Toxicity Classifier Streamlit App

Key Takeaways

hello everyone welcome to AI anytime channel so in previous video we created an endpoint of the toxicity classifier model that we created using supervised machine learning algorithms so we created this endpoint in fast API and we tested that API it was working completely fine so what we will do in this video is that we'll create a very simple streamlit application okay we'll take that model and we'll try to integrate and we'll have an UI where there will be an input text area and then we will give our text and in it will then detect or classify whether the text is toxic or non-toxic and once we build this application we'll also deploy this on share streamlit okay through GitHub so if you see currently I'm on my GitHub uh repository toxicity classifier app if you haven't uh downloaded this notebook or the API endpoint you can download it from here and you can test it out so now what we are going to do okay let me go back to my vs code you can see I'm on my vs code just give me a minute and yeah so you can see this this was our API dot Pi file where we created the endpoint you can see with our endpoint was predict and we what we did we we have loaded our model here okay and once we have this model and TF IDF we created this endpoint called predict and then you know we uh ran that model and got the response okay now what we will do here guys will create an app so I'm calling it app.pi and here we'll use streamlit which is a web framework you know to uh create powerful data science application okay it's it's very easy for you know the developers who have very limited knowledge of web Technologies okay okay so if you see uh on requirement.txt I have added streamlit Library we have to install it as well I have already installed it so I'm not going to install what I'm going to do now I'm going to just import streamlit add HD and rest will be same so I'm just going to take a code snippet from here and what I will do I'll just use this here so what am I doing here we have extremely it as St we are also importing pickle to load the model and TF IDF features and then we have numpy for mathematical operations and then we have psychic learn packages you know to run and do the processing of you know feature extraction and other things right so now what we have to do here let's create some functions okay so what we will do first function will be you know to PF IDF this is going to be our first function I'll just take the code from here so what I'm going to do take this code let's call it load TF IDF load TF IDF and then I will just pass this here and I'll just return this return PF IDF and the next is to load the model so I'll say load model it will be again like same so we'll have uh let's take the model code from here NB model and let me just copy and paste MB model looks fine and we'll just return this NB model so now we have created two functions one is for loading the TF IDF here you can see we have saved here right with pickle and the other is the nav based model OKAY the toxicity model the next thing is let's write the function for toxicity as well so toxicity detection and here we will pass a text as an input and then we'll predict the first thing is to load the TF idea because we have to do the transform we have to use the transform function right so TF IDF and we'll use load TF IDF so this is the variable where we are using this function right we are loading it so load TF IDF now the next is we'll use this transform so you can see let me just write over here you can see now we are uh using the transform we are applying the transform function on this text which is an input parameter and then we are converting it to an array so to array and once we have this text T of IDM now we have to load the model as well so model and then here we'll have load model so here we are storing our model guys okay so model and then we'll do couple of these lines let's copy and predict so you can see the prediction it will not be n b model here because our variable has been changed to model now so model dot predict and then we have text DF idea variable we are passing it over here to not get the prediction that's what we are doing map the predicted class twisting I think that is this is good so now what we can do we can just say return class name this is what we are returning okay so this is the function that we have created so now we have three different functions here the first where we are loading the TF IDF the second we are returning the nav based model and the third for the toxicity detection where input parameter is text and we are doing all our operations on these functions okay and now in we have the class name the result now what we will do we'll create an estimate application and in that let's first do St Dot header I'm calling it toxicity you know detection app this is going to be this HD dot header toxicity detection app now let's also use a server header kind of a step you know just to let end user enter the text below into the text below and now what we have to do hd.serve header uh enter the text below now what will do HD Dot let's call it text input St dot text input and here we'll have entered your text enter your text now this is the text input that we have HD dot header hd.sub header and text input SE dot texture now what we will do we'll use this function on this texture right so now let's say call it a result so result equals uh text uh what we're going to do now guys that we have to load this toxicity reduction we are not going to fast text which will be text input we have to pass this variable yet which is storing our input text let's do one thing let's do not do that okay so what we can do we can say if text input let's first check if there is a value also or not is not none and then we create a estimate button we call it SD dot button so that will be if HD dot button and let's call this button analyze analyze and this result you can see this is another result and now what we can do uh result let's call it HD dot sub header detected result is below something like this and we can do HD dot info we can use the status messages uh we have several status message option in streamlit being info success warnings you know there there are an error as well so we'll use H dot info because this is a kind of info that we are you know presenting on the UI saying XT dot info what we can do guys we can just write the result is and we can do this plus and that will be our result guys so result and okay excuse me this is going to be the hd.info HD dot info is done sub header result now let's run this guy and see if we are able to get the desired response so what we did we first imported these libraries streamlit for the web UI and then pickle for loading the models and the TF IDF then we have numpy and Cyclone packages then we created two functions you know to load the TF IDF and the model and then we created one more function called toxicity detection where we have written our code for toxicity detection using these functions and then we created a streamlined snippet where we have headers operators and input text area where the end user will give the input text and then we'll run the model on top of that input text so let's now run this so how we run an estimate application is like we do streamlit run my file name is app.pi so you can see this is the application that we have is a toxicity detection app enter the text below enter your text now in this I'll just say I'll just use a toxic okay what did it right there such a so I'm just using this uh statement here such a f word you are right and let's analyze so you can see the detected result is below with the result is toxic okay the result is toxic for this now let's try something else the product or the product was you know fantastic and I really liked it and suppose this is my review right and I want to just find it out if this review you know done by the end user is toxic or non-toxic so you can see the result is non-toxic in this case right so now let's do one thing okay what we'll do if you come back here on vs code you'll see that you know we have created this estimate application and use the same kind of you know found here on this GitHub repository toxicity classifier app and you know in the next video we'll deploy this okay on share extremely okay this that you see that app.pi okay so this streamlit application the toxicity detection app you know we will deploy this application on shared streamlit through GitHub okay so they kind of provide the space like you know around 1GB where you cannot deploy your application stimulate application in free so in today's video uh just to summarize we created a toxicity detection app that takes your input text and then classify classifies that between toxic and non-toxic so if you like this video guys please like this video and uh subscribe the channel if you have not subscribed yet and please share the video and Channel with your friends and to peer okay thanks for watching the videos guys see you in the next video