<p>In this section we will:</p>
<ol style="list-style-type: decimal">
<li>Generate data with known parameters</li>
<li>Fit a set of regression models using subsets of the complete dataset (e.g. bootstrapping)</li>
<li>Compare processing times for sequential vs. parallel execution</li>
</ol>
<p>For more on motivations to bootstrap, see <a href="http://www.sagepub.com/sites/default/files/upm-binaries/21122_Chapter_21.pdf">here</a>.</p>
<p>Make up some data:</p>
<pre class="r"><code>n <- 100000 # number of data points
x1 <- rnorm (n) # make up x1 covariate
b0 <- 1.8 # set intercept (beta0)
b1 <- -1.5 # set beta1
p = invlogit(b0+b1*x1)
y <- rbinom (n, 1, p) # simulate data with noise
data=cbind.data.frame(y=y,x1=x1,p=p)</code></pre>
<p>Let’s look at the data:</p>
<pre class="r"><code>kable(head(data),row.names = F,digits = 2)</code></pre>
<pre class="r"><code>ggplot(data,aes(y=x1,x=as.factor(y)))+
geom_boxplot()+
coord_flip()+
geom_line(aes(x=p+1,y=x1),col="red",size=2,alpha=.5)+
xlab("Binary Response")+
ylab("Covariate")</code></pre>
<div id="sampling-from-a-dataset-with-sample_n" class="section level3">
<h3>Sampling from a dataset with <code>sample_n()</code></h3>
<pre class="r"><code>size=5
sample_n(data,size,replace=T)</code></pre>
</div>
<div id="simple-generalized-linear-model" class="section level3">
<h3>Simple Generalized Linear Model</h3>
<p>This is the formal definition of the model we’ll use:</p>
<p><span class="math inline"><em>y</em><sub><em>i</em></sub> ∼ <em>B</em><em>e</em><em>r</em><em>n</em><em>o</em><em>u</em><em>l</em><em>l</em><em>i</em>(<em>p</em><sub><em>i</em></sub>)</span></p>
<p><span class="math inline"><em>l</em><em>o</em><em>g</em><em>i</em><em>t</em>(<em>p</em><sub><em>i</em></sub>)=<em>β</em><sub>0</sub> + <em>β</em><sub>1</sub><em>X</em><sub><em>i</em></sub></span></p>
<p>The index <span class="math inline"><em>i</em></span> identifies each grid cell (data point). <span class="math inline"><em>β</em><sub>0</sub></span> - <span class="math inline"><em>β</em><sub>1</sub></span> are model coefficients (intercept and slope), and <span class="math inline"><em>y</em><sub><em>i</em></sub></span> is the simulated observation from cell <span class="math inline"><em>i</em></span>.</p>
<pre class="r"><code>trials = 10000
tsize = 100</code></pre>
<pre class="r"><code> ptime <- system.time({
result <- foreach(i=1:trials,
.combine = rbind.data.frame) %dopar% {
tdata=sample_n(data,tsize,replace=TRUE)
M1=glm(y ~ x1, data=tdata, family=binomial(link="logit"))
## return parameter estimates
cbind.data.frame(trial=i,t(coefficients(M1)))
}
})
ptime</code></pre>
<p>Look at <code>results</code> object containing slope and aspect from subsampled models. There is one row per sample with columns for the estimated intercept and slope for that sample.</p>
<pre class="r"><code>kable(head(result),digits = 2)</code></pre>
<pre class="r"><code>ggplot(dplyr::select(result,everything(),Intercept=contains("Intercept")))+
geom_density(aes(x=Intercept),fill="black",alpha=.2)+
geom_vline(aes(xintercept=b0),size=2)+
geom_density(aes(x=x1),fill="red",alpha=.2)+
geom_vline(aes(xintercept=b1),col="red",size=2)+
xlim(c(-5,5))+
ylab("Parameter Value")+
xlab("Density")</code></pre>
<p>So we were able to perform 10^{4} separate model fits in seconds. Let’s see how long it would have taken in sequence.</p>
<pre class="r"><code> stime <- system.time({
result <- foreach(i=1:trials,
.combine = rbind.data.frame) %do% {
tdata=sample_n(data,tsize,replace=TRUE)
M1=glm(y ~ x1, data=tdata,family=binomial(link="logit"))
## return parameter estimates
cbind.data.frame(trial=i,t(coefficients(M1)))
}
})
stime</code></pre>
<p>So we were able to run 10^{4} separate model fits in seconds when using 3 CPUs and seconds on one CPU. That’s X faster for this simple example.</p>
</div>
</div>
<div id="your-turn-1" class="section level2">
<h2>Your Turn</h2>
<ul>
<li>Generate some random as follows:</li>
</ul>
<pre class="r"><code>n <- 10000 # number of data points
x1 <- rnorm (n) # make up x1 covariate
b0 <- 25 # set intercept (beta0)
b1 <- -15 # set beta1
y <- rnorm (n, b0+b1*x1,10) # simulate data with noise
data2=cbind.data.frame(y=y,x1=x1)</code></pre>
<p>Write a parallel <code>foreach()</code> loop that:</p>
<ul>
<li>selects a sample (as above) with <code>sample_n()</code></li>
<li>fits a ‘normal’ linear model with <code>lm()</code></li>
<li>Compiles the coefficient of determination (R^2) from each model</li>
</ul>
<p>Hint: see <code>?summary.lm</code>.</p>
<p><br><br><br><br><br><br><br><br><br><br><br><br><br></p>
<pre class="r"><code>trials = 100
tsize = 100
result <- foreach(i=1:trials,.combine = rbind.data.frame) %dopar% {
tdata=sample_n(data2,tsize,replace=TRUE)
M1=lm(y ~ x1, data=tdata)
cbind.data.frame(trial=i,
t(coefficients(M1)),
r2=summary(M1)$r.squared,
aic=AIC(M1))
}
ggplot(data2,aes(y=y,x=x1))+
geom_point(col="grey")+
geom_abline(data=dplyr::select(result,everything(),Intercept=contains("Intercept")),
aes(intercept=Intercept,slope=x1),alpha=.5)</code></pre>
<div id="writing-data-to-disk" class="section level3">
<h3>Writing data to disk</h3>
<p>For long-running processes, you may want to consider writing results to disk <em>as-you-go</em> rather than waiting until the end in case of a problem (power failure, single job failure, etc.).</p>
<pre class="r"><code>## assign target directory
td=tempdir()
foreach(i=1:trials,
.combine = rbind.data.frame) %dopar% {
tdata=sample_n(data,tsize,replace=TRUE)
M1=glm(y ~ x1, data=tdata, family=binomial(link="logit"))
## return parameter estimates
results=cbind.data.frame(trial=i,t(coefficients(M1)))
## write results to disk
file=paste0(td,"/results_",i,".csv")
write.csv(results,file=file)
return(NULL)
}</code></pre>
<p>That will save the result of each subprocess to disk (be careful about duplicated file names!):</p>
<pre class="r"><code>list.files(td,pattern="results")%>%head()</code></pre>
</div>
<div id="other-useful-foreach-parameters" class="section level3">
<h3>Other useful <code>foreach</code> parameters</h3>
<ul>
<li><code>.inorder</code> (true/false) results combined in the same order that they were submitted?</li>
<li><code>.errorhandling</code> (stop/remove/pass)</li>
<li><code>.packages</code> packages to made available to sub-processes</li>
<li><code>.export</code> variables to export to sub-processes</li>
</ul>
</div>
</div>
</div>
<div id="spatial-example" class="section level1">
<h1>Spatial example</h1>
<p>In this section we will:</p>
<ol style="list-style-type: decimal">
<li>Generate some <em>spatial</em> data</li>
<li>Tile the region to facilitate processing the data in parallel.</li>
<li>Perform a moving window mean for the full area</li>
<li>Compare processing times for sequential vs. parallel execution</li>
</ol>
<div id="generate-spatial-data" class="section level2">
<h2>Generate Spatial Data</h2>
<p>A function to generate <code>raster</code> object with spatial autocorrelation.</p>
<pre class="r"><code>simrast=function(nx=60,ny=60,theta=10,seed=1234){
## create a random raster with some spatial structure
## Theta is the scale of an exponential decay function.
## This controls degree of autocorrelation,
## values close to 1 are close to random while values near nx/4 have high autocorrelation
r=raster(nrows=ny, ncols=nx,vals=1,xmn=-nx/2, xmx=nx/2, ymn=-ny/2, ymx=ny/2)
names(r)="z"
# Simulate a Gaussian random field with an exponential covariance function
set.seed(seed) #set a seed so everyone's maps are the same
grid=list(x=seq(xmin(r),xmax(r)-1,by=res(r)[1]),y=seq(ymin(r),ymax(r)-1,res(r)[2]))
obj<-Exp.image.cov(grid=grid, theta=theta, setup=TRUE)
look<- sim.rf( obj)
values(r)=t(look)*10
return(r)
}</code></pre>
<p>Generate a raster using <code>simrast</code>.</p>
<pre class="r"><code>r=simrast(nx=3000,ny=1000,theta = 100)
r</code></pre>
<p>Plot the raster showing the grid.</p>
<pre class="r"><code>gplot(r)+
geom_raster(aes(fill = value))+
scale_fill_gradient(low = 'white', high = 'blue')+
coord_equal()+ylab("Y")+xlab("X")</code></pre>
</div>
<div id="tile-the-region" class="section level2">
<h2>“Tile” the region</h2>
<p>To parallelize spatial data, you often need to <em>tile</em> the data and process each tile separately. Here is a function that will take a bounding box, tile size and generate a tiling system. If given an <code>overlap</code> term, it will also add buffers to the tiles to reduce/eliminate edge effects, though this depends on what algorithm/model you are using.</p>
<pre class="r"><code>tilebuilder=function(raster,size=10,overlap=NULL){
## get raster extents
xmin=xmin(raster)
xmax=xmax(raster)
ymin=ymin(raster)
ymax=ymax(raster)
xmins=c(seq(xmin,xmax-size,by=size))
ymins=c(seq(ymin,ymax-size,by=size))
exts=expand.grid(xmin=xmins,ymin=ymins)
exts$ymax=exts$ymin+size
exts$xmax=exts$xmin+size
if(!is.null(overlap)){
#if overlapped tiles are requested, create new columns with buffered extents
exts$yminb=exts$ymin
exts$xminb=exts$xmin
exts$ymaxb=exts$ymax
exts$xmaxb=exts$xmax
t1=(exts$ymin-overlap)>=ymin
exts$yminb[t1]=exts$ymin[t1]-overlap
t2=exts$xmin-overlap>=xmin
exts$xminb[t2]=exts$xmin[t2]-overlap
t3=exts$ymax+overlap<=ymax
exts$ymaxb[t3]=exts$ymax[t3]+overlap
t4=exts$xmax+overlap<=xmax
exts$xmaxb[t4]=exts$xmax[t4]+overlap
}
exts$tile=1:nrow(exts)
return(exts)
}</code></pre>
<p>Generate a tiling system for that raster. Here will use only three tiles (feel free to play with this).</p>
<pre class="r"><code>jobs=tilebuilder(r,size=1000,overlap=80)
kable(jobs,row.names = F,digits = 2)</code></pre>
<p>Plot the raster showing the grid.</p>
<pre class="r"><code>ggplot(jobs)+
geom_raster(aes(x=coordinates(r)[,1],y=coordinates(r)[,2],fill = values(r)))+
scale_fill_gradient(low = 'white', high = 'blue')+
geom_rect(mapping=aes(xmin=xmin,xmax=xmax,ymin=ymin,ymax=ymax),
fill="transparent",lty="dashed",col="darkgreen")+
geom_rect(aes(xmin=xminb,xmax=xmaxb,ymin=yminb,ymax=ymaxb),
fill="transparent",col="black")+
geom_text(aes(x=(xminb+xmax)/2,y=(yminb+ymax)/2,label=tile),size=10)+
coord_equal()+ylab("Y")+xlab("X")</code></pre>
</div>
<div id="run-a-simple-spatial-analysis-focal-moving-window" class="section level2">
<h2>Run a simple spatial analysis: <code>focal</code> moving window</h2>
<p>Use the <code>focal</code> function from the raster package to calculate a 3x3 moving window mean over the raster.</p>
<pre class="r"><code>stime2=system.time({
r_focal1=focal(r,w=matrix(1,101,101),mean,pad=T)
})
stime2
## plot it
gplot(r_focal1)+
geom_raster(aes(fill = value))+
scale_fill_gradient(low = 'white', high = 'blue')+
coord_equal()+ylab("Y")+xlab("X")</code></pre>
<p>That works great (and pretty fast) for this little example, but as the data (or the size of the window) get larger, it can become prohibitive.</p>
</div>
<div id="repeat-the-analysis-but-parallelize-using-the-tile-system." class="section level2">
<h2>Repeat the analysis, but parallelize using the tile system.</h2>
<p>First write a function that breaks up the original raster, computes the focal mean, then puts it back together. You could also put this directly in the <code>foreach()</code> loop.</p>
<pre class="r"><code>focal_par=function(i,raster,jobs,w=matrix(1,101,101)){
## identify which row in jobs to process
t_ext=jobs[i,]
## crop original raster to (buffered) tile
r2=crop(raster,extent(t_ext$xminb,t_ext$xmaxb,t_ext$yminb,t_ext$ymaxb))
## run moving window mean over tile
rf=focal(r2,w=w,mean,pad=T)
## crop to tile
rf2=crop(rf,extent(t_ext$xmin,t_ext$xmax,t_ext$ymin,t_ext$ymax))
## return the object - could also write the file to disk and aggregate later outside of foreach()
return(rf2)
}</code></pre>
<p>Run the parallelized version.</p>
<pre class="r"><code>registerDoParallel(3)
ptime2=system.time({
r_focal=foreach(i=1:nrow(jobs),.combine=merge,.packages=c("raster")) %dopar% focal_par(i,r,jobs)
})</code></pre>
<p>Are the outputs the same?</p>
<pre class="r"><code>identical(r_focal,r_focal1)</code></pre>
<p>So we were able to process the data in seconds when using 3 CPUs and seconds on one CPU. That’s X faster for this simple example.</p>
</div>
<div id="parallelized-raster-functions" class="section level2">
<h2>Parallelized Raster functions</h2>
<p>Some functions in raster package also easy to parallelize.</p>
<pre class="r"><code>ncores=2
beginCluster(ncores)
fn=function(x) x^3
system.time(fn(r))
system.time(clusterR(r, fn, verbose=T))
endCluster()</code></pre>
<p>Does <em>not</em> work with:</p>
<ul>
<li>merge</li>
<li>crop</li>
<li>mosaic</li>
<li>(dis)aggregate</li>
<li>resample</li>
<li>projectRaster</li>
<li>focal</li>
<li>distance</li>
<li>buffer</li>
<li>direction</li>
</ul>
</div>
</div>
<div id="high-performance-computers-hpc" class="section level1">
<h1>High Performance Computers (HPC)</h1>
<p><em>aka</em> <em>supercomputers</em>, for example, check out the <a href="https://www.buffalo.edu/ccr.html">University at Buffalo HPC</a></p>
<div class="figure">
<img src="assets/CCR.png" />
</div>
<p>Working on a cluster can be quite different from a laptop/workstation. The most important difference is the existence of <em>scheduler</em> that manages large numbers of individual tasks.</p>
<div id="slurm-and-r" class="section level2">
<h2>SLURM and R</h2>
<p>You typically don’t run the script <em>interactively</em>, so you need to edit your script to ‘behave’ like a normal <code>#!</code> (linux command line) script. This is easy with <a href="http://cran.r-project.org/web/packages/getopt/index.html">getopt</a> package.</p>
<pre class="r"><code>cat(paste("
library(getopt)
## get options
opta <- getopt(
matrix(c(
'date', 'd', 1, 'character'
), ncol=4, byrow=TRUE))
## extract value
date=as.Date(opta$date)
## Now your script using date as an input
print(date+1)
q(\"no\")
"
),file=paste("script.R",sep=""))</code></pre>
<p>Then you can run this script from the command line like this:</p>
<pre class="r"><code>Rscript script.R --date 2013-11-05</code></pre>
<p>You will need the complete path if <code>Rscript</code> is not on your system path. For example, on OS X, it might be at <code>/Library/Frameworks/R.framework/Versions/3.2/Resources/Rscript</code>.</p>
<p>Or even from within R like this:</p>
<pre class="r"><code>system("Rscript script.R --date 2013-11-05",intern=T)</code></pre>
<div id="driving-cluster-from-r" class="section level3">
<h3>Driving cluster from R</h3>
<p>Possible to drive the cluster from within R via SLURM. First, define the jobs and write that file to disk:</p>
<pre class="r"><code>script="script.R"
dates=seq(as.Date("2000-01-01"),as.Date("2000-12-31"),by=60)
pjobs=data.frame(jobs=paste(script,"--date",dates))
write.table(pjobs,
file="process.txt",
row.names=F,col.names=F,quote=F)</code></pre>
<p>This table has one row per task:</p>
<pre class="r"><code>pjobs</code></pre>
<p>Now identify other parameters for SLURM.</p>
<pre class="r"><code>### Set up submission script
nodes=2
walltime=5</code></pre>
</div>
<div id="write-the-slurm-script" class="section level3">
<h3>Write the SLURM script</h3>
<p><a href="https://www.buffalo.edu/ccr.html">More information here</a></p>
<pre class="r"><code>### write SLURM script to disk from R
cat(paste("#!/bin/sh
#SBATCH --partition=general-compute
#SBATCH --time=00:",walltime,":00
#SBATCH --nodes=",nodes,"
#SBATCH --ntasks-per-node=8
#SBATCH --constraint=IB
#SBATCH --mem=300
# Memory per node specification is in MB. It is optional.
# The default limit is 3000MB per core.
#SBATCH --job-name=\"date_test\"
#SBATCH --output=date_test-srun.out
#SBATCH --mail-user=adamw@buffalo.edu
#SBATCH --mail-type=ALL
##SBATCH --requeue
#Specifies that the job will be requeued after a node failure.
#The default is that the job will not be requeued.
## Load necessary modules
module load openmpi/gcc-4.8.3/1.8.4
module load R
IDIR=~
WORKLIST=$IDIR/process.txt
EXE=Rscript
LOGSTDOUT=$IDIR/log/stdout
LOGSTDERR=$IDIR/log/stderr
### use mpiexec to parallelize across lines in process.txt
mpiexec -np $CORES xargs -a $WORKLIST -p $EXE 1> $LOGSTDOUT 2> $LOGSTDERR
",sep=""),file=paste("slurm_script.txt",sep=""))</code></pre>
<p>Now we have a list of jobs and a qsub script that points at those jobs with the necessary PBS settings.</p>
<pre class="r"><code>## run it!
system("sbatch slurm_script.txt")
## Check status with squeue
system("squeue -u adamw")</code></pre>
<p>For more detailed information about the UB HPC, <a href="https://www.buffalo.edu/ccr/support/UserGuide.html">see the CCR Userguide</a>.</p>
</div>
</div>
</div>
<div id="summary" class="section level1">
<h1>Summary</h1>
<blockquote>
<p>Each task should involve computationally-intensive work. If the tasks are very small, it can take <em>longer</em> to run in parallel.</p>
</blockquote>
<div id="choose-your-method" class="section level2">
<h2>Choose your method</h2>
<ol style="list-style-type: decimal">
<li>Run from master process (e.g. <code>foreach</code>)
<ul>
<li>easier to implement and collect results</li>
<li>fragile (one failure can kill it and lose results)</li>
<li>clumsy for <em>big</em> jobs</li>
</ul></li>
<li>Run as separate R processes
<ul>
<li>see <a href="http://cran.r-project.org/web/packages/getopt/index.html"><code>getopt</code></a> library</li>
<li>safer for big jobs: each job completely independent</li>
<li>update job list to re-run incomplete submissions</li>
<li>compatible with slurm / cluster computing</li>
<li>forces you to have a clean processing script</li>
</ul></li>
</ol>
</div>
<div id="further-reading" class="section level2">
<h2>Further Reading</h2>
<ul>
<li><a href="http://cran.r-project.org/web/views/HighPerformanceComputing.html">CRAN Task View: High-Performance and Parallel Computing with R</a></li>
<li><a href="www.stat.uiowa.edu/~luke/talks/uiowa03.pdf">Simple Parallel Statistical Computing in R</a></li>
<li><a href="http://download.springer.com/static/pdf/832/chp%253A10.1007%252F978-3-642-13872-0_64.pdf?auth66=1415215123_43bf0cbf5ae8f5143b7ee309ff5e3556&ext=.pdf">Parallel Computing with the R Language in a Supercomputing Environment</a></li>
</ul>
</div>
</div>
<!-- Google Analytics -->
<script>
(function(i,s,o,g,r,a,m){i['GoogleAnalyticsObject']=r;i[r]=i[r]||function(){
(i[r].q=i[r].q||[]).push(arguments)},i[r].l=1*new Date();a=s.createElement(o),
m=s.getElementsByTagName(o)[0];a.async=1;a.src=g;m.parentNode.insertBefore(a,m)
})(window,document,'script','https://www.google-analytics.com/analytics.js','ga');
ga('create', 'UA-2684666-2', 'auto');
ga('send', 'pageview');
</script>
<!-- disqus -->
<div id="disqus_thread" style="margin-top: 45px;"></div>
<script type="text/javascript">
/* * * CONFIGURATION VARIABLES: EDIT BEFORE PASTING INTO YOUR WEBPAGE * * */
var disqus_shortname = 'rdataanalysis'; // required: replace example with your forum shortname
var disqus_identifier = "{{ page.url }}";
/* * * DON'T EDIT BELOW THIS LINE * * */
(function() {
var dsq = document.createElement('script'); dsq.type = 'text/javascript'; dsq.async = true;
dsq.src = '//' + disqus_shortname + '.disqus.com/embed.js';
(document.getElementsByTagName('head')[0] || document.getElementsByTagName('body')[0]).appendChild(dsq );
})();
</script>
<noscript>Please enable JavaScript to view the <a href="http://disqus.com/?ref_noscript">
comments powered by Disqus.</a>
</noscript>
<!-- give the footer some space -->
<br/>
<br/>
<footer id="site-footer">
<div id="footer1">
<a href="http://www.adamwilson.us"><img src="img/wilson.png" alt="logo" width=40px></a>
<a href="http://adamwilson.us/#contact"><i class="fa fa-envelope fa-2x"></i></a>
<a href="https://twitter.com/AdamWilsonLab"><i class="fa fa-twitter fa-2x"></i></a>
<a href="https://github.com/AdamMWilson"><i class="fa fa-github fa-2x"></i></a>
</div>
<div id="footer2">
<a rel="license" property="http://creativecommons.org/ns#license"
href="http://creativecommons.org/licenses/by/4.0/" ><img src="img/cc-by.svg" alt="cc-by"/></a>
</div>
</footer>
</div>
<script>
// add bootstrap table styles to pandoc tables
$(document).ready(function () {
$('tr.header').parent('thead').parent('table').addClass('table table-condensed');
});
</script>
</body>
</html>