{ "id": 1274, "title": "Hard first day with R -", "url": "/competitions/WhatDoYouKnow/discussion/1274", "author_name": "", "post_date": "2012-01-15T01:22:03.510Z", "votes": null, "comment_count": 4, "views": 2924, "content": "

Dear Listeners,

\r\n

I had a hard first day with 'R'! (half day with importing the data and another half day with the following beginner problem:)

\r\n

Backround: The last ten year I worked with SAS (work) and the change to R is not so easy. Used to data steps I was not able to do some really easy computing and merging.

\r\n

 What I was trying to do:

\r\n

Computing the mean question difficulty for each question_id with 'questiondiff<-tapply(training$correct,training$question_id,mean)'

\r\n

So far so good,  the result is a vector with the right numbers, but I am not able to get the second dimension for merging it (1 to many) with the original training data by question_id for further steps. Do I have to tell tapply that I still need the question_id?\r\n Or maybe I am just to stupid for merging in R and there is no dimension problem at all...and so all the Ideas in my head for the real job have to wait....  

\r\n

I hope somebody can give me a hint. Thanks for your time. For me it's bedtime. Good night.

\r\n

Greetings from Germany-

\r\n

Vielauge 

\r\n

 

\r\n

 

\r\n

 

\r\n

 

\r\n

 

\r\n

 

", "messages": [ { "id": "8105", "postDate": "01/15/2012 01:22:03", "content": "

Dear Listeners,

\r\n

I had a hard first day with 'R'! (half day with importing the data and another half day with the following beginner problem:)

\r\n

Backround: The last ten year I worked with SAS (work) and the change to R is not so easy. Used to data steps I was not able to do some really easy computing and merging.

\r\n

 What I was trying to do:

\r\n

Computing the mean question difficulty for each question_id with 'questiondiff<-tapply(training$correct,training$question_id,mean)'

\r\n

So far so good,  the result is a vector with the right numbers, but I am not able to get the second dimension for merging it (1 to many) with the original training data by question_id for further steps. Do I have to tell tapply that I still need the question_id?\r\n Or maybe I am just to stupid for merging in R and there is no dimension problem at all...and so all the Ideas in my head for the real job have to wait....  

\r\n

I hope somebody can give me a hint. Thanks for your time. For me it's bedtime. Good night.

\r\n

Greetings from Germany-

\r\n

Vielauge 

\r\n

 

\r\n

 

\r\n

 

\r\n

 

\r\n

 

\r\n

 

", "rawMarkdown": "", "votes": null }, { "id": "8109", "postDate": "01/15/2012 17:44:06", "content": "

Usually I would use ddply from the plyr package as a convenience wrapper to tapply that has some added features.

\r\n

However, it can be pretty slow, so I believe the use of tapply is justified.

\r\n

Having said that, for your specific problem, tapply implicitly transforms "training$questionid" into a factor and then computes the mean for each of the resulting levels. You should be able to recreate those levels in integer form with just a "unique(training$questionid)".

\r\n

You can then make a matrix with two columns via cbind, or a full data frame via a data.frame call.

\r\n

Best of luck!

", "rawMarkdown": "", "votes": null }, { "id": "8111", "postDate": "01/15/2012 22:36:38", "content": "

Keep it going, mate! Learning R was my new years resolution, and I am totally hooked to it now.. 

", "rawMarkdown": "", "votes": null }, { "id": "8112", "postDate": "01/15/2012 23:02:33", "content": "

Hi Vielauge,
\r\n
\r\nI'm not sure I understood your question, but have you tried this:
\r\nas.integer(dimnames(questiondiff)[[1]])
\r\n
\r\nOr, in a data frame:
\r\ndata.frame(question_id=as.integer(dimnames(questiondiff)[[1]]), questiondiff)
\r\n
\r\nJames

", "rawMarkdown": "", "votes": null }, { "id": "8123", "postDate": "01/16/2012 19:07:25", "content": "

Hi everybody,

\r\n

many thanks for your responses. 

\r\n

@Shea Parks

\r\n

Thank you. Your answer works and is definitely better than googling and searching in a lot of help documents for hours (with no result).

\r\n

@James Petterson

\r\n

I will try your answer too, but have not done it so far. I am sure I can learn something. (At first sight it looks quite similar to Shea Parks idea).

\r\n

@shaz24

\r\n

I will see. I am already addicted to kaggle, but R could be a problem. With two small childs (2 and 5)  and regular work there is not too much time to get into R. I guess 44 days will not be enough for me to get to top twenty with such elemental programming\r\n difficulties. But maybe I can get the benchmark with a total different method. We will see.

\r\n

Have a good time. All of you. Bye.

\r\n

Vielauge

\r\n

 

", "rawMarkdown": "", "votes": null } ], "comments": [ { "id": 8109, "author_name": "biasvariance", "author_url": "", "post_date": "01/15/2012 17:44:06", "content": "

Usually I would use ddply from the plyr package as a convenience wrapper to tapply that has some added features.

\r\n

However, it can be pretty slow, so I believe the use of tapply is justified.

\r\n

Having said that, for your specific problem, tapply implicitly transforms "training$questionid" into a factor and then computes the mean for each of the resulting levels. You should be able to recreate those levels in integer form with just a "unique(training$questionid)".

\r\n

You can then make a matrix with two columns via cbind, or a full data frame via a data.frame call.

\r\n

Best of luck!

", "votes": null, "replies": [] }, { "id": 8111, "author_name": "shaz24", "author_url": "", "post_date": "01/15/2012 22:36:38", "content": "

Keep it going, mate! Learning R was my new years resolution, and I am totally hooked to it now.. 

", "votes": null, "replies": [] }, { "id": 8112, "author_name": "jpetterson", "author_url": "", "post_date": "01/15/2012 23:02:33", "content": "

Hi Vielauge,
\r\n
\r\nI'm not sure I understood your question, but have you tried this:
\r\nas.integer(dimnames(questiondiff)[[1]])
\r\n
\r\nOr, in a data frame:
\r\ndata.frame(question_id=as.integer(dimnames(questiondiff)[[1]]), questiondiff)
\r\n
\r\nJames

", "votes": null, "replies": [] }, { "id": 8123, "author_name": "vielauge", "author_url": "", "post_date": "01/16/2012 19:07:25", "content": "

Hi everybody,

\r\n

many thanks for your responses. 

\r\n

@Shea Parks

\r\n

Thank you. Your answer works and is definitely better than googling and searching in a lot of help documents for hours (with no result).

\r\n

@James Petterson

\r\n

I will try your answer too, but have not done it so far. I am sure I can learn something. (At first sight it looks quite similar to Shea Parks idea).

\r\n

@shaz24

\r\n

I will see. I am already addicted to kaggle, but R could be a problem. With two small childs (2 and 5)  and regular work there is not too much time to get into R. I guess 44 days will not be enough for me to get to top twenty with such elemental programming\r\n difficulties. But maybe I can get the benchmark with a total different method. We will see.

\r\n

Have a good time. All of you. Bye.

\r\n

Vielauge

\r\n

 

", "votes": null, "replies": [] } ], "raw_markdown_by_id": { "8105": "", "8109": "", "8111": "", "8112": "", "8123": "" }, "source": "meta" }