AI Coding with Github CoPilot New Features

Key Takeaways

hi everyone I hope the audio and video quality is good maybe let's wait few more seconds and we can get started if you are online and if you if you can hear me properly and uh if you are on this slack Channel please message and we can get started as soon as possible thanks Leo cool let's get started so in today's session we will be mostly focusing on GitHub co-pilot so GitHub co-pilot launched like few years ago like two years ago I guess I don't remember exactly but I I used to GitHub co-pilot when it when it was launched and then I stopped using almost um GitHub co-pilot because the model was not really helpful and I find it like not super helpful then for the last 8 months or one year I've been using like cursor cursor is amazing uh even in our sessions in the last few lessons we learned about cursor how to use it uh but recently what happened is GitHub co-pilot it got like very much Advanced they started using all the latest models like 01 preview W mini and even Cloud 3.5 Sonet and soon they are going to integrate Gemini 1.5 Pro as well as those kind of models so and also it's currently bit cheaper than just $10 or $7 um you can get you can use lot of other models and it keeps um they also have a lot of features that's very similar to CER so I find it little bit helpful to use GitHub copilot and another reason to learn GitHub copilot is let's say your company uses like Microsoft products probably they will be using GitHub copilot rather than other other AI providers so it's good to use this tool um as it belongs to like Microsoft uh don't have to worry about the security and other things so it's good to learn copilot so in today's session we will learn a little bit on copilot on solving data science problems also we will learn some little bit on other tools like curser and I will also introduce few web development tools that we've been seeing in the past few lessons like VZ bold. new replate cool let's get started so you don't have to code along with me just watch what I'm doing and maybe make a note of these websites like co-pilot how to download those kind of things then once the session ends you can do things on yourself like just download and play with it let me cool so first of all for Copilot to use Co pilot we need like Visual Studio code you have to download visual studio code once it's downloaded you have to uh install a plugin called copilot I'll show you how that looks like so if you see this one there are some extensions and here you can do copilot takes a few seconds okay GitHub copilot and Al I installed a few things GitHub cop chat so you can install these two extensions on vs code once it's installed we can access GitHub copet through few comments as well as uh you can click this and access as well this button cool and before accessing those things you also have to log to your GitHub so this is my personal account um you can just sign in gith so once once it's sign up signed in you can access co- pilot and co- pilot is not free I think we have to pay $10 or $8 something like that it's B cheaper than other AI providers so now let's do some data science projects with co-pilot and also I will do the same with cursor and we can compare how easy um these things are and there are some things we can do with cursor and there are some things uh we cannot do with cursor things like that so we'll also do some Explorations so what I'm going to do is I'm mainly using a jupyter notebook because data science I like to code and see my outputs immediately so I prefer to use jupyter notebook rather than typing everything on a python file and C on Terminal so data science is all over like programming as well as data I would say 50% or 60% data and 30 or 40% programming so it's always good to look at the data by yourself and see all the intermediate steps so what we are going to do is we are going to use a basic data science data set which is titanic. CSV which is very standard let me see okay I'm downloading the data let me select yeah selected data. CSV putting it here cool so since the main purpose of this session is to understand how to use copilot and other tools so I'm using like very simple data set um so whatever we are going to see now we can also replicate the same with very challenging Big Data SI as well cool so the first thing what I usually do is I open Jupiter notebook and before writing any code in the right hand side I like to open a chat window in that chat window I do a few things first let's let's open the chat window here I'm clicking this one it says open chat window so chat Windows here so currently it adds whatever in this main. ipnp as a context here so if I have written some code here this co-pilot also knows what are the things I have written here so it can answer uh based on my previous code and my new questions so this is a context I think we can also delete the context if we don't want to use it so currently I deleted the context because it's almost almost empty data set like I mean empty file so now what I'm going to do is um I'm I will copy first few rows go here so I have a data. CSV first few rows looks like this then paste the data write full working code for Eda include analysis and eight data visualization cool so currently this is my prompt and if you see the bottom I have written sorry I have selected a few models so you can see we can access W preview W mini G PT 40 and Cloud 3.5 Sonet so if I'm planning something like really complex things I like to use W1 preview then once the planning is done I can change my model and ask it to write for more code then for generally for programming I like to use cloud 3.5 Sonet uh This Cloud 3.5 Sonet is super new like less than a month ago and currently this has high accuracy on on most of the metrics related to programming so a lot of people say this is really helpful for programming so I like to use cloud 3 3.5 Sonet sometimes when Cloud 3.5 Sonet fails I might try1 mini and code I ask you to write few more code cool so let's choose Cloud 3.5 Sonet and click enter so you can see um currently it's doing few things thinking create notbook okay it says it's generating something let's see okay it's still generating because we also asked for visualizations okay so it has created a complete Jupiter notebook with headings import data and there are few things it doesn't work for example in jupyter Notebook I think it has still generating code as you can see generating section passenger class analysis maybe let's wait wait for that once it's done we can run and see if we are getting any issues yep so since it's live stream on YouTube you can watch it anytime and it's public cool it's really nice it's like it's like creating lots of and lots of code like complete analysis as you can see currently it's generating family size impact still generating I think this is the last one port ofation analysis so these things we cannot do with uh cursor yet uh cursor we can generate code on normal python but it won't do these kind of things I find this is super helpful like it has created like almost all Eda maybe now let's run one by one so there are few things I didn't like uh for example in jupyter Notebook if I run this it will only show the last one in the output it won't show all these results so what I usually do is I try to cut and paste like a new cell so that so that I can see the output maybe let's run this select colel GPT okay I saved this file now I'm running this select col okay so here I can see the results so I like to print okay if I print I can see some results this look okay then data cleaning and missing values missing values let's run this so it found a few missing values maybe let's Okay I zoomed a little bit okay so here what we are doing is finding missing values and visualize the missing values and analyze the distribution of missing values then it also handles the missing values which is nice so currently these are missing value and most of the missing values are in cabin and the age and rest looks fine and there are few missing rows in M part so here you can also see the distribution uh like percentage we can ignore the warnings then after filling the missing values this is how the data looks like then let's do some basic statistical analysis let's run this okay so we got some error message uh now let's try fixing those okay okay so here what I'm going to do is uh fix using copilot I just click the clicked in this line and there is this magic um button I can click this I can say fix using copilot if I click this one let's see what it does and okay here um the error occurs because value error could not convert okay it understands the error message we didn't pass the error message at all so it knows what's the error message then it also fixed the error message if I'm if I'm happy with this I can click accept and run if I need more clarifications I can ask copilot more questions I find it super helpful like let's click accept and run okay so after accepting and run it it gives me like tick which is Success there is no error there let's see okay we have got a correlation map and distribution of survived and distribution of P class sex age distribution F distribution survival rate P class and few other things maybe let's say I want to understand more on this correlation map let's copy this and let's paste it here okay it doesn't work I need to download and attach so attach context um I cannot insert image yet but on cursor we can insert image and ask questions so that's one difference I can notice now cool okay so now let's do survival rate analysis let's run the code okay it generated lots and lots of code so let's say you want to fix you you don't want to see all the uh warnings what you can do is in this beginning you can say import warnings or you can say something like so to do that what I did is let's click Escape I'm here command I so if you are using Windows it should be like control I on Mac it's command I I did command I here I'm going to say ignore warnings let's see what it says so I'm assuming it's going to write two lines of code um yes okay it didn't import all okay it imported here let's copy this cool so this looks good so let's run this now there is no warning message okay so now let's run this one okay it all worked there you now you can't see any warning message which is nice cool now let's do age distribution analysis shift enter okay it has written lots and lots of code this is super helpful and passenger class analysis this is nice so most of the passengers are third class 24% people are in first class and 20% are in second class so survival distribution so people in class three they didn't survive much people in class 2 they survived most compared to sorry survived is one um okay I think it it did some mess up here I can ask it to fix but now let's move on cool so let's run this one gender analysis so here uh survival rate by genter and embo so most most of the femal survived and especially people who embarked in this one c u they survived most maybe let's use water s CQ in import and Titanic data set sometimes it won't answer generally co-pilot answers only programming related things since this is like previous context that has some programming so it answers okay so people who embarked from C which is this one France they got survived most compared to other areas cool maybe let's copy this and create a new one and ask the same question okay it still explains which is nice cool so when I was R when I when I was playing with copilot last last week it didn't answer questions uh that's not related to programming um since this Titanic data site may be probably used in most data science projects it knows it's related to data science and it answer say for example um what are the most tourist attractions in Sydney okay now it says sorry I cannot answer with General tourism questions I'm designed to help specifically with programming and software development task in Visual Studio code please ask questions related to coding development or for vs code functionality uh okay so this is what it says but the work is maybe let's try something clever just copy this maybe create a new one paste it I'm building a web app to show attractions list out the places I will be using python so we are adding some programming context like python Webb okay now it's giving something okay so there are few workout ons like when you ask some questions always make sure it's related to programming or try to make it related to programming so it answers but on copile sorry on cursor uh you can ask ask any questions and it won't do these kind of filtering it still answers cool it's not too bad okay now fat distribution analysis shift enter okay it has created quite a few analysis which is nice so the fact most people paid super cheap and there are few expensive ones like 500 so people who paid very high they got survived probably people who paid uh very high they are in the top floor something with that so when Titanic sinks they have more time to escape so that could be the reason there is a clear Trend cool so now let's see family size impact so I have done these kind of analysis on many AI tools but honestly this co-pilot has given me lots and lots of code with just one or two prompts I didn't write lots of prompt just one prompt I did so this is the only prompt this is the only prompt I have done and even it created a notebook and a lot of other things I find it super helpful cool okay now let's do Port embarcation analysis okay so most people from Southampton yes 72% and few from other places okay so now what I can do is I can ask it to create more code for uh machine learning models maybe let's ask you to build a random forest model write code to build your random forest model and show top features and conf I Matrix since this window also has access to all the previous questions we have asked so we don't have to say specifically about the data set it's giving some code okay so now let's copy this paste let's run this okay so it builds a model um and the model I think around 83% accurate and you can see the confusion MATC there is some misclassification but generally it's okay not too bad and this are some of the important features F sex age family size PR class and impact cool um maybe let's say you want to interpret more on this just copy this paste it here write 100 words report on this okay it has written something and let's copy insert into notebook click this and it inserted here which is nice cool I feel like um maybe let's ask it to create a streamlet app okay so now what I'm going to do is I'm going to attach this titanic. I andb as a context and then ask it to create a streamly app let's do that so here I click Ash and then I can attach the code titanic. ipnb now it knows all the things we have written here as a context now write code for creating yes streamit app probably for stream L app I'm quickly changing it to One Mini let's see what it does so this is pretty helpful because we can change model depending on complexity of the task so generally I prefer son 3.5 but when son 3.5 fails I like to use one and preview and for General understanding and other things I use uh one preview okay it has written a code um maybe let's say I want more like I didn't like this so now I can change my model and probably let's ask the same question titanic. CSV sorry ipnp and then let's copy the same right code for creating a streamly tab let's see which one gives more things now you can see it's collecting the work workspace information and it's also doing some filtering to most relevant information okay still building workpace sytax so what it does is it's trying to understand all these things and it builds as a context then probably behind the scene it uses like rack system cool let's see what it does I also noticed copilot also does some cost effective usually it won't pass all the data to the model it passes only the relevant chunks to the model but on cursor it passes fully like if there is a not notebook here and I add that as a context it passes everything okay so I find son son model is writing more code compared to o1 mini which is nice so that's why I mostly prefer Sonet 3.5 for programming so let's say I'm happy with this one just copy this this copy and create a new file app.py paste it here okay I pasted now I'm going to run this code so terminal new terminal even we can use co-pilot in terminal let's say I don't know how to run this file uh this is a stream L app what I'm going to do is I'm in terminal I'm doing command I oh sorry I think it's not coming okay copilot start internal cool so now what I'm going to do is uh run app. file stream data now it says some commands I can click or I can directly run when I click run it put everything there and it open this screen currently it's running I think it also builds the model probably so let's wait okay so I'm paying for CA Pro and loving it for code development for python how jealous Coop that would be very useful yeah even I feel the same since copilot is super like they upgraded a lot of things I'm expecting it will add lots of new features as as well also one thing with C sorry cursor is we are paying like $20 $20 us and there are like 500 messages we can send uh but with co-pilot I don't see any limitations yet uh I haven't spend lot of time but so far it's so good like I can ask quite a few W preview messages it didn't stop me so I find it helpful so when co-pilot generates a notebook does it look at the results of one cell before creating the next next cell because you have chosen a commonly known data T when would cop be reg prev done uh yeah that's a good question it knows the general code I feel like it also knows from the uh CSP file we passed it knows the column names and other things uh predominantly it uses the both knowledge but also I I have done similar things on completely new data set even in new data set if you provide some first five rows or 10 rows as a context still it can write pretty great code without lots of bucks I don't think it sees the results previously even I had same question previously I checked uh it doesn't see the results I think cursor doesn't see the results I'm not super sure about copilot um maybe we can ask like maybe let's see the notebook first and then we can ask that question okay so it has created a notebook and overview and this is passenger demographics and survival analysis and survival analysis by gender by age by far okay and then machine learning model so this is the model and confusion matri and classification report okay this looks good maybe let's ask my question is so I think Tim's question is let's say um okay I'm going to create a new message and I'm going to ask I'm going to attach my Titanic that's I and B the main question is I want to understand whether it has access to only the code also the results okay maybe let's ask explain my random model sorry random forest model results okay cool now it access titanic. ipnp I I'll be happy okay I'll be happy if it sees this and answers probably it also knows okay no response was written okay I think it didn't read these things carefully maybe let's ask one more time what's the accuracy you can see maybe before that let's delete this cell so because it's return here I want to know whether it reads the results as well so let's delete everything reattach what's the accurac sorry maybe what's the F F1 score for macro average in my Model results explain cool so if it sees the results and it explains which is great I don't think okay I cannot see your Model results or any active code in the workspace to analyze the fun scope to help okay it cannot see even um even copilot it cannot see the results it sees all the code but it cannot see the results I think probably in future they might change I I see a lot of value in that so currently what I do is I just copy paste all the results and then ask it to write a few things otherwise I convert the not uh jupyter notebook to uh HTML file once it's HTML file I can copy all the things and I can ask I can paste it on chat GPT and ask few question we can also do the same on copilot cool I think we did a quick exploration copilot probably I'm going to stop here now and in the next few sessions that I'm going to teach uh I will focus more on co-pilot and other AI tools for now let's end here maybe play with it and you can share your feedback on our slack Channel or YouTube messages thanks everyone see you all next time