Preparing for a Real-Life Machine Learning ProjectIn the <a href="https://techfastly.com/machine-learning-101-iii-insurance-data-set">previous post</a>, I shared with you my trial with six different machine learning models and being able to measure their accuracy scores.In the preceding Insurance dataset post, I had left out Strings data (text). In this post, we will revisit the scores of the models after including the string data, which will help us in determining the values in these strings that the algorithms would enjoy.As always, we follow the following series of steps:Step 1- Load the Data.Step 2- Check Data Types &amp; Standardize the data types.Step 3- Split the data into Test &amp; Training sets.Step 4- Run multiple algorithms and compare the Accuracy Score.Step 1:&nbsp; Load the DataThis time, we load the large CSV file which has both strings and integers. We have also kept the same logic based on which we arrived at the different labels of 'Excellent/Good/Poor' drivers.As before, we use the print() command to verify if the pd.read_csv works successfully. The screenshot below proves that this is the case.<figure class="kg-card kg-image-card"><img src="https://iaxds62ruoojbg1k4180.cleaver.rocks/images/posts/43771dfbf52e3962411c87ad6b01dcee.png" class="kg-image" alt="" loading="lazy"></figure>We now wish to view the different data types the various 'features' are in.Pandas support a data type method (d.type) that allows us to query the data type in a single line of code. To verify whether we executed the d.type command, we once again rely on the print() command. The output on the screen confirms the data types in our file to be a mix of integers, float, and objects (strings). Algorithms in scikit-learn understand only numeric data. So, we need to convert all object data into numeric data.As this is a common problem faced in machine learning, we will use a pre-programmed command 'Label-Encoder()'&nbsp;method. Using this method converts string data into numerical data.<figure class="kg-card kg-image-card"><img src="https://iaxds62ruoojbg1k4180.cleaver.rocks/images/posts/bb6f67f7ef2f6a073cf0ec4b2db8619f.png" class="kg-image" alt="" loading="lazy"></figure>We implement the 'Label-Encoder' in the following manner:From sklearn import preprocessing#import the necessary module from sklearn import preprocessing# create the Labelencoder object = preprocessing.LabelEncoder()#convert the categorical columns into numericencoded_value = le.fit_transform(["basic", "premium", "extended"])print(encoded_value)Please note, this method assigns the numeric values to the classes. The classes are assigned with values in the order of their first letter, as in the original list.The output on the bottom right of the screen tells us that 'basic' transformed into 0, 'extended' transformed into 1, and 'premium' transformed into 2.<figure class="kg-card kg-image-card"><img src="https://iaxds62ruoojbg1k4180.cleaver.rocks/images/posts/1ce83bbea886ace820c15745c4279f27.png" class="kg-image" alt="" loading="lazy"></figure>Now, to transform all the different columns and all the unique entries, we use the following code:print("'State Code' : ",file['State Code'].unique()) print("'Education : ",file['Education'].unique()) print("'Gender : ",file['Gender'].unique()) print("'Location Code : ",file['Location Code'].unique()) print("'Marital Status : ",file['Marital Status'].unique()) print("'Sales Channel' : ",file['Sales Channel'].unique()) print("'Vehicle Class : ",file['Vehicle Class'].unique()) print("'Vehicle Size' : ",file['Vehicle Size'].unique())And upon running the program, we can view the unique classes for each column in the screenshot below:<figure class="kg-card kg-image-card"><img src="https://iaxds62ruoojbg1k4180.cleaver.rocks/images/posts/7ba62cd2099e700cd8ec0628c2f68e9c.png" class="kg-image" alt="" loading="lazy"></figure>We now use the encoded_value = le.fit_transform(["basic", "premium", "extended"]) command, and convert all the strings into numeric values. Upon running the program, we observe that there are no errors:<figure class="kg-card kg-image-card"><img src="https://iaxds62ruoojbg1k4180.cleaver.rocks/images/posts/f8fc82d704a032f6e4dda64aedf1ae2e.png" class="kg-image" alt="" loading="lazy"></figure>To verify if the data conversion has taken place appropriately, we use the head() function to examine the first few lines of our data set. To replicate all the remaining columns of string, we use the following code:file['State Code'] = le.fit_transform(file['State Code'])file['Coverage'] = le.fit_transform(file['Coverage'])file['Education'] = le.fit_transform(file['Education'])file['Gender'] = le.fit_transform(file['Gender'])file['Location Code'] = le.fit_transform(file['Location Code'])file['Marital Status'] = le.fit_transform(file['Marital Status'])file['Sales Channel'] = le.fit_transform(file['Sales Channel'])file['Vehicle Class'] = le.fit_transform(file['Vehicle Class'])file['Vehicle Size'] = le.fit_transform(file['Vehicle Size'])Upon printing the file (print(file.head())), we can see that all strings got converted to numeric outputs.<figure class="kg-card kg-image-card"><img src="https://iaxds62ruoojbg1k4180.cleaver.rocks/images/posts/ba4ec3b9a5e7c1171a583884e617b2e8.png" class="kg-image" alt="" loading="lazy"></figure>Step 2: Train the DataLike we did in the previous exercise, we load the file into an array.Once the data loads into an array, the data can split into 'Training' data and 'Testing' data. As we did last time, we will train the data on 80% of the data set and test it on the remaining 20%.To do so, we use the following commands:array = file.valuesX = array[:,0:17]Y = array[:,17]Upon running the program, there are no errors reported, as shown in the screenshot below:<figure class="kg-card kg-image-card"><img src="https://iaxds62ruoojbg1k4180.cleaver.rocks/images/posts/d1cf69ec6806b7b4a56685f342c46f65.png" class="kg-image" alt="" loading="lazy"></figure>To confirm the correct selection of X (features) and Y (labels), we use the print command:print(X)<figure class="kg-card kg-image-card"><img src="https://iaxds62ruoojbg1k4180.cleaver.rocks/images/posts/b61ec440f454e2c859632e04890962b9.png" class="kg-image" alt="" loading="lazy"></figure>print (Y)<figure class="kg-card kg-image-card"><img src="https://iaxds62ruoojbg1k4180.cleaver.rocks/images/posts/f92c6e3dc69b555adf2d071f0a594700.png" class="kg-image" alt="" loading="lazy"></figure>We re-use the commands that we used the last time to split the data to 'Train' and 'Test'. We import the module from the sklearn library. We use a 'random state' in our data splitting to ensure that the different algorithms we use receive the same amount of data to avoid any bias. from sklearn.model_selection import train_test_split X_train, X_validation, Y_train, Y_validation = train_test_split(X, Y, test_size=0.80, random_state=1) Running the library does not result in any errors, as evidenced in the screenshot.<figure class="kg-card kg-image-card"><img src="https://iaxds62ruoojbg1k4180.cleaver.rocks/images/posts/915b493fcecaf16a0215e5548b31d991.png" class="kg-image" alt="" loading="lazy"></figure>At this point, our data has successfully split into Training and Testing data.Step 3: Run Multiple ModelsAgain, as we did in the last post, we use the following six models for testing and evaluation:Linear Models:o Logistic Regression (LR): The model looks for a pass/fail relationship between variables.o Linear Discriminant Analysis (LDA): Looks for a linear relationship between two or more classes of objects.Non-Linear Modelso K-Nearest Neighbors (KNN): A pattern recognition used for both regression and classificationo Classification and Regression Trees (CART): Use of ?decision trees? to go from observations to conclusionso Gaussian Naive Bayes (NB) : Calculation of pre and post probabilities approacho Support Vector Machines (SVM): Supports labeling of unlabeled data in an unsupervised environmentTo run these six models concurrently, we can treat these models as a list of models to be used using [].Add the following to your code:from sklearn.linear_model import LogisticRegressionfrom sklearn.tree import DecisionTreeClassifierfrom sklearn.neighbors import KNeighborsClassifierfrom sklearn.discriminant_analysis import LinearDiscriminantAnalysisfrom sklearn.naive_bayes import GaussianNBfrom sklearn.svm import SVC#importing the modelsmodels = []models.append(('LR', LogisticRegression(solver='liblinear', multi_class='ovr')))models.append(('LDA', LinearDiscriminantAnalysis()))models.append(('KNN', KNeighborsClassifier()))models.append(('CART', DecisionTreeClassifier()))models.append(('NB', GaussianNB()))models.append(('SVM', SVC(gamma='auto')))#adding the modelsTo append these models to a list named models, we must import the models from sklearn. We once again run our code and find no errors. Hence our code should look like as follows:<figure class="kg-card kg-image-card"><img src="https://iaxds62ruoojbg1k4180.cleaver.rocks/images/posts/e83ca11cdb012e7afefd754dced241e4.png" class="kg-image" alt="" loading="lazy"></figure>Step 4 - Check Accuracy Score for each ModelTo check the accuracy of each of the six models, we create a list titled results and names.results = []names = []for name, model in models:kfold = StratifiedKFold(n_splits=10, random_state=1, shuffle=True)cv_results = cross_val_score(model, X_train, Y_train, cv=kfold, scoring='accuracy')results.append(cv_results)names.append(name)print('%s: %f (%f)' % (name, cv_results.mean(), cv_results.std()))We will also have to import:from sklearn.model_selection import cross_val_score (to Evaluate a score by cross-validation) andfrom sklearn.model_selection import StratifiedKFold (Provides train/test indices to split data in train/test sets)Upon running the program, we can see the results for the different models we used in the screenshot below:<figure class="kg-card kg-image-card"><img src="https://iaxds62ruoojbg1k4180.cleaver.rocks/images/posts/ab0c38179a14bd500c6a46c2bcfad4c2.png" class="kg-image" alt="" loading="lazy"></figure>If we compare the results of the previous exercise and this, we will find that with an increase in data ingestion with the algorithm, the accuracy of predictions has increased.<figure class="kg-card kg-image-card"><img src="https://iaxds62ruoojbg1k4180.cleaver.rocks/images/posts/ec67ee6774db4c5160be7c355b4e7991.png" class="kg-image" alt="" loading="lazy"></figure>The above table does provide evidence that data points contained in strings when converted into numeric values have assisted in increasing the accuracy of the algorithms.

Harmit S Kamboe

Machine Learning 101 Continued IV- Insurance Data set - Final Post

<h3 id="writing-an-algorithm-with-just-six-lines-of-code-inspired-by-a-google-developers-youtube-video-from-2016">Writing an Algorithm with just six lines of Code- Inspired by a Google Developers' <a href="https://www.youtube.com/watch?v=cKxRvEZd3Mw">YouTube</a> Video From 2016</h3>This post is part-1 of a series of multiple posts that will help us get somewhat familiar with Machine Learning.In our work lives, many of our projects touch Machine Learning directly or indirectly. Apart from a few folks on the team, seldom do others get a chance to work on an algorithm. The very word, algorithm, brings up images of pages and pages of complicated code. The reality, however, is much different.The kind of a Machine Learning model we will code today will be a Supervised Learning Machine Algorithm. What this means is that we will feed some sample data (referred to as 'training data') to our 6 line algorithm and get the desired prediction.As an example, if we were writing an algorithm that was going to predict if an email is a spam or not based on some rules we provide, the entire process would look like the image below.<figure class="kg-card kg-image-card"><img src="https://iaxds62ruoojbg1k4180.cleaver.rocks/images/posts/01d6bfbb18d5f1db9b981927afce024e.png" class="kg-image" alt loading="lazy"></figure>The different emails coming in would be the input. The classifier or our algorithm would then classify emails as spam or not spam. The resulting segregation would be the output of the program.In this transcription of a Google Video, I will walk you through all the steps I took to set up the right development environment and then understanding the six lines of the actual algorithm. The use case I have taken here is that of an algorithm that will determine if a new car insurance applicant is likely to be a Great/Average/Bad driver. From a business standpoint, of course, based on the prediction of how good a driver is, a different price can be provided to different applicants.However, at the end of the exercise, you can save the algorithm and reuse it for any other use cases.The steps I am describing below have been broken further into major categories:Step 1 ? Setting up the Development/Coding EnvironmentStep 2 ? Writing the AlgorithmStep 3 ? Running the AlgorithmLet's get started.Step 1 ? Setting up the Development EnvironmentWe need to install two things:A. A programming language to write the Machine Learning Code in andB. A code editor that helps us detect errors easily as we write this codeYou can achieve both objectives by visiting <a href="https://www.anaconda.com/">https://www.anaconda.com/</a>. Anaconda is an open-source data science platform that provides access to Python and R, two programming languages, and their editors in a convenient package.Steps to download Anaconda and getting ready to write code in the Code Editor:1. Go to the individual version of Anaconda page - <a href="https://www.anaconda.com/products/individual">https://www.anaconda.com/products/individual</a>2. Choose the right version of Anaconda to download.<figure class="kg-card kg-image-card"><img src="https://iaxds62ruoojbg1k4180.cleaver.rocks/images/posts/5d4a47b27113b94353e77fd8c8b3e97c.png" class="kg-image" alt loading="lazy"></figure>3. Suppose you are on a Windows machine, you can check whether you are on a 32 bit or 64-bit machine by typing in 'system information' in the Windows search bar at the bottom left.<figure class="kg-card kg-image-card"><img src="https://iaxds62ruoojbg1k4180.cleaver.rocks/images/posts/0cd0f1eb94df60b3297b64a90d652760.png" class="kg-image" alt loading="lazy"></figure>4. Double click on the Anaconda software once you have downloaded it. Choose all the 'recommended options' as you install Anaconda.<figure class="kg-card kg-image-card"><img src="https://iaxds62ruoojbg1k4180.cleaver.rocks/images/posts/b57c48cebefdf7f602ab3d0b2f799157.png" class="kg-image" alt loading="lazy"></figure>5. Access the Anaconda Navigator by using the windows search bar.<figure class="kg-card kg-image-card"><img src="https://iaxds62ruoojbg1k4180.cleaver.rocks/images/posts/61cfa28ab83959ca99164559dc155438.png" class="kg-image" alt loading="lazy"></figure>6. Click to launch Spyder- the code editor that you will be using to write the code<figure class="kg-card kg-image-card"><img src="https://iaxds62ruoojbg1k4180.cleaver.rocks/images/posts/349e3b289c006f8e173ec35e6e991765.png" class="kg-image" alt loading="lazy"></figure>7. At this point, this is what your screen will look like<figure class="kg-card kg-image-card"><img src="https://iaxds62ruoojbg1k4180.cleaver.rocks/images/posts/31a50301a5bdcf2efaad640110cfc197.png" class="kg-image" alt loading="lazy"></figure>8. The Code EditorAs you write code in the left-hand panel, Spyder will let you know if you have entered any wrong syntax. You are now ready to write code for your algorithm.Step 2Writing the AlgorithmThere are three key steps at a macro level to write the algorithm that we need to undertake:<figure class="kg-card kg-image-card"><img src="https://iaxds62ruoojbg1k4180.cleaver.rocks/images/posts/5fbae98c585271be3239989e90b49daf.png" class="kg-image" alt loading="lazy"></figure>The Training data collection exercise is where the algorithm gets the data from which it can learn. The Training Classifier ingests this Training Data and detects patterns in the data. At this point, the algorithm is ready to make predictions on any input data that it may get.As a general rule, an algorithm usually gets 80% of all data. We use the remaining 20% of the data to test the algorithm. But more on this will be discussed much later in this series of posts.Line 1At this point, we need to import scikit. Scikit-learn is a free software machine learning library for the Python programming language. It features various classification, regression, and clustering algorithms.In-Line 1, all you have to write is 'import sklearn'. Don't worry about the 'alert' that the code editor displays in orange. Any time you wish to run your program, click on the green arrow circled below.Training Data<figure class="kg-card kg-image-card"><img src="https://iaxds62ruoojbg1k4180.cleaver.rocks/images/posts/8f4d0bdfab64f933d1b549dbd00c29ef.png" class="kg-image" alt loading="lazy"></figure>To train the algorithm, we need to provide it with some relevant data. Here is an example of what this input data may consist of in our insurance example:<figure class="kg-card kg-image-card"><img src="https://iaxds62ruoojbg1k4180.cleaver.rocks/images/posts/37722296c86a7b4d099b3ff5211f8a13.png" class="kg-image" alt loading="lazy"></figure>For this small example, we will stick with the four input variables of Zip Code, Age, Gender, and Car Type. On a live project, of course, there may be many other variables that can be relevant and required.In terms of the terminology used in the field of Machine Learning, the four input variables of Zip Code, Age, Gender, and Car Type are called 'Features'. The desired output of a Good/Average/Bad driver is called a 'Label.Let's take a very small sample of 3 records:Instead of filling in actual Zip codes, Age, Gender, and Car Type with actual values, using numeric values make it easy to write the algorithm. If that is not the case, we would have to deal with integers versus strings, etc. So let's avoid that by using numeric codes for each feature field.<figure class="kg-card kg-image-card"><img src="https://iaxds62ruoojbg1k4180.cleaver.rocks/images/posts/ca0e93b6d70909096b739e940b7e7c70.png" class="kg-image" alt loading="lazy"></figure>In our simple example, we will input the numeric codes above as a representative value for each feature.Step 1 ? As we are getting ready to enter now important data, it is a good time to save this file with a real name. Use the file's 'Save As' command and save it as 'car-insurance'.<figure class="kg-card kg-image-card"><img src="https://iaxds62ruoojbg1k4180.cleaver.rocks/images/posts/5b307c59adae93e289be9c3ba608474e.png" class="kg-image" alt loading="lazy"></figure>Please make sure that you change the file type to be as 'Python' when you are saving the file.Line 2Add in the features as line 2 into the algorithm.features = [[1, 2, 1, 2],[1, 5, 1, 4],[3, 3, 2, 6]]Your screen should now look like the following screenshot:<figure class="kg-card kg-image-card"><img src="https://iaxds62ruoojbg1k4180.cleaver.rocks/images/posts/b952848dca112257e062495803046405.png" class="kg-image" alt loading="lazy"></figure>Line 3Add in-line three, which is the 'labels' for the algorithm to learn.labels = [0,1,2]Based on the code above, your excel file/Google Sheet of data should look like as follows:<figure class="kg-card kg-image-card"><img src="https://iaxds62ruoojbg1k4180.cleaver.rocks/images/posts/46a739e968be2ad37089da456800f968.png" class="kg-image" alt loading="lazy"></figure>The program now has the input data that it needs. We now need to train the algorithm.Line 4At this point, we need to import the Decision Tree module from scikit for data to be analyzed using Decision Trees.A single line of code can easily achieve this-clf = tree.DecisionTreeClassifier()The word 'clf' refers to the 'classifier'. Your screen should now look like this:<figure class="kg-card kg-image-card"><img src="https://iaxds62ruoojbg1k4180.cleaver.rocks/images/posts/64b0f9e2ef0314cb683c720754edfb73.png" class="kg-image" alt loading="lazy"></figure>The red X on line 4 suggests an error. If you hover your mouse over the red X you see the code editor being helpful and providing you the reason for the error:<figure class="kg-card kg-image-card"><img src="https://iaxds62ruoojbg1k4180.cleaver.rocks/images/posts/55a469713854896d7aba7c4cddf7359f.png" class="kg-image" alt loading="lazy"></figure>The error relates to 'tree' not being a defined element. Line 1 needs to be revised from the sklearn to 'import tree' to resolve the error. As you can see from the screen below, the error has now gone away.<figure class="kg-card kg-image-card"><img src="https://iaxds62ruoojbg1k4180.cleaver.rocks/images/posts/5784e8e08a10ea153440c66f67b1ff21.png" class="kg-image" alt loading="lazy"></figure>Line 5At this point, we are ready to trigger the Training Classifier to ingest the rules and detect the patterns from the sample data. To do so, we need to add the following line of code-clf = clf.fit(features, labels)The 'fit' here refers to the command which instructs the algorithm to find patterns in the data.<figure class="kg-card kg-image-card"><img src="https://iaxds62ruoojbg1k4180.cleaver.rocks/images/posts/b11e8d863dd33c5e47d16405d6954f3f.png" class="kg-image" alt loading="lazy"></figure>The PredictionYour algorithm is now ready to accept input data for a sample case and provide a prediction of the kind of driver a prospect is likely to be, based on what it has learned.Let's assume that a new prospect has the following variables for their Zip Code, Age, Gender, Car Type - 3, 2, 3, 4To output the prediction for the values above, please input the following line of code-print(clf.predict([[3, 2, 3, 4]]))At this point, your screen should look like this:<figure class="kg-card kg-image-card"><img src="https://iaxds62ruoojbg1k4180.cleaver.rocks/images/posts/9873accecc44c819e498243d80998a14.png" class="kg-image" alt loading="lazy"></figure>To run the program, click on the green arrow and see the output on the right side of the pane to view the prediction.<figure class="kg-card kg-image-card"><img src="https://iaxds62ruoojbg1k4180.cleaver.rocks/images/posts/242bafde1d3024e2b08dc470cb15637a.png" class="kg-image" alt loading="lazy"></figure>Congratulations!! Your algorithm predicts that based on the values inputted, this new prospect driver will be a bad driver. Based on this information, we can show this prospect as a higher car insurance quote. This algorithm can now be saved and reused for other use-cases with modification. In the next post, we will examine how you can analyze results for thousands and tens of thousands of records. Only when we feed in thousands of records, we will see our algorithm properly train itself and be able to provide reliable outcomes.

Machine Learning 101 For Non-Programmers- I

Premium reporting on technology, startups, VC funding, and software from Techfastly tagged with programmers. Read about programmers here.

Top Writers

Subscribe to Magazine