Github Copoilot live demo!

Key Takeaways

so hi everyone I am kadaa and today I'm here to teach you or to tell you um how to get the most out of GitHub co-pilot so GitHub co-pilot is an AI pair programmer that helps you to code a lot faster and it also goes beyond your editor and I'll be showing you a lot of those features today so I'm cadesa I work at GitHub as a developer Advocate I've been a software developer for about four years and I am Jamaican so if bust out in some Jamaican twang because I can't figure out how to debug the code live forgive me forgive me um my Jamaican vernacular will definitely come out but today is not about me it is about GitHub co-pilot we're going to skip over this slide and go straight into it so just in case you're unfamiliar GitHub copilot is an AIP programmer and it lives in your editor as an extension in your IDE it also is in your terminal it's on github.com and it's in most multiple places on github.com to help you go a bit faster so you can use it in Visual Studio Visual Studio code movm and Jet brains idees like pie charm over there um you can also use it with multiple languages Ruby um Java JavaScript typescript python of course and I've even seen somebody translate code from one language into cobal or from cobal into another language which I thought was pretty slick um I've also SE develop scene developers use it with uh terraform to do ISC stuff so there are a lot of use cases with GitHub co-pilot now how can you make the most out of this tool I have six tips to share with you today and then we're going to jump into a demo where I'll be attempting to build either successfully or I will fail in front of you an ml model with GitHub co-pilot live so the first tip I have for you today when using the tool is to make sure that you have your files open open files means you'll be able to give the tool context as to what you're building and what you're doing all right the second tip is to provide a top level comment and a top level comment means you know how sometimes when you're trying to talk to your cooworker about what you're building you really can't figure it out so you give them a like a overarching view of what you're trying to build so that they understand where you're going it's the same thing with GitHub co-pilot you want to let it know where you're going what you're trying to do because if you don't tell it it won't no right um it's not like a it's not a guessing game with GitHub co-pilot the third tip for you is to set includes and references and setting includes and references essentially mean tell it what version of a tool you want to use so let's say you're trying to build a flask API instead of just saying build a flask API for me say build a flask API using version X um to do XYZ you know what I mean be specific in your requests and you will get back better responses the fourth tip is that meaningful names matter and this is essentially like having good programming practices and principles within yourself and within your editor is only going to give you better results essentially meaning garbage in garbage out one thing about GitHub co-pilot is that it will show you whether or not you're coding properly uh and so if you're finding that the tool is giving you code that looks like a little iffy think about what's currently in your editor okay um five the fifth tip is to provide specific function comments and so separate from providing that overarching view of what get of of like what you're trying to do specific function comments mean like let's say you're trying to ride fetch airports to fetch a list of airports from your um database instead of just saying Fetch and then write your function say fetch airport data or fetch airports and this will give GitHub co-pilot additional context into what you're building with llms it's all about context llms are large language models it's all about context so the more context you can provide the tool the better results you will get the sixth tip for you is to provide sample code so sometimes it's important to tell GitHub co-pilot like what you're looking for if it's not giving you the thing that you want so like I know sometimes for me I ask it to do something and then like it doesn't do the thing and I'm just like oh my gosh but as soon as I give it an example of what I'm looking for it's almost like turning on a button and it's like oh this is what you meant here you go and I have a much better time using it one thing about using GitHub co-pilot in your editor and Beyond is don't get frustrated with the tool it is an AI it's not here to replace your job it's not here to do your work for you it's here to help you go a bit faster it's all about being a more productive and efficient developer now I have a a demo today with for you and if you go to g.o for / Pyon demo you will see the repository that we'll be using today and today I will attempt to build an ml model live with GitHub cop pilot so let's see what we can do so I have a data set here and this demo is actually based on a larger Workshop that we do uh with software developers on using the tool and the first step is to build an AI model so that's what we'll be doing today the data set that we have is uh a list of observations of us domestic flights from 2013 and it has a lot of information this is the data set flights. CSV and it's just a Big Blob so let's go into our notebook to get a bit of uh a clear understanding of what's going on in the data to do some exploration so I have some notes right here on my tablet just so I don't forget anything because whenever I do live demos my brain goes to Mush and I can't seem to remember a thing all right so if I look over here that's why I'm looking over here so so the first thing I want to do is to say if I hit command I or control I on my keyboard it will bring up inline copilot chat co-pilot inline chat and so that's what we'll be using right now so I'm going to say I want to load a data set and see the first few lines all right using pandas that looks good I'm going to click accept and this is not the path because I didn't provide it with the path so it's at data flights. CSV and there we go click run and that should run for me and as with any other data set uh one of the first things you want to do okay let me remove print and then it should look the way I want it to look awesome and so we have a lot of information here now as with any data set you want to make sure that you clean the data so like remove any null values uh remove any outliers so let's do some of that work right now so if I click on new cell I can say and this is going to be like that top level comment for the function find so I'm just going to start typing just so it's in a function I'm going to hit tab remove sum because I don't need it to be summed then I'm going to say I'm going to replace these values all right there we go in place equal true that looks good to me and so the next thing I want to do is to I want to normalize the data to remove outliers think it's step delay and arrival delay yep and here you see it's giving me a suggestion but it's it's missing arrival delay right here if I go to the next suggestion there it is so I can click accept to get the second one and if ever I'm working and I don't know what any of the code means I can highlight it I can go to copilot chat and I can say explain and it should grab that context that line that I highlighted because it's highlighted and there we go it's telling me exactly what's happening all right that's actually not what I wanted to do so I'm going to remove this and I'm going to say uh find the outliers from this is why I have notes calculate the Z scores is what I wanted to do actually which it should be the same it should be the same code I think or similar all righty there we go now I want to find the absolute Z scores that's exactly what I wanted to do and then I want to okay and now I should have some clean data to work with now if I want to see how how many rows and columns I can ask a question like how do I see how many rows and columns I have and it should tell me to use the shape attribute which I expected and so I call this a dummy question and so sometimes when you're working you don't remember how to do certain things and so you want to ask that question but if you don't want co-pilot to keep that context within the conversation you can hit this x button right here in vs code to remove it now let's keep going something else you can do right here is you can ask questions about your data so if I do CU what does and there was a strange name up here this one so I can say what does this mean and it's going to give me an answer right here so like I can have a conversation right in my editor about the data I data I have because it has the context off the data in like its memory or whatever we we call it um but let's keep going so let's say I want to create a chart this is also something I can do using GitHub copilot so I can say I want to calculate the average arrival delay by carrier and sort in ascending order then create a bar chart how do I do that that it's going to say to calculate this here's a step-by-step guide and then it's going to give me the code to do that work I can hit this inserted cursor icon here check the work and these are all items in my data so I can hit the I can run it and there we go we have a bar chart of the average arrival delay by carrier now let's commit this work and then we're going to go into creating the model so something else is you can get like commit messages so if I hit this sparkle button that's not a read me let's try again okay let's just commit it regular way so I'm just going to say updates for now but make sure your comment messages are a lot more has a lot more information let's keep going so if I wanted to create a model I can ask GitHub co-pilot and give it this prompt it's quite long so I wanted to say create a model to predict the likelihood of a flight being delayed by 15 minutes on a given day of the week and the arrival airport use logistic regression split the data into training and testing data sets and train the model and then calculate the accuracy of the model and then this function should give me an output of the accuracy score that's what I'm expecting so let's see what what it gives us and as you see like I provided so much information right there like I told it exactly what I wanted to do I told it exactly what um algorithm to use and you know just to give it a guide so I can get the best results so I'm less frustrated you know what I mean okay delayed so just looking at this I know that delayed is not a part of my data set so what I can do is I can actually ask it this give it this prompt again and then if I hit hash and I do file I can attach my CSV file to the conversation because if you notice I don't have my CSV file open so it's not getting the context of that information from the file if I enter I should get a different response it's giving us a plan then it's giving us the code it's loading the data and doing all the pre-processing okay didn't need that but let's see what we get and if we get an error we'll debug it together don't need this don't need this pre-processed I don't believe I need that I don't believe I need this I'm going to remove these and I need day of the week destination airport name I actually want to use destination airport ID array Del 15 yep random State 20% M do fit y prediction all right do you see any errors sometimes like like I said earlier I can't see so let's run it and see what we get and we have an accuracy score how about that so in a few minutes I was able to build a very simple relatively simple ml model with GitHub co-pilot but like think about how much faster and how much more efficient and how much easier your workload would be with a tool like this to help you by your side you know what I mean if I wanted to know what this accuracy score meant I can copy it I can go to my besti chatty that's not the official name it's GitHub capella chat and I can say what does the accuracy score dot mean and it's a measure of how right how cool is that if I don't understand something I can just say what does this mean please help me um and it gives me something right there something else I love to do with getup co-pilot is I like to write tests with it so how many of you actually write tests for your code right exactly exactly it's a problem so if I go into my server file so my server is consuming the models you know I can ask co- pilot to export the models so I can create an API and it did it earlier so this is just uh a faster going through it so if I click in this app and I hit this sparkle button right here copilot pops up I can do test hit the Run button and it's going to give me a file that I can hit the create it pops it in a new file for me and then of course I don't think this is going to run on the first try I wouldn't expect it to run on the first try but this is a great start for you to go in do some debugging and have some tests for your code it's very important to test your code if you're not going to do tdd it's okay but at least write some test to make sure that your code is working the way you expect it to be working and GitHub copilot can help you with that so we just did a bunch of stuff so let me go ahead and do a commit let me stage let's try the sparkle button again to see if it gives us a better commit message add unit test for Server endpoints and created a created the model we'll say so let's do a commit let's publish the brunch we can create pull requests in our editor but I don't want to do that today so let's go on github.com let's hit this button and let's open up a poll request now how many times have you open up a PR that looks exactly like this no information no context no nothing you just send it to your coworker with Vibes and say please review and they don't even know what you're talking about with GitHub copilot on github.com you can use this PR summary button it's my absolute favorite button because now I don't have to try to remember everything that I just did once I hit that button it is going to generate a summary for me and depending on how much work you did it's going to give you like links to the exact files the dips and such that you can use so like right here right it gave me links to like the airport test airports the test predict the test home so like different test files it didn't link to the notebook which I find very interesting um so I'll bring that back to the team um but if I go to to files change I can also chat about uh the code that I'm reviewing so let's say your cooworker sent you sends you this code to review you're reviewing it uh you can ask co-pilot about the code so I can like select a portion of the code this is a test file but just for you know example sake and I can say like what is this code [Applause] doing or I can ask it uh more intelligent questions of course about the code that I'm reviewing but I think this is such a great tool to get up to speed a lot faster with the work my co-workers are doing because I'm doing my own work I really don't have time to review the code and so this is going to give me a much faster outlook on what's happening in the in the repository or in the pr something else I like to do is let's say you were assigned a brand new repository so this is uh an API it's a fragrance recommendation system and essentially if you put in fragrances that you like it's going to give you suggestions on fragrances that you may like based on the notes of the of the of the perfume it's something I'm working on it's demo and so let's say you were given this file and you you were told build the front end for this app and you're just like what what is this API how do you use it the documentation is lacking it doesn't tell me how to get started with it there's nothing there are a lot of repos out there like this that we we're given right I can click on my copilot friend and I can say let's say I want to know like where the Canon algorithm need to use because that's going to be be the main driver of this API so I can say where is the and what I'm expecting is like links to like specific files that I can then click and look into and then I can then ask it further questions about what's happening in the code yep so if I click here there's a notebook there's the read me there's this fragrance recommend that file Pi file and I can go in here and I can view and then I can go ahead and ask additional questions about what's happening in the code I think that's pretty cool we also have something called knowledge bases knowledge bases is essentially like you can talk to your your private repositories on GitHub so primer is our open-source uh Design Systems so let's say I wanted to build the frontend UI to accept responses or like data from the user like except their their current fragrances that they like so I can say something like I want to um get a user's favorite fragrance and I'm trying to decide should I use the text input or the text box component I'm trying to decide which component to use based on our internal design system what it's doing here it's going in it's looking at primer and it's telling me okay I can use text input component except it's designed to accept a single line of text which make it suitable for short free form input like a fragrance name now I can go in and say what if I want to to get like five okay it gave me a different component that I wasn't even thinking about with an example of how to implement that I think that's pretty cool there are a lot of cool things I think about get up co-pilot something else I love is that we can do uh what we call general purpose chats so if I go [Music] out great so I can hit general purpose chat and actually this morning a professor came up to me at the booth and he says hey I teach Python and I teach C I am looking for information on teaching my students how to use AI with C I don't know anything about C and so I went to get upub co-pilot and I said are as a general chat are there any resources if I can type or repos on AI with CP um learning resources for students pretty General question and I was hoping to get a list of repositories that I could share with the professor and that's exactly what I got and I was so impressed I was like Wow here you go sir here's what I found for you in like a couple seconds um thanks to GitHub co-pilot and if you see this Bing sign get up get up cellet actually did a web search using Bing to find these resources and then link them directly to me which I think is what pretty cool and look at that I was able to share this repo and it's an awesome repo with a lot of resources and how to use AI with c and net we also indexed I believe over 100,000 of the most popular open source I keep hitting the mic sorry of the most popular open source repositories and so I can go through a repository like Matt plot lib and I can say I can say something like uh how do I create a subplot with mt plot lib hey and it will give me an example so I know a lot of times we well I know I was I was so scared of getting started and open in open source uh in the open source community and like contributing to open source and I think this feature is amazing because somebody can go on an open source project that they want to contribute to they can ask questions about the project they can get started with the project um they can go to an issue and ask GI up co-pilot to summarize the issue and it will summarize the issue for you and help you along the way I think that's what pretty cool the coolest thing I think that we just launched with GitHub co-pilot is called co-pilot workspaces and it's essentially what we think is going to be like the future of sofware development we'll see where this leads but I have a demo video here to show you if you please and the sound won't work so I'll probably just like talk over the video so here he has an issue that he's been assigned and he's going to hit the co open in workspace button what co-pilot workspace is going to do is it's going to go through your code base and it's going to say okay here's the current implementation that's a bit blurry isn't it here's the current implementation implementation of what's in your code it's not there here's a proposed solution of how we can implement this in your code base then the developer is going in and he's going to add some like additional uh specifications that he wants then he's going to hit this generate plan button and co-pilot workspace is going to give him a whole plan of how we can do this he's going to add more files once he's finished with adding his additional specification he's going to hit Implement plan and GitHub co-pilot workspace is actually going in and it's implementing the code like it's it's actually implementing the plan that it just suggested and I think that's amazing but as you see the developer is still in charge of reviewing the code making sure it's accurate making sure that it works to what we're looking for and he opened it up in a code space because copilot work space provided him with a link that he can open up in a code space to make adjustments he's doing some debugging on the tests now he's going to create a pull request he's going to get a description that was generated by GitHub co-pilot and all the work that it did he's going to hit that create pull request button and workflow complete he just did that issue so quickly with the assistance of AI now if you want to learn more on how you can maximize your usage off GitHub co-pilot there are two blog posts that I would recommend that you read on the blog. github.com it's uh using GitHub compiled in your IDE tips tricks and best practices and then 10 unexpected ways to use GitHub copilot I wrote them but they're pretty great I promise they have a lot of great tips I would love for you to share these tips on social media just let me know you can find me online at it.d it's that lady Dev love for you to tag me and let me know what you think about this session and that's all I have for you today thank you [Applause]