Build your first machine learning model in Python

Key Takeaways

a portion of this video is sponsored by discover data science powered by wiley more on them in just a moment in this video i'm going to show you how you could build your first machine learning model in python and we're starting right now so we're going to build our first machine learning model in python and we're going to do that using the scikit-learn library and the coding environment that we're going to use is going to be google colab it's free and it's quite powerful and so let's fire it up so typically when i create projects on google codelab one of the first thing that i would do is i would give the notebook a name so we're going to give it a name of first project and the next thing is i like to add documentation or text to the notebook so i'll add a text cell here and move it up or you can also adjust the location of it by using the down button here and then i'll double click here and i'll click on this button which is the equivalence of a hash symbol which will give the text a heading one if you have two of it it will be a hitting two so the great thing about using headings is that it allows you to neatly organize your jupyter notebooks so i'll show you so here we're also going to make the text bold my first ml project and we're going to use two hash symbol or you could click this button twice and then i'll also make it bold so typically we're going to start the project by loading in a data set so let's find a data set to analyze and for that we're going to the github of data professor and if you scroll down one of the pinned repository will be called data click on it and then there's a lot of data sets here that i have compiled over the years as a content creator so a reasonably simple and unique data set that i would like to use here is the delani data set and i think it's this one let me have a look okay so this is a data set of the solubility of molecules and they are important in the fact that they are crucial for biologists and chemists in determining whether a molecule is soluble in water or solvent and whether they will be good drug candidates and so let's have a look here you can see that the data set here is in the format of a csv and essentially it is a comma separated value file so if you click on the raw link here you're going to see the native file let me zoom in and you're going to see that the first row will comprise of the names of the columns and each word that you see here is the name of the column and it represents a single cell and then you have the comma to separate it and therefore the first word here and the second word here and the third word and etc are separated by commas therefore hence they are called comma separated values because the comma will separate the values and so here how many columns do we have we have one two three four five so we have five columns and then we have correspondingly from lines two until the end of the file they represent our data sets and so typically what i like to do is i normally will put the y variable or the dependent variable or you could also call it the output variable or the y variable so there are so many names for it and so they are the variable that you want to predict as a function of the x variables which are the ones here that are highlighted so you might be familiar with the equation of y equals to f of x right so y is the last column here the y variable equals to the function of x so we have several x here so it is a multivariate analysis okay so what we want to do is we're going to import this particular data set so i'll click here on the raw link and then i'll click here in the address bar right click and click on copy and now we're going to read in the data set into the jupyter notebook so the python library that we're going to use in order to do that is called pandas and so we're going to import it as follows import pandas as pd so pd is sort of a alias for the pandas library so from here on we're going to call pandas as pd as mentioned here and then we're going to read in the data set in the csv format and then we're going to assign it to a variable called df and df is an acronym for a data frame so let's do it we're going to type in pd because we want to use pandas and then we're going to use the function from pandas library called read csv and then ask the input argument which is inside the parenthesis we're going to type in the address of the file or you could also type in the file name so you can see here that we could directly within the data set from the url that we had just copied from github and so let's do it and then once read in we're going to print it out by typing in df and i'm going to hit on the play button here to run the cell and so because it is just loading you're going to see that it's connecting so it's going to take a short moment it's initializing and now it's connected and now we're ready now you can see the ram and the disk that are assigned to this particular cloud computing unit that we have here in the notebook all right and so once you have run the cell you're going to see the output which is right here which you could also close if you don't want to see it again or you could play it again to display it again so we're going to see the contents of the csv file in the tabular format here you're going to see here that the first column here that you see is the index number so officially it's not a column so it is the index name and here you're going to see the columns moloch p mo weight num rotatable bonds aromatic proportion so these four variables represent the x variables and so when we build a machine learning model to predict the y variable or the log s and therefore log s is equal to the function of all of the x variables here so in other words we're going to use the four variables here to make a prediction on the log s variable okay and so the next thing that we want to do here now is that we want to split the data frame into the x and into the y and so let's do the most simplest thing is to obtain the y variable so let me show you i'll create some text cell i'll make it both data preparation we have here the first level here one hash symbol we have here two hash symbol this is let's make it a two hat symbol let's make this three because we want it to be a sub section of this one and so we're going to call this data split data separation or data separation as x and y okay and so we're going to create the y and we're going to type in df and the name of the last column here is log s so that's how we're going to get the y and let's see okay and these are the y log s and now we want to get only the x variables so we want to remove d log s so we're going to do that type in x equals to d f dot drop parenthesis and then we're going to see we want to draw log s and we want to have it axis equals to 1 because x is equal to 1 will allow the drop function to work with the data as column mode however if you use x is equal to 0 it will work it in the row mode let's see if that's correct it is correct you see that the log s now gone and that we have four columns here and prior to that we have five columns so the number of rows remain the same at 1144 so now we have x and y in the separated form so the next thing that we want to do is we're going to split the data set we're going to split it as the training set and the testing set so let's do it so remember how many we need we need three hash symbols here so we're going to add text cell click it three times and then type data splitting and we're gonna use the scikit-learn package for that so you want to type in from sklearn dot model underscore selection and then you want to import the train test split training test split and now we're going to type in x train x test y train y test equals to train test splits x and y and we're going to have the test size to be 0.2 and let me see i want to have the random state to be assigned a specific number so that every time i run the code cell i will get the same data split so we're going to have random state equals to let's say 100 and now we're going to run it so we should now have four new variables here and let's have a look at the x screen and we see that we have 915 rows and four columns let's have a look at x test we have 229 rows and also four columns so x tests or x string will come from the x variable so we started out with 1144 and so 80 of thousand one forty four is nine hundred and fifteen and twenty percent of one thousand one hundred forty four is two hundred and twenty nine and so the training set here will have eighty percent of the data and the x test here or the test set will have 20 of the data and i've actually written a blog post about this particular topic of building your machine learning model in python using scikit and i've drawn several illustrations explaining about the data split so let me go and let me show you and it's this article how to build a machine learning model a visual guide to learning data science so here we have the x and y that i mentioned already and i've color coded here as orange and pink for the x and y respectively scroll down and here here's the data split so here you have the initial data set and then you perform data splitting where eighty percent of your data will go into a container that you call the training set and then the remainder or the twenty percent will go to a container that you call it the testing center and the typical ratio is 80 to 20 for the training set and the testing set so typically we use the training set to build a model and then we want to use the testing set to serve as sort of a unknown data that you want to test training set for you want to evaluate whether the model that you have built using the training set whether it performed in a robust manner against an unknown data that you simulate using the testing set okay so before continuing further a quick word from our sponsor and so a short message from our sponsor discover data science powered by wiley which is the premier information hub for the field of data science with in-depth guides on careers degrees and industry-leading programming languages discover data sciences goal is to provide accessible resources and materials for prospective students and professionals through discover data science expert driven articles and publications you'll learn more about which data science degrees help accomplish your professional goals the tools and skills that are necessary for a successful career in the field which career paths appeal to your personal interests how to land a job in data science and as you know data science jobs are rapidly expanding on a global scale with a growing need for qualified data science professionals it's never been a better time to earn your degree and pursue a career in this rewarding field you can begin your data science journey by visiting discoverdatascience.org powered by wiley or visit the link in the description below alright and so let's continue with the tutorial okay and now we're going to build the model so let me add a text cell and i'll add here to be two hash we'll make it bold let's call it model building and here we're gonna add another one we're gonna say linear regression let's have it as three hash we have two here we have three here so we have it in a hierarchical form so if you click here you're going to see the table of content of your code and so the benefit of organizing your text cells in hierarchical form is that you could see the table of contents here and then you could click through the various sections so actually instead of making load data having two hash symbol i'm gonna make it into having one so it's gonna be the same as the title and then you're gonna see that this one moved to the left a bit and now we're gonna make uh data preparation to be one as well one hash we're gonna make data modeling to be one hash like that and now we're gonna make data separation to be two data splitting to b2 linear regression to b2 okay and now it looks good to me okay and now we're going to continue by populating the code cell underneath the linear regression so we're going to use scikit-learn from sklearn.linear model import linear regression so you're going to see here that scikit-learn has several functions that you could use not only to prepare your data set but also to build a machine learning model and here we're going to build a typical linear regression model and now that we have imported the function we're going to create a variable called lr to stand for linear regression we're going to type in linear regression function here which will be represented by lr and then on the next line we're going to run lr dot fits which means that we want to train the empty linear regression model on the following data set which we specify to be extreme comma y train and then we run it you could click here or what i like to do is i like to use the keyboard shift enter which is quicker for me and the model is built and now that the model is built we want to apply this particular model to make a prediction so let me add the text here so that we could annotate it a bit more we could say training the model and make it bold find the model to make prediction and we're going to call it y underscore lr underscore train underscore pred so we're going to apply the model to make a prediction on the training set and the prediction to notify that we're going to use spread and then to make note of the algorithm that we're using to train the model we're gonna specify to be what lr here and then we're gonna start with the y underscore so this naming convention will be helpful when we have several machine learning algorithms that we want to try out and also whether our prediction is made on the training set or the testing set so type in lr.predict and then i'm going to specify xtrain to be the data because we want to make the prediction here on the x strain so essentially we're going to do the recall it's going to be making prediction on the original data set that it has been trained on and so that will allow us to evaluate the performance of the algorithm so here we're going to call it y underscore lr underscore test underscore thread equals lr.predicts and as you've guessed why underscore test let's do it let's print out the results y underscore lr train thread y underscore lr test spread actually let's just make it like that okay so these are all of the predictions have a look here so these represents the 80 of the data and there you go the remainder 20 has been predicted and we have the predicted value and the next part here is we're going to compare the predicted value with the original value or the actual value and we're going to call the new section here to be model performance we're going to say evaluate model performance because we want to compare x strain here no not not x string y train with the here ylr train so you're going to notice that they are the actual value and the predicted value okay so in just a moment i'm going to show you a scatter plot of these two values and if they lined up in a diagonal trend line and see whether they have high dispersion or low dispersion so if the dispersion is low we will expect that the performance will be good okay and so now that we see the data that we are going to use we're now going to actually perform the model evaluation let me delete it here first lead delete add the code cell and we're going to type in from sklearn.metric import mean underscore squared underscore error and we're gonna use the r2 score function lr underscore train mean squared error equals to mean squared error function y train underscore y now train thread and so these are the two variables that we have taken a look just a moment ago and now we're going to calculate the mean squared error we're going to calculate the squared correlation coefficient using the r2 score function y train and you guessed ylr train thread and so these two blocks are for the training set now we're going to do the same for the testing set mean squared error one test r underscore test red lr test number two equals r2 or and we have y test and the ylr test underscore print run it let's run values here okay they're reasonably similar performance here so we could tidy it up a bit by saying the lr mse and then we say training or to train equal or colon print and then we're going to have this one here we're gonna reuse it r2 that will be our training r2 okay here and now we're going to turn this to be test test this would be test this would be test okay there you go so instead of having four of these we're going to delete them so you could highlight multiple cells just by highlighting it and then you could click on either you want to move it or in this case i'm going to delete them okay so we see all of it at a glance here however we could tidy up this particular layout a bit more let me show you lr results and then we're going to create a pandas data frame we're going to call this linear regression lr i mean mse lr test mse lr test underscore r2 and then we're going to transpose it and let's have a look it looks like that and now we need to change the column names here zero one two three four so what we want to do here lr results dot columns and we're gonna rename it we're gonna call the first column to be method second column to be training mse and then we have training our square and then we have test mse and then we have test r square run it there you go it looks much cleaner much cleaner than this in a tabular form and so the great thing about having it in a pandas data frame like this is that if you evaluate more and more machine learning models like random forest k-nearest neighbor support machine neural network then you're going to have a data frame that will allow you to easily compare you could also sort by column the performance and that will help you to evaluate which one was the best so here you have already built a linear regression model and we're gonna try out another one which is the random forest see we have two hash symbol here so we're going to add a text cell add two and then we're going to call it random forest random forest and then you can see it here but notice that you don't see the bold text because it needs to be in a hash symbol which will give it a heading one heading two heading three you know like the hierarchical ordering so if you want this to appear here then we need to add more so this is two then we need to make three here add three and you're gonna notice it appears here at three it might be good because you could also you know hop around the notebook like this you know click on the various topic of your choice and then you could skim through your jubilee notebook and also the great thing is that you could take a look at your table of content without you know scrolling up and down to see what's the name of the cell because sometimes your output might be quite long here and it might take some time right but it'll be much quicker to just navigate by clicking on the particular link so we can see here that we have training the model here so we could just add section called training the model training the model and then let's just add the headings and then applying the model to make a prediction then evaluate model performance so we could move this up a bit so we're going to train the model using the random forest algorithm so from sklearn dots ensemble import random forest regressor so a pointy note here is that this particular tutorial video makes use of regressor because we're building regression models and it is because the y variable which is called log s let me show you log s right here it's a quantitative value so if the y variable is quantitative we're going to build a regression model whereas if it is categorical then we're going to build a classification model okay so in this tutorial the log f is quantitative therefore we built the regression model because random forest here has two versions random force regressor and random first classifier and here we're using the regressor so we're going to create a rf variable to house the random forest algorithm and we're going to specify some of the parameters for the model here maximum depth of 2 and the random state of what about 100 because in the prior random state we used 100 and now we're going to train the model so we're going to type in rf fits and then we're going to use xtrain and y train and then we run it to train the model and the model is trained we're now going to make the prediction in here so actually we could just copy the code cell above here scroll down and we're going to change this to be rf rf lr to be rf part f and now it looks correct to me and we're going to run it okay and now we're going to do the model performance evaluation i'm going to copy the code here paste it we're going to use the mean squared error and we're going to use the r2 score and here instead of lr we're going to replace that to be rf okay so replace all of the lr to be rf and be mindful maybe you might type in wrong like me um just a moment ago to be fr so our f here will be r f now r here will be r f r f and r did i say r f just a moment ago i meant to say lr and now it's rfk let's run it and let's copy the code here which we use to make the table and we're going to change this to brf again and this will be random for rest rf and here r f f show the table okay and now we have two tables we have the linear regression table and we have the random first table so why don't we combine the two tables together okay let me create another level see what level is this random for us with the two hash so one two model comparison and now we're gonna compare it so we're gonna combine the two results table into one and let me see df models equal pd.concats and then i'm going to specify the name of lr results and rf results see do i have x equals to zero because i want to combine it in a worldwide manner let me try if it works all right it worked yeah so x is equal to zero if you want to combine in a row-wise manner whereas if you use axis one it will be in a column-wise manner so here we're stacking them on top of one another okay so you can see now that the two are in the same table but then the index number is a bit off so we need to reindex that so let me see if it's as simple as doing this index okay but it also added a new column here we just say draw it true oh and now it worked we could have also added this at the back of here one again and the number is correct okay but i'm just going to separate it so that it looks a bit more tidy and you could see it but you know how to make it into a wine liner you could just copy here and paste it at the end here so here you can see that we have already compared linear regression model and the random force model let's have a look at the scikit-learn okay and if you click here regression and so here you could find other regression model that you like and you could use it to build your own in the colette notebook here and then you could then add the resulting performance into the data frame here to make your comparison and so now we're going to perform data visualization to take the predicted value and the actual value and make a scatter plot let's do it let's say data visualization of prediction results and we're going to make use of the matplotlib library so we're going to import matplotlib dot pi plot as plt let's say plt scatter we're going to assign to the x axis y train and to the y axis be l of train thread and let's make plots okay this is our first attempt let's make it a bit lighter you could adjust the darkness of the samples that you see here that are represented by the circles using the alpha option we're making alpha to be 0.3 so that regions that are highly overlapping will be a bit darker whereas those that are not overlapping will be lighter color and you're going to see that the x and y axis is not yet labeled we're going to do that plt y label predicted log f f c f label experimental s okay now we have the labels here why don't we make it have plain width and height make fixed size to be five and five okay why don't we make the dots here to be another color the color option and we're going to use color green and what about a trend line let's add a trendline for that we're going to use numpy get the fidget line creating a z variable np dot polyfit and then the y train and then lr train brad p equals z then we're going to color to it 60. okay there we go so we added this red line as trendline that are fitted with the data here so congratulations you built your first machine learning model in python using the scikit-learn library so you can see how easy it is now to build models in python particularly for your tabular data sets and so please feel free to build more models and you could tweak the learning parameters and as i have shown you this api documentation from scikit-learn you could go through the documentation you could click on an algorithm that you're interested in read about it and look at some of the parameters that it allows you to adjust so give it some try let me know in the comments down below what models that you are building and have fun thank you for watching until the end of the video if you reach this far drop a snake emoji so that i know that you're the real one and while you're at it please smash the like button subscribe if you haven't already make sure to turn on notifications to be notified of the next video and as always the best way to learn data science is to do data science and please enjoy the journey