I understand that you are using regsubsets() from the {leaps} package. If that is true, you're not simply subsetting your data based on self-specified criteria.

The answers anyone gave you so far concern regular subsetting of data with general subsetting functions from {base} and {dplyr}, subset()

and select(), respectively.

This shows how important it is to phrase questions as clearly as possible. It also helps to be precise when naming a function or package you are working with, because you might otherwise mislead readers.

With regsubsets() you select a number of best fitting linear models based on something like the Bayesian or Akaike Information Criterion (BIC, AIC). That can help you in subsetting your set of possible variables for your model.

But it is vastly different from normally subsetting your dataset. So to give you a clear answer, we would need a more precise question. What are you really trying to do? If you just want to subset you data, use the base function subset(). If you still need a faster method for that (simple subsetting of data), have a look at this post:

http://stackoverflow.com/questions/8936525/fast-subsetting-in-r

Best,

Angelo

The questio is: What's your objective? 

try creating a new object as in: subset1

I decided to use backward selection, it seems working but i could decrease until 53 variables, which is still better than 90. I ll try dplyr package.

thank you for opinions. 

Idris

I understand that you are using regsubsets() from the {leaps} package. If that is true, you're not simply subsetting your data based on self-specified criteria.

The answers anyone gave you so far concern regular subsetting of data with general subsetting functions from {base} and {dplyr}, subset()

and select(), respectively.

This shows how important it is to phrase questions as clearly as possible. It also helps to be precise when naming a function or package you are working with, because you might otherwise mislead readers.

With regsubsets() you select a number of best fitting linear models based on something like the Bayesian or Akaike Information Criterion (BIC, AIC). That can help you in subsetting your set of possible variables for your model.

But it is vastly different from normally subsetting your dataset. So to give you a clear answer, we would need a more precise question. What are you really trying to do? If you just want to subset you data, use the base function subset(). If you still need a faster method for that (simple subsetting of data), have a look at this post:

http://stackoverflow.com/questions/8936525/fast-subsetting-in-r

Best,

Angelo

Thank you for your response and I could except that I didn't give right explanation here we go, I used regsubset because it's reduction ur data as using smallest AIC value. However, thank you again ur link but that one is not proper for me. I said i cannot use regsubset() because it takes time. Eventually it is going to work but i need another library do same work or has a based idea like using smallest AIC or BIC or so on. 

I thought to use regression model and took out not significant variables from my model. However, I still want to use a library and a method for subset selection. 

I always open to new ideas or opinions. 

Idris

Thanks for your elaboration of the question. In this case my link and the other proposed answers will really not help you. We therefore have to find something else.

First: Are there no assumptions about the model you try to fit? Are there no variables that are more likely to predict the outcome based on theoretical considerations? Can you therefore not exclude some variables based ons such considerations? It feels rather strange to fully and blindly fish for the 'best' variables computationally.

You could also run something like a principal component analysis (PCA) on your variables first, to shrink your variable pool. See for example:

http://stats.stackexchange.com/questions/27300/using-principal-component-analysis-pca-for-feature-selection

and here is the {princomp} package in R to do that:

http://stat.ethz.ch/R-manual/R-patched/library/stats/html/princomp.html

Second: The routines used in the {leaps} package are direct adaptations of (what I guess to be rather efficient) Fortran code. So I don't think you will find a more computationally efficient package in R that does the same thing.

But, as I read in the documentation of {leaps}, the function leaps() is actually a more efficient way of doing what regsubsets() does. As stated in this documentation, this is because it uses an "efficient branch-and-bound algorithm" compared to the backward and forward stepwise or sequential replacement that regsubsets() uses. So why not use the leaps() function? - I tested it and it runs with a maximum of 31 variables only, so that doesn't seem to be an option with your original 90 variables. But maybe after the PCA or the theory-based selection.

There is also the {bestglm} package (http://cran.r-project.org/web/packages/bestglm/), which has a neat paper attached explaining a lot more about these best subest glm approaches, here: http://cran.r-project.org/web/packages/bestglm/vignettes/bestglm.pdf

But also {bestglm} depends on the {leaps} package and mainly uses these functions, as you can read in the documentation, here: http://cran.r-project.org/web/packages/bestglm/bestglm.pdf

I further found a comparison of the packages' functions and how to use them, here: http://rstudio-pubs-static.s3.amazonaws.com/2897_9220b21cfc0c43a396ff9abf122bb351.html

Two more things: You could of course also switch to another programming language, which would probably be faster, or use more computing power.

I would recommend using data.table package for your dataset. It has been immensely helpful for me to save time while read or subsetting data.

http://cran.r-project.org/web/packages/data.table/index.html

Dear Idis,

a bit late but you have two choices:

- change the default method="exhaustive" in the regsubsets() call to one of the following: "backward", "forward", "seqrep". The latter 3 are greedy and hence MUCH faster.

- use the more modern regularization methods such as the lasso or elastic net to perform variable selection. A quick start is given by the R library glmnet.

Let me know if you need more detail.

Best

ML