Build A Machine Learning Web App From Scratch

hi i'm patrick from the python engineer channel and in today's video we learn how to build a machine learning web application from scratch we are going to use real world data from the stack overflow developer survey and in the first part of this video you will learn how we analyze and clean this data and then we build a machine learning model and train it with the scikit-learn library then later in the second part of this video we built a web application in python using the streamlit library this library will make it extremely easy and is really beginner friendly so you can follow along without much experience so now before we jump to the code let me show you the final application so here we are in my browser and i have the final app running and we're going to build a software developer salary prediction app so we're going to predict the salary based on a few information that the user has to put in for example the first thing we have to put in is the country and here we get a nice selection box so we can select between different options then we have to choose the education level so here i can say for example bachelor's degree and then we get a nice little slider to select the years of experience and then we click on calculate salary and get the estimated salary so in this case it's 97 000 so let's increase the years of experience and click again and then we see we get a new higher value so this is what we should expect if we increase the years of experience so for this number in the background here we have our machine learning model running that has been trained on real world data and is now predicting this number based on these three information we give the model so in the first part of this video you will learn how to train and then save and reuse this model and then next to this prediction app we can also switch to a second site so here on the left we have a sidebar and we can choose the explore page and here we can explore the data that we used so in this case we used the real world data which is the stack overflow developer survey 2020 so you can find this at this website we can put the link in the description and you can read through this if you're interested in this and you can also download this data set so this is publicly available so again we can put the link in the description and in this video we used the latest data from 2020 but you can also download the data from the past years so yeah on this website you should download this csv file and put into your project folder so let's get back to our web app and here i simply plot some information to get a nice little overview about this data and also learn how we do plotting in this web app so the first thing i plot is a little pie chart to plot the number of data from different countries so we can see the biggest chunk is from the united states and then we also have a lot of data from the uk germany and india and canada and then it's getting smaller and smaller then if we scroll down we get the mean salary based on different countries so this might be a little bit small for you to see but the biggest average salary is in the united states which we can see here and we can also zoom in and zoom out in this plot so this is really nice so this is a bar plot and then down here as last example i show you how to use a line chart so here we plot the mean salary based on the number of years of experience so this is starting at 0 years and goes up to 50 years and for example we can see that at the beginning the salary is increasing almost linearly with the years of experience but then at some point it's stagnating and it's also jumping around so from this plot for example we learned that we cannot use the gears as only information for our model so we have to combine multiple data points and we will learn how to do this later so yeah this is the final web app that we're going to build and this is also fully responsive so we can make this smaller and then it still looks nice so to build this web app we are going to use streamlit this is a python library that makes it extremely easy to build beautiful web apps especially for machine learning and data science tasks so we will learn how to do this in the second part of this video and i promise you that this library is really beginner friendly so it takes care of all the back and stuff for us and we just have to write a little simple python script so yeah let's do this and now as first thing i want you to download this developer service so the 2020 csv file and put it in your project folder all right so here i have this public survey from 2020 open and this is a really large file so it's almost 100 megabyte so you will see that there is a lot of information in here for example if we scroll down all the way to the end then we see we have almost 65 000 data points so from 65 000 different developers and then we can also scroll to the right so you see that there are a lot of different columns in here so a lot of different questions that they answered and also for a lot of data points we have n a in here so not available so we can't use all the information so yeah in the first part now you will learn how we clean this data and then use only the information we want so let's start with the project setup and here i'm in a new directory so i call this ml app and the first thing i recommend is to create a virtual environment so you can either use a python virtual and and i can display the commands for this here but in this case i want to use a conda and so we create a new conda environment and by the way if you don't know how to use anaconda then you can find a tutorial about this on my page so we say conda create and then minus n and then we give it a name so in this case i want to call this ml for machine learning and then i also tell it the python version so here i want to use the latest one python 3.9 and then hit enter and yes so now it created our environment and now we want to activate it so we say conda activate ml and now this is activated so in my terminal we can see this because we see this on the right side and now we want to install the packages we need so for this you could either say pip install and then the libraries i tell you and this is actually the recommended way for streamlet but in my case i'm on a m1 mac and i had some issues with with pip installs so i'm just using conda install this should be just fine and the first thing we want to say is conda or pip install streamlit and hit enter all right so this installed streamlit and all the dependencies for example it also already installed numpy and panda so we can verify this if we say conda install numpy pandas and hit enter then it should be there already so we see all requested packages already installed but we need some more so we want to say conda install and then we also want matplotlib for plotting and we also want scikit learn for our machine learning model so again if you use pip then this is just fine but here i use conda and let's hit enter all right so now we have everything we need and now we want to create a two-parter notebook and there we play around with the data and train our model so we also want to install a kernel for this virtual environment so for this we use this command i python kernel install minus minus user minus minus name equals and then the name of this environment so this is ml in this case all right so now we can start our two-parter notebook server by saying two-parter notebook and hit enter all right so here we see the folder and you can see that i already have the csv file with the survey data in this folder and i also already prepared a notebook but in your case you want to click on new and then select the environment so in this case ml and this is why we just used this kernel install command so that we can see this here so click on new notebook and then you can start so in my case i already prepared the notebook and now i will walk you through it so if this is too fast and you can find this notebook available on github we can again put the link in the description so we want to import pandas and matplotlib and then we want to read this csv file by saying read pandas.readcsv so let's execute the cell and then let's display the first five rows by saying dataframe.head so here we can see the first five rows so as i said this data set is very large so we have 61 columns and also a lot of missing data so nan all right so let's start cleaning the data frame and we only want to keep a few columns so we want to keep the column country the add level which is the education level the years code pro which is the number of years of professional experience then the employment status for example if the developer is working full time and then the converted comp column so this is the salary converted to u.s dollars and this is a yearly salary so this is what we want to keep and then we rename this converted comp to salary so we can do this by saying dataframe.rename so let's run this and have a look at the head again and then we can see we only now have the country the education level years code pro employment and salary so here obviously you can also keep more data so the more columns or the more information we keep the better we should be able to train our model later but here we also want to do a trade-off because we don't want to bother the user with too much information that he or she has to fill out so yeah in this case i'm only using these these data points to predict the salary then so here we can see we already have a lot of missing information where the user didn't put in the salary so for our training data now we want only data where the salary is available so we want to drop all the columns where this is nan and we can do this by saying data frame equals data frame where the data frame salary is not null so if we execute this and have a look then we see the first rows already have a salary so now we have a salary for all the data points so let's have a quick look at the information by saying data frame dot info so here we see we have almost 35 entries and we have these columns and we also see the data type so i think object means this is a string now the next thing we want to do is we also want to drop all the rows where one of those columns is and not a number so we can do this by saying data frame equals data frame dot drop n a and then if we inspect data frame is null dot sum then we should get a zero for each column so now we only use data points where we have data so of course just getting rid of all this data is a radical approach of cleaning the data there are other approaches you can use for example you can fill in missing data with the mean of this column but in this case we still have enough data points available so it should be fine to just drop them then the next thing we want to do is we only want to keep the data points where the user was employed full time so we can do this by saying data frame equals where data frame employment equals equals employed full time so we only keep these rows and then we want to drop the column employment so we don't need this for the prediction so let's run this and again print the info so now we see we only have these columns and drop the employment the next thing we want to do is clean the country data so if we say data frame country dot counts and hit enter then we see how many data points we have from each country so we have the most data from the united states and we also see that a lot of information that for a lot of countries we only have one data point so now we want to get rid of all the smaller countries and in no way do i want to offend these countries so i'm sure there are great developers there but the thing is just if we use for example only one data point from monaco then this might actually confuse our model because it can't really learn from this one data point so for this now we want to get rid of the countries where we have only a few values now in order to clean the countries we have this function here shorten categories which gets the categories and this must be these value counts here and then a cut off value where we can play around with this and if the number of samples we have here is greater or equal than the cut of value and we keep this and otherwise we combine this into a new category that we call other so the other countries and then we execute this function so we say shorten categories and then we give it dataframe.country.value counts and the cut of 400 in this case and then we change our data frame by saying dataframe country equals dataframe country dot map and give it this country map and now if we have a look at the value counts again all right so now here we can see we only kept these countries and united states is still the biggest one and now here we have the new other category and here this is the biggest chunk now the next thing we want to do is let's inspect the salary range and for this i want to plot the salary against the country so let's execute this and here this is a box plot so here you see the box plot and we see the different countries on the x-axis and then we have the salary ranging from zero all the way to two million so the way a box plot can be interpreted is that this small area here that you see this is the median value and now all these points are actually outliers so there are a lot of outliers in this data frame and now actually the most data points are in this median area so what we want to do here now is we want to keep only the data where we have the most information so we can see that none of these boxes here reaches over this line which is 250 000 so we define this as our cut off value and keep only salaries where this is smaller than this range so we can do this with this command and we also only want to keep salaries where this is greater than 10 000 so i think if you're working full time then you should earn more than ten thousand dollars in a year so let's keep only this and then we also want to drop the other categories so we can do this with this command and now let's plot the same again so the same box plot here and hit enter so now we can better detect these boxes here and see where the median range is and yeah so now we still have outliers but not so many anymore then the next thing we want to clean is the years code pro so if we inspect all the different values we get then we see we have this as a string value and we also have less than one year and more than 50 years so what i do here is i want to convert this to a float and if it's less than one year then i just say this is 0.5 and here we do a cut off so for more than 50 we just say 50. so for this we create a function clean experience which gets x and here we say if x is more than 50 years so exactly this string then we return 50 if it's less than one year we return 0.5 and otherwise we convert it to a float and then we transform it by saying data frame years code equals data frame use code dot apply this function so let's hit enter and now we want to do a similar thing with the education level so if we have a look at all these unique values then we see we have all these different answers and here again we want to combine a few categories so i want to keep the bachelor's the master's degree then if it's a professional or a doctoral degree then i will say this is a post-grad and for all the other things that we say this is less than a batch class so for this again we create a function define clean education where we check if and then if this string is in x so if bachelor's degree then we return this is the bachelor's degree then we have the master's degree the postgrad and the less than bachelors and then again we apply this to the education level column so now hit enter and then let's inspect the unique values again and now we see we only kept these values all right so now we're almost done with cleaning the data but we still have one problem and this is because some of our data for example the number of years of experience this is a float but the education level and the country is a string and our model can't really understand the string so what we have to do is we have to transform the string values to a unique value a number that our model can understand for example for these four strings we can use the number zero one two and three and then in order to transform this data we can use the label encoder from sklearn.preprocessing so we import this and then create a label encoder education and now we fix this to the data and also transform it in the same step and then we assign it back to the education level column so now if we have a look at the unique values again then we see we no longer have strings in here but now we have the numbers zero two one and three so the order is not really important here but this label encoder here now knows which number belongs to which string and then later can convert this back again so yeah we did this for the education level and now we do the same for the country so we map each country to a unique number and for this we create a separate label encoder that we call le country and then we apply this fit transform again and now let's have a look at the unique values for the country and then again we see we have numbers here and no longer a string all right so now we have our final data that we can use and now we can finally start training our model so for this we have to split our data in the features x and the label so the label is the salary this is what we predict and the features are the rest of the columns so the country the education level and the years of experience so we get these columns by saying data frame drop the salary and the y is only the salary column so now we have x and y and now we want to try out different machine learning models and for this we use the scikit-learn library and in this case we deal with a regression problem and this means that here we don't predict a defined category like cat or dog but here we predict a number and the number can be anything from i don't know 0 to 1 million so that's why this is a regression and not a classification and for this the first obvious choice is the linear regression model so we can import this and then we create a regression model like so and then like for all the models in the sk learn library first we have to fit the data so we can fit x and y dot values and run this and then to predict new values we say our model dot predict x and here we get the predicted values so let's do this as well and one node here so normally you want to split your data into training and testing or even training validation and testing data and then do the training or the fitting only on the training data but here for simplicity we do this on the whole data so now we get the predictions back and now in order to um evaluate how good our model performs so for the regression plot problem we usually have a look at the mean squared error or you can also have a look at the mean absolute error so you can import this and now we want to calculate the mean squared error based on the actual labels y and the new predicted label so to get the actual number back we calculate the square root of this with numpy square root and this will be our error so let's have a look at the error all right so this is the error 39 000 dollars so this means that our model on average is off by this number and this is still pretty high if we think that the numbers range from 10 000 all the way to 250 000 and each time we predict the salary then on average we are off by this value then it's still pretty high so let's try out a different model so the next model we want to try out is the decision tree regressor so the concept is the very same we import this then we create a instance of this and then we fit the data like so and then we predict the values so let's call model.predict and again let's calculate the mean squared error and then the root of this and print this and now this time the root mean squared error is only 29 000. so it's already a lot better now let's try out one more so let's also try out the random forest regressor so this will basically combine multiple decision trees into a forest so again the same approach we import this we create one we fit the data and then we call model dot predict and calculate the error and have a look at this and this time the error is still 29 000 so it's not really better than the decision tree so what we do here is obviously a little bit cumbersome we do this for every model that we find and then we try out the different values but now what we didn't consider yet is that for each of these models we also can use different parameter values that we can put in for example for a decision tree we can use the max depth parameter and then it might perform very differently for different parameters so in order to find the best model with the best parameters um there is a so-called grid search cv method which stands for grid search with cross validation and the way this works is we import this and then we define the set of different parameters we want to try out for example for the max depth parameter we can try out these different values then we create a dictionary with this and this has to be the same name as the keyword argument for the in this case the decision tree regressor so you can find this in the sklearn documentation and then we create a grid search cv with this regressor and the parameters and in this case the scoring is the negated mean squared error so this is just what we want to use for a regression problem with the grid search and then again we call the fit method with x and y and run this and now what is happening here so this will basically loop over all the different parameters that we put in here and try out all the different parameters and each time evaluates the error and then chooses the best model for us so now this has been executed and now we get the best estimator by saying grid search dot best estimator underscore so now let's use this best estimator here is our regressor and now again let's train it on the whole training data and then call calculate the error and see what we get so this is the final error we get for our final regressor and this is actually higher again than this one we used here and this might be a little bit surprising because actually we search for the best one here but what could have happened here is that in the decision trees there is some kind of randomness in it even though we said random state equals zero so it might be slightly worse again but that's fine so we still keep this regressor now and continue with this so now let's learn how to apply this to new data like we have to do in the app later so let's have a quick look at the data frame again and we see we have these columns country education level and years code pro so let's say in the app later we get a new array and it will have these values so here we start with the strings that the user puts in so the country the degree and the experience so now our x looks like this and now what we do here is we want to apply the label encoder for the country so the country is the first or the zero column then we want to apply the label encoder for the education and then we also want to have this as a float array so let's apply this and then we see now we get these values and we can put this in our regressor and then predict this so yeah with these information we get this predicted salary and now this is what we have to do in our app but before we can use this in our app first we have to of course save our model and for this we use pickles so we can say import pickle and then we want to save the regressor the label encoder for the country and the label encoder for the education so for this we create a dictionary then we open a pickle file and here we want to open this in right binary mode so this is important for pickle here and then we say pickle dump the data so this here in this file so now if we execute this then this should save this data in our directory and we can check this by loading it again in read binary mode and then say data equals pickle dot load and then we can access the model the label encoder for the country and for the education by accessing the key so let's get them back from the data and run this and then let's run the prediction with our loaded model again and then we see we get the exact same number that we get here from the original model so saving and loading works all right so now we are done with our two-parter notebook so now we have the complete pipeline here so we learned how we can clean the data and then create our model and train it and save it but now we have to convert some of the code from here to our app so let's do this so now let's create our streamlet app and for this let's open your editor of choice and you can see that inside this directory now we have the csv file and we also have the saved pickle file and we have the notebook so yeah this savings step here worked so now let's create three new files so let's create one app.pi file then let's create a separate file for the predict page so we call this predict underscore page dot pi and let's create another one and we want to call this explore underscore page dot pi so let's start with the predict page and first of course we want to import these things we need so we import streamlet sst then we also need pickle to load the data so import pickle and we need numpy so we import numpy snp and now we need some code from our notebook so the first thing is we want to load the model so let's create a function define load model and here let's go back to our notebook here we want to copy this part of the code so we open our pickup file and say this is our data and then here we return the data then of course we want to execute this so we say data equals load model and now we want to access the different keys like we did here so we create the regressor the label encoder for the country and for the education so let's simply call this regressor so now let's create the prediction page and i again wrap this into a function and later you will see why i do this so let's create a function and let's call this show predict underscore page and here now we will build our streamlet application and the way this works is we create different widgets for example we can say streamlets.title and then here we can say software developer salary prediction so now this will display a title and for example we can also simply write text by saying streamlet dot write and then here we can use a multi-line string and we can use a markdown syntax here for example if we use three hashtags then this will be a h3 tag so let's say we need some information to predict the salary so now let's save this and now in the terminal so here again i want to say conda activate ml and now in order to run this file we say streamlit and then run and now for the first example yes let's use this predict page.pi so now if we execute this then this will start the server so now it will open up the app and as you can see we don't see anything and this is because we only defined the page but we didn't execute this so now in order to execute this actually i want to run this from the main app.pi file so here again we say import streamlet and then we import this show predict page so we say from predict page we import this function show predict page and then let's execute this here and now let's quit the server again by hitting control c and then let's say streamlet run and then we actually want to run the app.pi file and now if we go to this address again then now we see we have our app up and running so now this is working so now the next thing here is what we want to add is we want to add two select boxes for the countries and the educations so let me copy and paste this in here and this here is a tuple with all the different choices so all these countries and you can find them in your notebook for example at this position where we printed the value counts so all these different countries that we keep so we don't keep the other category so these are all the different choices for the countries and then the same with the education level so these are the exact names of the education categories that we keep and now in order to create a select box we can say country equals streamlets dot c lectbox and now this must get the number of choices so here the countries which can be a tuple or a list so here we use a tuple and we can give this a name as first argument so we can say country and then comma countries and now if we save this and go back to our app we can click on rerun and now here we see we have our select box so this is working so yeah pretty cool and now what happens here is if we choose a different value here for example germany then in our script this will be the value of this variable so now this will our country will be germany and then we can work with this so yeah let's copy and paste this in here and now we also want a select box for the education so let's call this education and then as a string we use education level and here we put in all the different education choices that we defined so education level so now we have this and now we want to select the years of experience so experience equals and for this we use a slider and we get this by saying streamlets.slider and then the name we say years of experience and then we can give this a min value a max value and a default value so let's say this should range from 0 to 50 and the start value is three so now again let's save this and let's go to our app and run this and then we see we have the education level as selection box and we have the slider with the default three so now this is working as well so let's add a button to start the prediction so we can add a button by saying streamlets.button and then the name for this button is let's say let's call this calculate salary and now again we can assign this to a variable so we can say okay equals streamlets.but so this means that if we don't click on the button then this is false and if we click on the button then this is true so now only if this is true if we clicked on the button then what we want to do here is we want to start the prediction so let's go back to our notebook and now what we want to do here is the same as i did here so we created this numpy array like we do here so let's copy and paste this then we transform the country the education wanted as a float like so and then we predict it so let's copy and paste this as well so now here we have the transforms for our numpy array and now here instead of these hard-coded values the nice thing is that we already have the information so the this is just the country variable that we get from this select box this is the education level and this is the experience level so now we have all the information that we need and now we simply need to calculate the predictions so we say salary equals and then we say regressor dot predict and we want to predict our x and now as last thing we want to display this so for this for example we can say streamlets.right or here i show you another thing so we can use the sub header which is kind of a smaller header and here let's use a f string and then let's say the estimated salary is and then this is in dollar and now let's use our variable here so this is the salary and this is a numpy array with only one value in it so we want to access the first value like so and we also want to display it with only two decimal values so we can format it like this so let's save this and let's go back to our app again and rerun this and then we have the button and now if we click this then we should see the results so let's try it out and yeah this is working so here we see we chose germany and we have less than a bachelor's and three years experience and this is our predicted salary now for example let's try this out so for example if we use the united states then this is higher than in germany so now we have the prediction page so let's add a sidebar and then let's also add the second page so for the sidebar what we do here is we go to the app dot py file and again we create a selection box by saying streamlets.selectbox and then here as first argument we give it the title so let's call this explore or predict and then a second argument as a tuple we give it the two choices so we want the predict page and we want the explore page so now let's first save this and have a look at the app again and here we see a error name error name st is not defined and this is by the way very nice in streamlit so we immediately see the error here in a nice way so if we go back obviously i have to import streamlet sst so let's save this again and check our app again and now we have this select box here explore or predict and now in order to create a sidebar the only thing we have to do is call st dot site bar and then dot select box so with this we can move any widget for example also a button or a slider to the sidebar so now if we go back then we should see the sidebar so yeah this is working now we have a nice little sidebar where we can choose between the different sites so now the only thing left to do is implement the explore page so let's implement the explorepage.pi so here again we import streamlet as st then we import pandas as pd then we import matplotlib dot pi plot as plt and now we want to load and clean the data the same way we did it in the notebook so let's go back to our notebook and let's copy all the things we need to clean the data so for this we want to for example want to copy these functions that we applied so we simply copy and paste this in here then let's have a look we also at some point we cleaned the experience like so so let's also copy and paste this in here then let me close this here and now let's also search for the clean education so we copy and paste this from our notebook so now we need a function to load the data so let's call this define load data and now here we load the data frame and then we apply all the transformations that we did in this notebook so here first of course we load the csv file and now let's go back through our notebook and apply all the transformations that we did here so only take only the columns we want then we rename the converted comp to salary so now we have to go through this notebook and copy and paste all the transformations that we apply so i will simplify this for you and just copy and paste all the steps that we used in here so yeah these are the exact same same steps that we used in the notebook and we apply them here and then we return the data frame from this function so now we have our load data function and of course we also want to execute this so we say data frame equals load data and now i want to show you a nice little tip that we can do with streamlit so now whenever we refresh the page then it will load the data again because it will execute this script from the top so in order to avoid loading this again and again and applying all these steps again we can actually cache this step simply by using a decorator and this is st dot cash so now this once we have executed this one time then it will cache it and it is available the next time again so this is a nice little trick that we can do with streamlet to improve the performance of our app so now we come to the actual page that we want to display and again for this we create a function and we call this show explore page and here again we give it a title and then a smaller title by saying streamlets.title and then streamlets.right and now for this little example i want to display three different charts one pie chart one bar plot and one line chart so for the pie chart we want to plot the value counts of the country so we can access the country like so and then we can get the value counts by calling value counts and now we want to put this into a pie chart and for this we use matplotlib and then let's simply copy and paste this for you so we create a subplot and then we can call access.pi with this data and the labels is data dot index and then here are some more arguments to get a nicer look and now in order to actually display this so first let's write a little heading first by saying streamlets.right and then here as a multi-line string in markdown format like so let's use um let's use a h4 tag and then we say number of data from different countries and now we want to actually plot this figure here and we can do this in streamlit by saying streamlets.pie plot and then the figure1 and here we can display any mudplot clip figure that we created so this is pretty nice so now let's save this and now let's actually show this and for this we have to do one change here in the app.pi so we also want to import this function so we say from explore page we import the show explore page and now from our select box we want to save the selection in a variable so we call this page and then we say if page equals equals if this is the predict value then we show the show predict page and otherwise we say sorry this must be here otherwise we say show explore page and now this should work so now if we go back to our app and rerun this and then we get a key error converted comp so let's have a look at where we load the data and here i renamed this to salary but then i access it here and here and here again so let's actually do the rename step at the very end so it doesn't really matter so like here and now let's go back and now this should work so let's rerun this and now let's select the explore page and i think we have one more typo so again let's go back and here we want the country of course so country and save it again and rerun it and now it's working so yeah here we have our pie chart so let me make this site smaller again so let's add two more charts the bar chart and the line chart so first let's again write a little heading so we say streamlit dot right the mean salary based on the country and now the data we want to plot is we get it like so so we want to group the data frame by the country and then access the salary for each country and calculate the mean value and also sorted in ascending order so we get this like so and then in order to plot this we can use the built-in bar chart of streamlit by saying streamlet.bar chart with the data like so and now this is all we need so now if we go back and again rerun this then here we see the bar chart so this is working pretty nicely and now we see we have all the different average or mean salaries for each country and now as last thing let's also use a line chart so for this let's again write some text streamlet.write the mean salary based on the experience and here we do the same thing we call this group by and but here we want to group by and this is called years code pro and then again the salary and we calculate the mean and sort it and now we create a line chart by calling streamlets.linechart with this data and save this and run it again and reload this and now yeah this is working as well here we have our line chart all right so yeah this is our final app and this is all that i wanted to put in in this example obviously you can play around with this and show different data as well so yeah i recommend that you do this and also i recommend that you play around with this model here and maybe try to improve this on your own for example you might try to use more than just the three information that we put in for example you can also try to use the age and use this as well for training but yeah that's it for now and hopefully you enjoyed this little tutorial and now know how you can build a machine learning web app from scratch with real data and then i hope to see you in one of my next videos bye you

Channel: Patrick Loeber

Views: 201,751

Keywords: Python

Id: xl0N7tHiwlw

Length: 53min 20sec (3200 seconds)

Published: Sat May 01 2021