Chat with a CSV | LangChain Agents Tutorial (Beginners)

Key Takeaways

foreign [Music] good morning everyone how's it going today welcome to this new tutorial on Lang chain in this tutorial we're going to be building this application that you see in front of you it's very similar to the previous application that we had built in which we were able to ask questions to a PDF just that in this case we're going to be asking questions to a CSV right this is a very educational video because not only are you going to be able to build this application yourselves but this will also be useful for you as an introduction to the concept of agents in launching right so let me show you real quick how this works in order to demonstrate this application I'm going to use this data set the breast cancer Wisconsin data set which contains data from a lot of cells in different tissues samples and we're going to be able to ask questions about this data so I have downloaded it here I will upload it and then I'm going to be able to ask a question about the CSV for example I can ask a question what is the mean radius of the malignant tumors in the CSV and if I ask the question you will see that first it starts to to to think itself and then it's going to Output an answer the mean radius of the malignant tumors is 17.46 right I mean it's very useful it's very convenient and I'm going to be explaining to you how agents works with a beautiful diagram and all right so stick to the end for that all right I hope you enjoyed this video all right so as usual let me just tour you around the setup that we have here in for our project um I have my main.py application right here which is where I'm going to write all of the code um I have this pretty quick test that tests whether or not I'm importing my application so if I am importing the application this will not run if I am executing it directly then the main function will be executed with it which is where all of my code is going to live then I have my git ignore file just to get my sake to tell get which files I don't want to track then I have my DOT env.example in which I have the example variables that I'm going to be needing for this project to work in which case in this case we're going to be using on the openai API which is why I need this variable right here which is open AI API key remember that when working with line with blank chain we need the variable to be named exactly like this it cannot be open AI key or open AI Etc if we're working with blank chain it has to be this exact name and then I have my real.env file which is the one that actually contains my secrets okay if you want a more detailed explanation about the setup just let me know um the API Keys you can of course get them from platform.openai.com if you don't have an account you're going to have to create one account and then you're just gonna have to click on API keys right here and then create new secret key you can give it a name then create key then you copy this and you paste it right here right all right so that's it for the setup let's just start coding right now all right so the first things that we're going to be doing is we're going to start creating our graphical user interface okay in order to create this user interface we're going to be using streamlit which is a very lightweight and very very minimalist and intuitive way of creating graphical user interfaces in Python in order to do that of course you're going to have to have it installed so what you're going to do is to pip install streamlit like this I already have it installed so it's not going to take a huge time for me but to you it might take a little bit longer and once that is done you're gonna have to do import streamlit s s t because that's just the the usual the usual name we give it and then right here what we're going to do is we're going to SD set page config this is basically just going to allow us to get to tell uh streamlit that this is a streamlined application I am actually going to disable co-pilot because I always forget to do that before I start the videos but alright so St set main page config all right then right here this this function can actually take several parameters but in our case we're all we're just going to give it one parameter which is the page title and we're going to say ask your CSV and then why not let's add a little Emoji right here of Fame um of a chart yeah like this there you go now in order to create a title or a header you're just going to do header SD header like this and then just your title I'm going to say ask your CSV as well then just use the same emoji there you go and then of course afterwards we're gonna have to read our CSV streamlit comes with a very convenient input element which you can call just by saying St dot file uploader I mean input when I when I say input I mean file file uploader sorry about that and here it takes a label um we're going to call it upload your CSV file like this and then I'm going to specify the type that little file that I expect to get from here and let's say that I'm going to I'm going to store the the object that the user is going to upload in a variable called user CSV there you go now if I save it I'm going to be able to come back here and I'm gonna be able to do streamlit run and then I can just run my main Dot py there you go and I have my application right here um now as you can see just with a very few very few lines I was able to create this very quick graphical user interface with python which already is able to take my CSV file however so far we only were only able to upload the CSV file but we're we're not able to do much more than that so let's let's add a little bit more functionality to that and actually do something with the CSV file right all right so now that our application already has a CSV in itself and is able to use it to do things we're going to be using a launching agent in order to actually parse it and to read it okay why do we need a language agent because a CSV file is not readable by by a language model right it's not like a PDF file like you if you remember in our previous video and if you haven't watched it I invite you to watch it in our previous video we we only had text and our language model can take text and can read it and can understand it but a CSV file is basically just a table with data and the language model is not able to understand it what we're going to do in order to be able to read a CSV file is to take our language model to understand the question that we are asking and to have it generate the actions that it has to perform in order to understand the CSV file for example our agent is going to be able to perform actions in this way let's say like taken from the official documentation if we have our agent we're going to be using the create CSV agent pre-built agent from langching then we're going to create it like this and if we run it let's say for example if we want to say how many rows are there the agent is going to be able to think by itself and say oh it's asking me how many rows there are so I need to count the number of run of rows then it's going to be able to perform an action on this on this CSV file and then it's going to be able to observe the output of that action in our case the action that it performed was to calc to execute the function Len which Returns the length of your data frame and now it is going I mean once it reads the output of that action it's going to be able to think again and to give you a final answer and this is very convenient because that makes the agent pretty autonomous and it allows you to ask sort of I mean somewhat complex questions about your CSV file okay um here's some more a more um yeah here here's a more complete example for example what is the square root of the average age so it it'll first think that it has to calculate the average age then it has to calculate I mean then it knows how to calculate the mean but as you can see it's not actually knowing how to do it doesn't know the data right by reading it from the CSV it has to think in order to get the date okay let me show you a quick diagram of how this works all right so there we go I confected this really quick diagram just to show you what is actually happening um an agent in general I mean this is an introduction to any agent not just to say to our particular CSV agent which is built to be able to perform actions using pandas and other libraries on the data set an agent is basically a tool that is able to make decisions by itself and to use tools to interact with its environment okay in this diagram as you can see an agent works like this let's say we have the we are our user and our user makes an input the input is it can be a question or something like that in our example it can be what is the age of Leonardo DiCaprio's girlfriend uh to the 0.43 power all right I think the agent of course he doesn't know that but in order to think of a solution it's going to use a language model from our question it's going to think that oh I need to find who is Leonardo DiCaprio's girlfriend and then from that thought he's going to be able to choose a tool from from a set of tools that it has available let's say that it can have a calculator a coding interface or for example the web it might be able to search the web or to search the web or things like that okay in our case since it doesn't know who is Leonardo DiCaprio's girlfriend is going to go and search the web then there you go it search the web it searches the web and then it finds out that his girlfriend is Camilla for example um and then it's going to read that output from the tool that it used and it's going to interpret the output it's going to it's going to be able to ask itself do I already have the solution and then again through the language model is going to be able to think and it's going to be able to say oh no I still don't have the answer and if it doesn't have the answer it's going to go back to this step right here and it's going to be like all right so I need to use another tool and then it's going to select a new tool now it's going to be like all right so now I know who is this who is his girlfriend now I'm going to uh maybe it's going to have to use the search web tool again in order to be like all right so what's the age of Camila then once he has the H is going to come back and be like all right now I now have the H what do I need now I need to elevate it to the power of 0.43 and this time it may not use the search web tool it's going to be it's going to be able to use the calculator tool and then once it has that final final solution it's going to be able Again by using our language model that in our example is open AI um I mean the gpt3 from openai but in your computer you might be able to you might want to do it using Auto GPT or GPT for all and then it's going to be able to generate a solution and then read that solution to us and that's exactly what is happening in our CSV in our CSV agent is taking the CSV and being like all right so I need to do this and then it's going to be selecting the tools to use and then reading the output of whatever it did and this is like the most basic explanation that I can think of to how agents work and what is that we're going to be doing exactly right now so let me show you how this looks actually in the code um in our application okay let's do that all right so the first thing that we're going to do is that whenever the user already has uploaded their CSV we're going to test that that CSV is actually uploaded if user CSV is not none then we're going to be able to use it and in order to use it we need the user to ask a question so I'm going to be like user question and the user question is going to be a tech a very simple text input something like this text input that comes this is a component from streamlit and I'm going to say ask a question about your CSV all right there you go now if we look at how this looks like let me just open the terminal and just run it again remember that in order to run to run a streamlined application you have to do extremely run and then your file you cannot do just like Python and then your file that will not actually execute the streamlit application so there you go now if we upload our CSV file we're going to be able to ask a question but so far it doesn't do anything so let's do something with that question that the user gave us and in order to do something we're going to be actually needing a launching so as I mentioned before we're going to be using the pre-build launching agent called from we're going to be installing importing it from langchain.agents and we're going to import create CSV agent and then from line chain llms which is the language models we're going to be able to import open AI now I am using openai but it is in this step that you can actually use whichever Lan language model you want uh launching has support for a huge deal of language models if you come right here to modules agents nope two models and then in models if you go to llms and you go to Integrations you'll see the huge list that they have for language model providers okay so yeah don't don't hesitate to use your whichever you want um now what I'm going to do as well is I'm going to import from the 10th oh in order to do this actually I will I'm going to need to pip install python.env which is the package that is going to allow me to read the variables from my DOT end file okay and in order to do that I'm going to have to import from the 10th 1.10 I'm going to import load.nf and in order to enable my these variables to actually be read I'm going to have to execute this function at the beginning of my file okay now launching should be able to get this API key without myself doing anything else okay all right so now we have imported our language model and our agent so we're going to initialize our agent in order to do that I'm going to I'm going to first initialize my language model I'm going to say that my language model is going to be open AI and this I mean in the case of open AI this one takes several parameters which are optional in my case I'm going to set a temperature of zero um I believe it used to go from 0 to 10 but in in launching apparently it's from zero to one so zero is non-creative and one is completely creative um a good way to think about this is that if you set it to zero it's going to return exactly the same thing every time you send the same prompt if you set it to the maximum it's going to be more creative and give different answers every time all right I'm going just going to set it to zero and then once that I have my language model I'm going to be able to initialize it inside of my agent so oops come to initialize a variable called agent and I'm going to use my create CSV agent function that I imported from langching just a moment ago and this function takes llm I mean my language model that I previously created as the first parameter um and secondly it takes um sorry and secondly it takes the the file the CSV file that I'm going to be using and remember the CSV file I saved it right here from the file uploader I'm just going to pass it right here and what why do I have a problem positional cannot appear after keyword all right so just not going to use there you go I'm just going to be using only positional arguments and then last but not least I'm just going to set a verbose option to true what does this do variables actually just make do you remember that when we had our example right here we see the green text right here with all the thinking process of the model of the agent this is basically just going to print that at the same time if we set it to false it's not going to print anything it's just going to do it by itself all right um something that you might want to consider right here is that um in particular about the CSV agent our model is actually running code okay so it's using pandas or other libraries to get information about our about our data set so it's running code by itself this can potentially be quite dangerous if you make it right code that is malicious for your file or or something like that okay so um just keep that in mind that it's running code by itself it's not asking anyone for permission to execute the code so yeah just keep that in mind um once we have that once we have that we're going to we're going to be able to actually use this agent to answer our question so if we have our question let me just show you so far like right now what it what this looks like I'm going to do if my question is not none and my question is different to nothing I'm going to just estimate your question was and then the user question I'm going to just wrap it within F string then let's remember that we have to in order to run it we do extremely drawn Main and there you go now I can actually upload my CSV then ask a question hello and then your question was hello okay everything seems to to be working correctly now we can actually use our model to treat that question right here okay so in order to do that here right here we're going to have our response and the response is going to come from our agent all right so in order to do that we need to call our agent first so we're going to do agent dot run in order for it to run some instructions right here we're basically just going to tell it the question that our user that our user has and so in order to do that we're just going to pass in the user question right here and then if I save this and actually you know what I'm not going to run this right here I'm going to run it at a terminal run just beside it to show you the thinking process at the same time as it's reading the file okay so I am right here this is exactly the same directory as where I am in my code so in order to do that I'm just going to do as usual streamlit run in my file is mine.py and as you can see it runs just like usual there you go and now what I'm going to do let me just put this one right here this one to the right um there you go there you go now I'm going to upload my CSV file I'm just going to put myself right here and then now I'm actually able to ask questions so I can say the previous question that we asked before was what is the mean radius of a malignant tumor of the malignant tumors then if I click enter I'm going to see all the thinking that our agent is doing right here so let's see the thought is I need to find the mean radius of the malignant tumor so it can actually think and it has to decide what it's going to do now it decided to use the python interpreter as a tool and it passed this input so I mean this is going to return the the mean value of the radius for the malignant cells and there you go so I so now once that it executes this it can read the observation then it can think now we have the final answer and then you can just return the final answer in our program um we can actually test it a little bit further we can ask what is the mean area of malignant tumors or how about something a little bit different the main um what does it mean smoothness of benign tumors for example if I click enter you can see that it's going to think again well I already run something here um you can see I need to find the mean smoothness of benign tumors again it thinks that it has to use the python interpreter it runs this code it observes the answer and then it thinks I now know the answer and then it just gives you the answer um using the language model okay um so yeah I mean this is how how to build this very quick application and a very quick introduction to agents so there you have it I hope this was an a good introduction to the concept of Agents I hope that you liked this application I will leave in the description the link to the repository on GitHub so that you can download it and play with it play around with it um just I don't recommend that you copy and paste the code from there I recommend that you actually write it yourself so that you can um like it gets into your head and yeah I hope that it was educational and please subscribe for more videos about Ai and software development So yeah thank you very much for watching and I'll see you again next time [Music] [Music]