Machine Learning Tutorial Python - 15: Naive Bayes Classifier Algorithm Part 2

Key Takeaways

here is a list of topics we are going to cover in this video in the end we have an exercise for use of watch tilde and I have this file with spam emails where I have the text body of the email and the first column says whether it is ham or spam ham means it's a good email it is not a spam okay and you can see that whenever you have spam you have Tom's like free message or free entry or winner you know this just by reading it it definitely looks like a spam I have loaded this particular CSV file in a panda's data frame as usual and I'm going to now do some data exploration to see what's going on with this data the first thing I am doing is just grouping it by category and describing it so that I know that there are four thousand eight hundred and twenty fives hem and 744 for the seven spam okay so that much I know that a good amount of spam in our data set the spam column is text I want to convert it to a numbers because we all know that machine learning models they understand numbers they don't understand text so we have to convert the category and message both the columns into numbers somehow okay the first one is very easy category ham and spam we can use 1 and 0 so that's what we are going to do and the way to do that is by using this apply function here when I say DF category it takes the category column and applies this particularly aMDA function on it the lambda function takes each individual values and checks if it is spam then it will put a return a value of 1 otherwise 0 and we will create a new column called spam here and when you execute this you can see that all the ham for 0 the spam are 1 here once that is done we can import the train to split method from Ascalon as usual here I have imported it and I'm gonna keep my tastes size to be 25% and when you run it it's it is splitting the samples into train and taste dataset okay once this is done we still have a message column which is text so that text column we definitely want to convert into indoor numbers now the way we'll do this is using count vectorizer technique in count vector as a technique let's say you have these four documents or the email bodies with all this text one of the ways to convert this into matrix or a vector is you find out the unique words in each of these documents and you'll find that there are nine unique words combinely in all of these documents now you can take each of these documents and you can treat these nine unique words as features or kind of like a column and you can build this kind of matrix okay here this is the first document second document and so on the first document you say end is 0 the occurrence of end is zero times you see that there is zero document appeared once first appeared once similarly in the second document the document appeared too so you say document here document here so this is a simple technique of representing versus count okay and we can use these individual columns as feature for our problem I took this example from SK learn documentation and here is a code snippet from SK learn documentation it explains the same thing but with the Escalon api and these are the AP as we are going to use for our data set I used count vectorizer and created the metrics which I showed you in the picture so it created probably many features that's why you see dot dot dot and these features are equal to number of unique words in our corpus corpus is basically all the unique words that you see in this huge data so you realize it will be many columns in over metric now naivebayes have three kind of classifiers like Bernoulli multimode multinomial and Gaussian in the last tutorial we use caution I bias here is the Quran Sarang what's the difference between the three and this guy accepting really gave a good answer where burn only is basically when your features okay features not the target variable when your features are 0 or 1 they're binary in nature that's when you use this multinomial is when you have discrete data for example your movie rating ranging from 1 to 5 Gaussian naive Bayes is when you have normal distribution or a bell curve in your features ok we are going to use multinomial naive bayes here for our problem and the way you do that is by writing this code just to say oh the typing time I'm just copying and pasting the code but you kind of get it you run model dot fit function on your x train count and VY train remember the X train count is basically the text which is the emails converted into a number metric once the model is trained it is ready to make a prediction so let's have two emails okay so we have this two image now the first one looks like a good email where a friend is asking another friend to go for a football game and the other one clearly looks like a spam and when you run it you see it detected the second email as one which is it's a spam all right now let's measure the accuracy or the score and the way you do that is first X test you need to convert it into count because our model is designed such that it works only on numbers and then you can feed it to model for election and you find that the model performs really well with 98 posts probability so you can see that for spam non-spam type of problem the naive Bayes model works really great now you found that the converting it into a metric create a little bit of inconvenience in dumps of when I was supplying X test count I had to perform this transform method also when I was trying this test emails I had to call this transform method before giving it to my model Escalon has a nice feature called pipeline where you can define a pipeline of your transformation here what we are doing is on our raw data we applying some sort of transformation before feeding it into our model right now we use only count vectorizer some people use more than one one transformation people you like tf-idf and so on so in that case if you have SQL on pipeline it is super useful and convenient and I am going to show you how to use the pipeline so the first thing you do is you import the pipeline like this and then you create the pipeline using your pipeline steps so my first step is count vectorizer so just to remind you on what I am doing here is I'm trying to simplify the same code base so the code work till here ok our model is ready it score 98% fine but since we had to perform this transformation steps I am writing the same core using a simple API and here I created a pipeline with two steps first step is convert my text into the vector of count vectorizer and then apply the multinomial naive bayes and when I have my classifier created what I am doing is I am going to train it now this time when I train it I can train directly on X train so remember what is X train Xtreme has this text okay in the previous example we use X train count we converted X into count and the entry in the model here we can directly feed the text into our model because internally this pipeline will convert to a vector first and then it will apply now naivebayes on that so when you run it you can see this works okay and again you can project the performance of our classifier it is same 98% okay and just to verify if the emails prediction works okay you can run this and you can see the first email is not spam the second one is spam and those are these two emails all right now comes the most interesting part of my tutorial which is the exercise I always give this example that if you want to learn swimming by watching swimming videos you're not going to learn swimming okay what do you do you have to move your butt and jump in the swimming pool similarly if you want to know coding you have to code and I have prepared these exercises with so much effort for you so why don't you take your laptop and just work on this naive based exercise all you have to do is load Escalon datasets find it ahead and classify those wines into one of the three categories using naive bayes you can use gaussian and multinomial classifier and tell me which one performs the best in the comments below i had a link of this file exercise file in the video description below also the tutorial code that was shown in this video that code link is also available in the video description below so all this code is available so you can just go ahead and try it there is a solution link by the way but a good student will not click on the link without trying it first on his own so I assume you all are beautiful good students you will try this thing on your own and then only look at the solution to match your answer thank you very much if you liked this video give it a thumbs up share this with your friends subscribe to my channel and please post a feedback in the comments below it really helped me it helped me improve my content thank you bye