Hey everyone, this lecture is going to be about building an actual r package. So we've already seen kind of what is in R package and what are the different components of the package. So I thought in this lecture I'd like to make do a little demo, and going to show exactly you know how to put all of the pieces together. In particular, how to do it in our studio which actually makes it pretty convenient So the package we're going to build today. Is going to be a very small package. It's only going to be two functions. And the functions in the package are going to be. Are going to build a prediction model for a high dimensional data set. Using just the top ten predictors. So if you want a little bit of a background on kind of what the rationale for this model is, I'll put a link up. You can look at the link right here. and, you can read more about it. So that's the package that we're going to build. It's going to be called the top ten package. And so, we'll do that in R studio. And you can see how you know everything is done. Okay, so the first thing that we need to do is start up R studio. So, let's just start it up here. And I might as well just make it full screen, while we're at it. Okay, so, this is the kind of the default, screen for R studio, the first thing we're going to want to do when we start on R project, sorry on R package project, is to kind of start a new project, I'm going to open up a project menu here and say new project and we want to start this project in a new directory, so I'll click on that then what we want to do is build in R package, and the name of R package. You can create whatever name you want, but I'm just going to call it the top ten package. Cause that's the predictor model that we are building, and it's going to be in the sub directory of my home directory which is fine. So I'll just hit create project. Okay. So you can see. That that puts me in the directory here, where where the packaged files are going to live, and you can see these are the package files here and you can see, it's kind of put up a bunch of default files in there, there is a description file, a namespace file and then, if you look in the R directory there is a, there is a file, there is a code file but you'll notice that there is actually nothing in it. So, we'll have to fill that up in just a second. let's. So let's just navigate this, file directory a little bit more. In the top level directory see there's a description file we talked about. If you open it up you'll see it's pre filled a bunch of the fields for you. It put the package in. Name in there for me. But we're going to have to fill our, you know, with the the title, the author, the maintainer, the description, etcetera. So why don't we do that just right now. So the title of this package is going to be, the top, excuse me. Just the we'll call it, building a prediction model from the top 10 features. And version 1.0 is fine. The date is fine. Who wrote it, so that's going to be me, let's say. Uh,the maintainer's going to be me, but the important thing, of course, about a maintainer is that there has to be an email address there. So it'll be me again. [NOISE] And then the description is just a little bit longer. Than the title so I'll just say you know build functions for [SOUND] prediction models from selecting. Actually building, let's say building and predicting from models from selecting the top ten predictors in a data set. Something like that. It doesn't have to be too long. And then the license, we'll just choose a nice, open source license like, Gnu Gpl. So, we'll just call that Gpl version 3, and that's our description file. So I'm just going to save that, and, by hitting command s. And then I'll just close it up because we don't need it anymore. So, I think at this point there really aren't any other files that we can fill out, so we might as well just start coding away. Okay, so the first thing we're going to want to do is build R top ten functions. Let's call it top ten. And we will take two arguments x and y. So the x is going to be the kind of matrix of predictors and y is going to be the vector of responses so p is the number of columns that is going to be p predictors. Uh,I think it doesn't really make sense to look at the top ten predictors if there aren't at least ten predictors, so we're going to say, we're going to check to see if p is at least ten. And if it's not we'll just stop. >> [SOUND] So the way this model works is basically, for each predictor in your kind of matrix of predictors, we fit a, a univariate regression model of the, of the response on each individual predictor. And then for each individual predictors, so there is going to be all these regression models that we fit and for each predictor, there is going to be a p value associated with that, that given predictor, you know, depending on how strong the association is. So, so for every predictor we're going to have a p value, and then what we're going to do is we're going to sort the p values from kind of smallest to largest and then we'll take the top ten smallest in this case p values, which indicate kind of the, the predictors of the strongest associations. So we'll take those top ten predictors, and then we'll fit a separate regression model with those ten predictors in it and that will be kind of the final prediction model. >> Alright so we're, first, so we've already checked to see if there's at least ten predictors. And we'll initialize our vector of p values here. This is going to be an empty vector of zeros. And the original loop through each of the predictors and righ, and fit univary regression models. And fit lm y to the x, and it'll [UNKNOWN] i. And then we want to, we want to get the p values out, so we need the summary of that model fit. And then the p values for that predictor, p value for that predictor, is just going to be the, key in the coefficients element. And it's the forth, column here. And so we are going to accumulate all of those p values. And then once we've done that, we want to, we'll look at. We want to sort the p values. We'll look at all of the ordering of p values. In order by default, we'll sort them for lower to giver the index indicies for the smallest to the largest. And then okay. So then and then we just want to take the top ten. So the so the the order, what we really want, just the, as one through ten here. And then we want a, so now we create a new data set called x10. With the kind of top ten predictors, from the old data set, the original data set, and fit this final linear model, [SOUND] and then grab the coefficients, from the model and so that's kind of what the top ten function returns. So we've got our top ten function, which kind of fits the model, it, it goes through all the p values and picks the top ten. So I'm going to write one for function that that takes the coefficients from the final fitted model, and it takes some, kind of, additional input vectors, and gives you a prediction for the new values, right, so it takes the coefficients and some data and gives you predictions, a predicted response for each one. So we will call that, lets call predict10 and it's going to take a matrix of predictors. Now, this matrix can only have ten predictors in it because that's what the model, final mode says and then a vector coefficients is called b. And then it's pretty simple. We are going to do, first thing, we are going to do is I am going to add the intercepts to this matrix and then we're going to do a matrix multiplication of the new kind of data and the coefficients from the model. So, I'll do x times b is the matrix multiplication and then I want to drop the dimension eh, so that it gives me a vector I'll call drop. So this just returns a vector predicted responses. Okay, so we've got our code for the package it's pretty straightforward. Just this two functions, but in order to have a package of course, we need a, a bunch of other stuff; we need some documentation, and we need to be able to specify the namespace file. So the way that we're going to documentation in this package is we're going to use Roxygen 2 package. And so the, what the, what's nice about this, the format is that allows you put all the documentation in the code file itself. And then what the Roxygen 2 package does is it strips out the documentation that you put in the code file and it makes the man pages, it formats it in the appropriate way. So, you don't have to worry about, documenting separate, writing out separate documentation files. Two nice things about this is one is that it kind of keeps you focused on one file. You don't have to kind of constantly switch back and forth. And the other thing is that since the documentation is actually close, physically close to the code, there's a better chance that the documentation will stay up to date because you'll be able to see if there are any discrepancies between the documentation and the code itself. So, let's go ahead and do that. So the first function I want to document here is the top ten function and so the first, so one of, I need to give it a, so the first you want to do is give it a little hash then the and then the apostrophe there, and I'll give it a title so I'll call it building a model with top ten features, and you can see that our studio nicely kind of insert those extra hashes for you. And I'll give a little title, description here so this function develops a prediction algorithm based on the top ten features in x that are most predictive of y. Okay seems reasonable. And then we want to start documenting some of the arguments so first we can use the parameter here say, there's an x argument and say this is a n by p matrix of n observations and p predictors. There's another parameter here called y, [SOUND] and this is a vector of length n representing the response, okay, and I will save the return value Is a vector of coefficients from the final fitted model with top ten features. There's a couple other things we can fill out. Such as the author, and that's me. If there's some extra details we can write out, somebody will write some details here for the reader. I'll say this function runs a univariate regression of y on each predictor in x and calculates a p-value indication the significance of the association. And, the final set of ten predictors is taken from the ten smallest. P values, okay, so that's a little description there. You might want direct readers to the LM function just in case, you know, they want to know where the hava linear models fit. Now what's important about, actually important about this function. First of all, it's going the be the function that we want to export. So we want to make sure that we export in the main space file. So we gotta put the export directive here. And then furthermore, because this function uses the LM function, which comes from the stats package, we need to import that function, so that we can use it. So we need to import from the stats package, the LM function. So that's going to go into the name space file. Alright, so that's our documentation for the top ten function. Now let's go down here and look at the predict ten function. So we gotta do, probably gotta do the same thing here. Start a little documentation with the prediction of the top ten features. And this will say this function instead of coefficience. Produced by the top ten function. And makes a prediction for each of the values provided in the input x matrix. Okay, it's in that little description there. So what parameters does this function have well.there's x there's b, then we know it returns a numeric vector containing the predicted values. Alright so what's x, this is an by 10 matrix, containing n new observations. And then what is v? V is a vector of coefficients obtained from the top ten function. Alright. So, then, of course, we also want to export this function. So we hit the export directive. This function doesn't export anything special. So, we don't need an import directive. Alright. So, let's save that with command s here. Okay, so we've written our R-code. We've written some documentation in the R-code, and so now it's going to be time to kind of process this information and start building our package. So, let's see how we do that in R Studio. So in R Studio we can go over here to the build tab and you'll notice that this bill tab only exists if you, if your, if you've created a project that's specifically an R package okay. So let's try to build our package and then load it into R. Okay? So you can see that at the top, in the top window here, the R session has restarted, and then the top ten package has been loaded. Okay? So, one of the things you'll notice, if you go over here to the left, in the files, if you, let's go in the man directory here, Oh, and you see that there's a, there's a default kind of, documentation file here that talks, that just kind of describes the package. But you notice that the documentation files have not been created by the Roxygen 2 package. And so what we actually have to do is we gotta go into here and configure our build tools. And then we, what we want to do, we want to generate documentation with Roxygen and we want to create the RD file from the manual. Let's create the namespace file from the documentation in the, that we wrote. And then we'll do it when we build and reload. 'Kay? This over here on the left this package documentation file is not required, so I'm just going to delete it because I don't feel like writing it. Alright, so let's build and reload R package again. And you'll notice now that the the two documentation files here for our code have been written, so if you look in the top10.rd file you can see all this information has been extracted from the the documentation that we wrote in the R file, so this is all of the stuff that we just wrote, and for the predict ten function again, another set of documentation here. So, Let's take a look at our top ten package here. I can do library, its already been loaded but lets take a, library help top ten, ops, and you can see here's the description file that we wrote out as a bit of information these the exported functions over here, ahm, if I look at, if I, I can print out the code for top ten. And you'll see that's the code that we wrote. If I do question mark top ten, you'll see that the documentation file that, has been built is right here. So, it, it's formatted nicely here. We can look at the documentation for predict ten. And you see that here's the documentation that we wrote. Should be all familiar to us. Let's bring this down. And so that's R package is pretty much all, most of the way there. If you want to see if it passes R command check, it's very simple. You just click the check button here and it will run our command check, it will build the package and then run R command check dot. And so you see all the tests are going by. [BLANK_AUDIO] Okay, so you can see that R command check finished and it looks like we passed all of the tests. You see everything is okay over here so that is a good sign. So I'm not going to upload this package this to command, but but we could upload to command because we would see that we pass our all the tests in R command check. So this very simple R package, we just wrote a little. We wrote two functions, but you can see that in R studio there are a lot of handy tools. For putting a package together and for building documentation files. And so try I,I encourage you to try it out and write, write some functions, maybe take some of the homework that you've done already and build in R package out of the functions that you've written. It's not too hard and particularly when you use the tools in R studio, you can go very quickly.
