http://cos.name/en/ In the name of statistics, unite! en Tue, 07 Sep 2010 07:58:16 +0000 http://cos.name/en/topic/12#post-66 Thu, 19 Aug 2010 19:44:33 +0000 digitalcat 66@http://cos.name/en/ <p>Good tip. I had a similar solution once but instead of tinkering with the color gradient I just log transformed my data. </p> http://cos.name/en/topic/17#post-65 Fri, 23 Jul 2010 21:02:47 +0000 yihui 65@http://cos.name/en/ <p>I guess it might be the package <code>reshape</code> by Hadley... </p> http://cos.name/en/topic/17#post-64 Fri, 23 Jul 2010 17:15:00 +0000 EzGraphs 64@http://cos.name/en/ <p>I recently have been using the sqldf package to reshape data frames.</p> <p><a href="http://www.r-chart.com/2010/07/make-r-speak-sql-with-sqldf.html" rel="nofollow">http://www.r-chart.com/2010/07/make-r-speak-sql-with-sqldf.html</a></p> <p>What is the best method that you know of for reshaping data frames? </p> http://cos.name/en/topic/16#post-63 Wed, 21 Jul 2010 22:41:48 +0000 pink_lover 63@http://cos.name/en/ <p>i need help on LDA (linear discriminant analysis). I want to decrease a high dimension data to one dimension using LDA. I find a code on line, but it need class label. If I know nothing about the data, and I do not know what is the label, how could I decrease the dimension?<br /> Thanks </p> http://cos.name/en/topic/2#post-62 Tue, 20 Apr 2010 15:20:21 +0000 kirkland 62@http://cos.name/en/ <p>It's definitely easy. Find a reference book about SPSS and follow the instruction in it, you are sure to solve all your problems, though you might understand nothing about the underlying theories. Anyway, as long as you know how to interpret the results, everything will be fine. </p> http://cos.name/en/topic/1#post-61 Tue, 20 Apr 2010 15:14:00 +0000 kirkland 61@http://cos.name/en/ <p>This website is awesome! </p> http://cos.name/en/topic/14#post-59 Fri, 05 Feb 2010 09:53:14 +0000 ryusukekenji 59@http://cos.name/en/ <p><a href="http://cos.name/bbs/read.php?tid=17606" rel="nofollow">http://cos.name/bbs/read.php?tid=17606</a></p> <p>I tried to browse over this website, while you may learn R and apply it on R-plus since its workable.</p> <p>R-plus gotta addon menu similar with S-plus, it will be more familiar for beginner who are not familiar with coding. </p> http://cos.name/en/topic/14#post-58 Tue, 02 Feb 2010 13:21:33 +0000 yihui 58@http://cos.name/en/ <p>R+ is the commercial product of the company XLSolutions, so you'd better ask them instead.</p> <p>R is free and you can also get free help here. </p> http://cos.name/en/topic/14#post-57 Mon, 01 Feb 2010 06:09:12 +0000 peng 57@http://cos.name/en/ <p>hi friends,<br /> I am new to R.I would like to know R-PLUS.Does any know where can I get the free training for R-PLUS.</p> <p>Regards,<br /> Peng. </p> http://cos.name/en/topic/1#post-56 Thu, 21 Jan 2010 11:06:09 +0000 Peter 56@http://cos.name/en/ <p>the forum is like a series of data instead of a simple chart,so... it is a little difficult for me to have a brief idea without the learning from COS in Chinese. </p> http://cos.name/en/topic/13#post-55 Sat, 16 Jan 2010 13:01:38 +0000 yihui 55@http://cos.name/en/ <p>just use '<code>y ~ .</code>' as your formula; the dot '<code>.</code>' means all the rest of variables in the data frame given in the <code>data =</code> parameter. </p> http://cos.name/en/topic/13#post-54 Tue, 12 Jan 2010 09:13:51 +0000 npobedina 54@http://cos.name/en/ <p>Hi!<br /> I'm trying to run logistic regression on a dataset which is contained in dataframe "data" ("y" is in the first col, and 28 parameters for the model).<br /> How can I write formula for function <code>glm</code> without listing explicitly all 28 paramaters?<br /> <code>glm(data[,1]~data[,2]+data[,3]+data[,4]+...,family=binomial)</code></p> <p>As an option I can use <code>glm.fit(data[,-1],data[,1],family = binomial(link=logit))</code>. But the obtained object cannot be used in function <code>predict.glm</code>.</p> <p>Thanks,<br /> Natalia </p> http://cos.name/en/topic/3#post-53 Tue, 12 Jan 2010 03:38:29 +0000 lovebluesky 53@http://cos.name/en/ <p>it's cool, i hope that i could reach your level some day </p> http://cos.name/en/topic/1#post-52 Fri, 18 Dec 2009 01:22:05 +0000 yihui 52@http://cos.name/en/ <p>Thanks, &amp; please bring more people here :-D </p> http://cos.name/en/topic/1#post-51 Fri, 18 Dec 2009 00:42:27 +0000 guan 51@http://cos.name/en/ <p>The English BBS should have some popularized...</p> <p>Today, I am here for support. </p> http://cos.name/en/topic/12#post-50 Thu, 03 Dec 2009 17:31:49 +0000 yihui 50@http://cos.name/en/ <p>Sometimes we need to fill in our graphical elements with gradient colors according to a certain variable (say, <code>z</code>). A special case is, <code>z</code> consists of a majority of small values and a few large ones. If we use the linear gradient (default in <code>scale_fill_gradient()</code> or <code>scale_colour_gradient()</code>, i.e. <code>trans = &quot;identity&quot;</code>), we will not be able to tell the differences among colors for points with many small <code>z</code> values. E.g.</p> <pre>## make up some data x = c(rnorm(100), 5:10) ## z contains 6 large values z = c(runif(100, 0, 1.5), runif(6, 10, 15)) ## gradient from red to green qplot(x, z, colour = z) + scale_colour_gradient(low = "red", high = "green") ## cannot tell the difference even if using scale_colour_gradient2() qplot(x, z, colour = z) + scale_colour_gradient2(low = "red", mid = "yellow", high = "green", midpoint = 1.5)</pre> <p>However, we can log-transform <code>z</code> to enlarge the difference among small values and make extremely large values much smaller. In this case, we can observe the gradient colors now:</p> <pre>qplot(x, z, colour = z) + scale_colour_gradient2(low = "red", mid = "yellow", high = "green", midpoint = log(1.5), trans = "log") ## or equivalently qplot(x, z, colour = log(z)) + scale_colour_gradient2(low = "red", mid = "yellow", high = "green", midpoint = log(1.5)) </pre> <p>Essentially we are using a nonlinear color interpolation (linear to <code>log(z)</code>, nonlinear to <code>z</code>). </p> http://cos.name/en/topic/11#post-49 Tue, 17 Nov 2009 18:19:27 +0000 yihui 49@http://cos.name/en/ <p>The raw data is like this ( in a 'wide' format, see <a href="http://www.public.iastate.edu/~pdixon/stat500/data/ratweight.txt" rel="nofollow">http://www.public.iastate.edu/~pdixon/stat500/data/ratweight.txt</a> ):</p> <pre>high beef 73 102 118 104 81 107 100 87 117 111 high cereal 98 74 56 111 95 88 82 77 86 92 high pork 94 79 96 98 102 102 108 91 120 105 low beef 90 76 90 64 86 51 72 90 95 78 low cereal 107 95 97 80 98 74 74 67 89 58 low pork 49 82 73 86 81 97 106 70 61 82 </pre> <p>And we want to reshape it into this form:</p> <pre>amount type gain high beef 73 high beef 102 high beef 118 high beef 104 high beef 81 high beef 107 high beef 100 high beef 87 high beef 117 high beef 111 high cereal 98 high cereal 74 ....</pre> <p>My solution is (<code>reshape()</code> is a function in base R):</p> <pre><em>## the code is nearly plain English; see comments below if explanation needed</em> <strong>rat = read.table("http://pdixon.public.iastate.edu/stat500/data/ratweight.txt", col.names = c("amount", "type", rep("gain", 10))) rat2 = reshape(rat, varying = list(3:12), idvar = 1:2, direction = "long")</strong> <em>## we know the column names are 'amount', 'type' and 'gain', so specify 'col.names' ## in read.table(); 'amount' and 'type' will be converted to factors by default ## because they are characters (which helps in aov() and lm()). we know the first ## 2 columns are id variables (combinations of treatment levels) and 3:12 columns ## are measurements; we want to reshape the 'wide' data into 'long' format. ## if we want to order the data according to 'amount' and 'type', just order() it; ## the order does not matter in data analysis after all. ## want to drop the 'time' variable? drop it with a negative index</em> # rat2 = with(rat2, rat2[order(amount, type), -3]) <em>## annoying row names? set them to NULL; again, row names does not matter</em> # rownames(rat2) = NULL <em>## we may begin ANOVA now</em> # aov(gain ~ amount * type, rat2) </pre> <p>I guess Stat579 students have other neat approaches to reshape this dataset. So I'm looking forward to solutions from Hadley's <code>reshape</code> package :)</p> <p>BTW, in case of unequal numbers per treatment, we may specify <code>fill = TRUE</code> in <code>read.table()</code> so that "blank cells" will be filled with <code>NA</code>'s. In the end we can omit the missing cells with <code>na.omit()</code>. </p> http://cos.name/en/topic/2#post-48 Wed, 04 Nov 2009 04:51:12 +0000 gradian 48@http://cos.name/en/ <p>it's possible :)<br /> he/she needs to learn only application of tests not theorical basis :D </p> http://cos.name/en/topic/9#post-47 Thu, 22 Oct 2009 18:59:16 +0000 yihui 47@http://cos.name/en/ <p>I'm really excited to see the first guest coming!!</p> <p>I started a new topic here: <a href="http://cos.name/en/topic/re-how-to-compute-weighted-cross-tabulation-to-sazimme2" rel="nofollow">http://cos.name/en/topic/re-how-to-compute-weighted-cross-tabulation-to-sazimme2</a> </p> http://cos.name/en/topic/10#post-46 Thu, 22 Oct 2009 18:57:12 +0000 yihui 46@http://cos.name/en/ <p>To sazimme2: I'm not sure I've understood your question well; here is an example for weighted cross tabulation.</p> <pre># to be reproducible, set the random seed first # so you'll get the same random numbers <strong>set.seed(123)</strong> # generate a random data frame for demo purpose <strong>dat = data.frame( sex = factor(sample(c("Male", "Female"), 50, replace = TRUE)), answer = factor(sample(c("Yes", "No"), 50, replace = TRUE)), weight = sample(1:10, 50, replace = TRUE) )</strong> # <code>dat</code> looks like this <strong>head(dat)</strong> <em> sex answer weight 1 Male Yes 6 2 Female Yes 4 3 Male No 5 4 Female Yes 10 5 Female No 5 6 Male Yes 9 </em> # ordinary table without weights # i don't want to write dat$something, so use with() <strong>with(dat, table(sex, answer))</strong> <em> answer sex No Yes Female 12 13 Male 10 15 </em> # ordinary table is equavalent to vector *length* of each cross-group # or, weight = 1 for each case <strong>with(dat, aggregate(weight, list(sex = sex, answer = answer), length))</strong> <em> sex answer x 1 Female No 12 2 Male No 10 3 Female Yes 13 4 Male Yes 15 </em> # table with weights: just sum up weights for each cell <strong>(wtable = with(dat, aggregate(weight, list(sex = sex, answer = answer), sum)))</strong> <em> sex answer x 1 Female No 77 2 Male No 63 3 Female Yes 75 4 Male Yes 69 </em> # need 2x2 matrix form? <strong>matrix(wtable$x, nrow = 2, ncol = 2, dimnames = list(levels(dat$sex), levels(dat$answer)))</strong> <em> No Yes Female 77 75 Male 63 69 </em> #~~~~~~~~~~~~~~~~~~~~~~~~~reshape~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~# # well, i'm behind the times now; you guys are learning more # advanced tools in 579 - let's do it with cast() in reshape # rather than the traditional aggregate() <strong>library(reshape) cast(dat, sex ~ answer, sum, value = "weight")</strong> <em> sex No Yes 1 Female 77 75 2 Male 63 69 </em> </pre> <p>Sorry the code looks wordy. I just want to make the process clear and solutions diversified. Take a quick look at the reshape part if you want to reach the solution directly :) </p> http://cos.name/en/topic/9#post-45 Thu, 22 Oct 2009 10:00:32 +0000 sazimme2 45@http://cos.name/en/ <p>I am trying to do a cross tabulation of two variables. Sex and the answer to a survey question. Both variables are factors. This is easy to do using the table function. The problem is that these respondents have weights. Table does not have an option to put in weights. Any ideas? Thanks! I'm in ST 579 so relatively new to R so explanations where you explain why I should do what I should would be appreciated. </p> http://cos.name/en/topic/9#post-44 Sun, 11 Oct 2009 22:37:34 +0000 yihui 44@http://cos.name/en/ <p>Yes I also noticed this phenomenon long ago. Is that something called "double-buffering" which R does not support? </p> http://cos.name/en/topic/9#post-43 Sun, 11 Oct 2009 19:53:46 +0000 yanlinlin82 43@http://cos.name/en/ <p>The effect of the code running in Windows is much better than in Linux. I tried a lot but still can not find any way to avoid the flicker when drawing repeatedly in Linux. </p> http://cos.name/en/topic/9#post-42 Sat, 10 Oct 2009 19:56:56 +0000 yihui 42@http://cos.name/en/ <p>Thanks!</p> <p>Another silly demo to show how <code>xlim</code> works:</p> <pre>par(mar = rep(0, 4)) for (i in 1:200) { plot(1, xlim = c(0, 2), ann = F, type = "n", axes = F, xaxs = "i") text(2 - i/200 * 3, 1, "happy_rabbit -___-//", col = topo.colors(200)[i], cex = 3) Sys.sleep(0.02) } </pre> http://cos.name/en/topic/9#post-41 Sat, 10 Oct 2009 16:17:48 +0000 happy_rabbit 41@http://cos.name/en/ <p>up~up~ </p> http://cos.name/en/topic/9#post-40 Wed, 07 Oct 2009 21:59:31 +0000 yihui 40@http://cos.name/en/ <p>We will try our best to offer additional help for learning R in this forum. Questions and discussions are warmly welcome!</p> <pre>par(mar = rep(0, 4)) for (i in 1:360) { plot(1, ann = F, type = "n", axes = F) text(1, 1, "Welcome", srt = i, col = rainbow(360)[i], cex = 7 * i/360) Sys.sleep(0.01) }</pre> http://cos.name/en/topic/8#post-39 Fri, 28 Aug 2009 10:27:37 +0000 yihui 39@http://cos.name/en/ <p><code>install.packages(&#39;labdsv&#39;)</code> if it's on CRAN; otherwise you have to download the <em>Windows Binary</em> package (*.zip) and use the menu in RGui to install it. </p> http://cos.name/en/topic/8#post-38 Fri, 28 Aug 2009 03:09:17 +0000 lqfcb 38@http://cos.name/en/ <p>I have downloaded this package in d:/my documents, but I cannot install it to R, although I see the help. Please write the command for me. Thanks a lot! </p> http://cos.name/en/topic/3#post-36 Mon, 22 Jun 2009 20:56:44 +0000 yanlinlin82 36@http://cos.name/en/ <p>Yes, with the very flexible interface, R can do almost everything in computer if you want. However, some of the things doing by R are not the best solutions. We were just trying to show off the tricks, which may attract people into R community.</p> <p>Everything in western sciences is learnable. R is not difficult to a medical student, since there are so many applications of statistical analysis in the field, and I think you may have already learned some of the methods in your course. That's enough to enter the R world. </p> http://cos.name/en/topic/3#post-35 Sun, 21 Jun 2009 19:03:47 +0000 achates 35@http://cos.name/en/ <p>Your guys are amazing! It seems R could do everything~~really? I can't help thinking it . Is it difficult to understand and apply to for a medical student? You know, in clinic there are many procedures like disease recurrence and wound recovery, and there are many influence and questions , need to solve. haha, was I too greedy? I am just a medical man knowing a little about math and computer. </p> http://cos.name/en/topic/7#post-34 Sat, 20 Jun 2009 07:55:23 +0000 yihui 34@http://cos.name/en/ <p>I realized the argument <code>timeout</code> is unnecessary, so this will be enough:</p> <pre>shutdown &lt;- function(wait = 0) { Sys.sleep(wait) ifelse(.Platform$OS.type == "windows", shell("shutdown -s -t 0"), system("shutdown -h now")) }</pre> <p>This function has gone to <code>fun</code> now. You can see it in R-Forge. </p> http://cos.name/en/topic/7#post-33 Sat, 20 Jun 2009 06:05:01 +0000 Taiyun 33@http://cos.name/en/ <p>Cool, Wonderful job! </p> http://cos.name/en/topic/7#post-32 Sat, 20 Jun 2009 04:21:29 +0000 yanlinlin82 32@http://cos.name/en/ <p>Linux use the 'shutdown' command too, but with different parameters.</p> <p>We can use following command to shutdown immediately:<br /> <code>$ shutdown -h now</code><br /> Or following command to shutdown after 30 minutes:<br /> <code>$ shutdown -h +30</code><br /> Or shutdown at 23:30:<br /> <code>$ shutdown -h 23:30</code></p> <p>So the function can be:<br /> <pre><code>shutdown = function(wait = 0, timeout = 10) { Sys.sleep(wait) shell( paste(sep = &quot;&quot;, &quot;shutdown &quot;, switch(.Platform$OS.type, windows = sprintf(&quot;-s -t %d&quot;, timeout), unix = &quot;-h now&quot;) ) ) } shutdown()</code></pre> http://cos.name/en/topic/7#post-31 Sat, 20 Jun 2009 00:54:03 +0000 yihui 31@http://cos.name/en/ <p>Because we can shut down the Windows system by the command <code>shutdown</code>, we may also use the R function <code>shell()</code> to execute the command. For instance, <code>shell(&quot;shutdown -s -t 10&quot;)</code> will shut down your computer after 10 seconds (DON'T run the code before you have saved and closed all the programs - I'll not be responsible for your data lost!). If you regret doing so, you can execute <code>shell(&quot;shutdown -a&quot;)</code> immediately.</p> <p>I'm not sure whether the command is the same under Linux. (AFAIR, it's the same command but with different options?)</p> <p>We may include this function in the package <code>fun</code> as well:</p> <pre>shutdown = function(wait = 0, timeout = 10) { Sys.sleep(wait) # .Platform$OS.type? Windows? Linux? shell(sprintf("shutdown -s -t %d", timeout)) } </pre> <p>I think this function can be useful when we want our computer to be shut down automatically after a (time-consuming) task. </p> http://cos.name/en/topic/6#post-30 Thu, 18 Jun 2009 21:05:10 +0000 achates 30@http://cos.name/en/ <p>I do some research works using Logistic Regression by SPSS 17. </p> <p>10 predictors (5 numeric and 5 nominal )and 1 dependent(3 groups). 128 cases.</p> <p>After calculation, output remind me like that:</p> <blockquote><p> There are 256(66.7%)cells(dependent variables by subpopulations) with zero frequencies.Unexpected singularities in Hessian matrix are encountered. This indicates that either some predictor variables should be excluded or some categories should be merged.The NOMREG procedure continues despite the above warnings. Subsequent results shown are based on the last iteration. Validity of the model fit is uncertain.</p> </blockquote> <p>Does it mean that something happened cause some empty cells? Yes, there are indeed some empty cells . Do I need delete those cases?</p> <p>Thanks! </p>