So, the antidote to doing things by hand is that you should always, you should try to teach a computer to do whatever it is about to do. So, that's the so the idea is that you're going to, you want to try to automate, in some sense everything that you do in your analysis. So if you, if something and particularly if you're doing, working with data, the parts of your kind of project that work with the data, that process the data, you should always try to teach a computer to do it. for, there's two reasons why you should do this. One is that if you can, if you can teach a computer to do it, then you have a solid rules and solid, kind of concrete instructions for what to do. And so there's no room for fuzziness to to kind of play into the data analysis because the computer doesn't accept, generally speaking, won't accept that kind of fuzziness. They need exact programming instructions. And so, when you give instructions to a computer, essentially what you're doing, is you're writing down exactly what you mean to do and how it should be done. And that is essentially the definition of reproducibility, right? So, if you can teach a computer, it's almost guaranteed that what you're going to do is re, is reproducible to a large extent. And so even if you're going to do something just once it's useful to try to teach a computer to do it. And it may involve using a different, using a variety of programming languages. You might use R, you might use something else. But if you can get your computer to do something for you that's always better for reproducibility. It may be less convenient for you because it takes longer to program a computer. But it will, it will pay you back in the sense that everything that you do in this way will be reproducible. So, as a quick example something that's very common that you'll have to do in, in most projects is to download some data and this data might be available on the web. So for one example you can might go to the UCI machine learning repository go to that website. And then you might go to simply like you know, there's the Bike Sharing Dataset, so you click on the bike sharing data set web page. You click on the data folder and then inside the data folder is a link to the zip file which contains the data set. You can choose the kind of save linked file as in your web browser then you down, then, then you have to name a file on your local computer and a folder and then save it, right. So these are all things that are very natural to do. But you'll notice that the list of instructions that I had to discuss was actually rather lengthy for this very simple operation. So, an alternative is to just teach your computer to do all that. And in fact, in R, it's quite straightforward. You can use the download.file function. And, and then just download the file directly onto your computer. So, one of the nice things, actually, this, the the example has some advantages. Because, first of all you'll notice that the full URL to the dataset is specified, right? So, there's no ambiguity there. You don't have to click on the website, then click on the folder, then click on the zip file. You get the whole URL to the, to the dataset. It's right there. It's very specific And furthermore, the name of the file that you save to your computer is also specified. So, for example, if you download a file using a web browser, you always have the option of renaming the file to be something else. But if that's not documented then, then it's very difficult to link the renamed file to the original dataset on the web. But here I explicitly specify in the code that the local dataset is going to be called Bike-Sharing-Dataset.zip. And so. Now I know what file to look for after this code has been executed. I know exactly which directory I saved the data set to. So I saved the data set to a directory called ProjectData. And so now I don't have to wonder what directory the dataset was saved to, well, you know, for example in the, in, downloading it from the website through the web browser. And this code can always be executed in R, can, even if you only need to do this once. Someone else could execute this same exact code, as long as they're using R and they'll, as long as the URL and the website hasn't change, they'll download the same data site, the, the same dataset. Now, of course there are many things that are out of your control, so for, so for example, if you don't run the website then you're. You'll have to depend on the operators of the website not making lots of changes. But to, to the extent that things are in your control, the more that you can code things like this for example, this simple download.file function the better you'll be. Another general piece of advice is to use some verse, some sort of version control software, right. And there are many different types of software in this course and in this series in general. We, we tend to focus on Git which is a very nice version control software and it's coupled with nice websites like GitHub and BitBucket which can be used to kind of publish code and other types of projects. So, I think the main feature to keep in mind when you use a version control system in a data analysis project is that it helps to slow you down a little bit, okay. So you might, that might sound like a bad thing but slowing people down is often good, because we often, when you have a project and you're very excited about the data, and you want to get into it. You just start doing things and you start working with the data, and you're not really keeping track of what's happening. And so diversion control software whatever it is that you use will help to kind of make you stop, think about what's been done, think about what changes have been made, commit changes to a repository so that you can have a record of what's been done. And then kind of move along step by step and this is useful for a data analysis because you have a log of kind of what has happened and what and what direction you've gone. If you want to go back, you can go back and there's a record of what you found. So, if in, in the course of, for example, your exploratory analysis or whatever it is you're doing. You find something interesting. You have a history of kind of how you got there. And so, version control systems can be very useful for this type of for this purpose. Of course software like Git can be used to, you know track and tag snapshots of, of, the project. You can revert to old versions. All of this kind of stuff is pretty standard in version control systems. And so it's useful if you can to as you're progressing in a project to add project, to add changes in small chunks. So don't just add like 100 files at once in a massive commit. But, if you can break your analysis down into logical pieces and add them piece at a time, this will help you into in reading the history and understanding kind of what was done in the past. Keeping track of your software environment is very important for reproducibility because a lot of complex projects will involve chaining together many tools and merging different datasets and some tools and datasets may only work with certain environments. And so the software in the general computing environment can be, can in fact me critical for reproducing an analysis. So, there are a number of different features that you might want to keep track of in your software environment and not all of them are necessarily critical for every type of project but just keep a couple things in mind. Which are fairly common. The first is your computer architecture and this is generally, it's usually not too important it was probably more important in the past. But but a general understanding of what kind of computer architecture you're working on is useful if you need to communicate that to someone else. So the CPU that you are using, is it an Intel chip, is an AMD chip, an ARM chip. Who makes the CPU can be useful. Another aspect of that is whether it's 32-bit or 64-bit that can have an impact on some software. Are you using graphical processing units or GPUs. Things like that are useful to keep, to kind of have in your mind to keep track of. The operating system can be very important. So, are you using a Windows operating system, Mac OS, Linux, Unix? Some other version of Linux? This can be very important because some software only works on Windows. Some only works on the Mac, etcetera like that. So, if someone wants to reproduce what you've done. And you've used a piece of software that only runs on Windows. Then they're going to have to find themself a Window, a Windows machine or use a Windows emulator. Your software tool chain is very important and typically you'll be using lots of different pieces of software. Some of it will be standard for example the Web browser pretty much everyone will have a Web browser. But things like compilers and interpreters the shell that you're using, are you using the Bash shell or something like that. Different programming languages that you use, C, Perl, Python, R or whatever it is. If you're using different database back-ends, it's important to know which ones you're using to any data analysis software. All these things should be noted because people, if someone wants to reproduce what you've done, they're going to have to reproduce this entire environment, all the compilers that you used, etcetera things like that. And so, that's very important to keep in mind. Any supporting software particularly things like libraries software libraries or R packages, which are, and other types of dependencies are going to be important to keep track of. Any external dependencies so these are things that are kind of outside your computer. You know external websites. Are you downloading data from, from central repositories. Are there other remote databases that you'll be querying. Do you get your software from other software repositories. So things like that. Maybe important for your analysis. And for all of these things, it's usually important to keep track of the version numbers because as other people develop the software they're going to make changes that may break dependencies And so, if your project was done with a certain version of an operating system or software it may be that it's only reproducible using that version and that future versions are not or cannot be used. So, knowing the exact version number for everything if it's available, is important to keep a note of. So one small example for example, is in R. If you use the sessionInfo function. What this tends, what this will do is it'll tell you kind of as much as it can about your environment. You can see this is the output from my computer here. I'm using R version 3.0.2 Patched and you can use maybe the subversion or vision number which is 64849 so it's actually quite precise. I'm on a 64-bit Intel compatible computer x86 and I'm using an an Apple computer. And I'm in the kind of the the US English locale, so that's the kind of language environment. these, I have the following R packages installed, stats, graphics, grDevices these are all the base packages that come with R. I've got other packages installed. And then there are other packages that are, I'm sorry, loaded, and there are other packages that are not loaded but their name space is available, so things like knitr, markdown yaml etcetera. And so these are all kind of what my, my R environment looks like as I was doing my analysis. Of course, this is just a small piece of my entire software environment but it's useful to have this kind of detailed information in general.
