20. Machine Learning

20.1. Overview & Types

Simlytiks now has Machine Learning integrated for exploring data, allowing us to predict results and patterns. The ML features review inputs and outputs to data in order to learn how to predict another output. d3VIEW’s Simlytiks application offers three main types of machine learning: Regression, Classification and Clustering. As shown in the following image, regression and classification fall under supervised learning as they are task-driven, while clustering falls under unsupervised learning as they are data-driven.

Figure 1: Machine Learning Types in Simlytiks


Simlytiks has a wide variety of ML algorithms to employ. Let’s review the three main types of Machine Learning that can be utilized along with their specific algorithms in the application. To learn more about the Simlytiks application, please navigate to the Simlytiks Documentation at this link.

Regression Types

Supervised Regression ML types, aside from Logistic Regression, learn and predict numbers to plot out correlations between labels and data points. Here are the available regression algorithms in Simlytiks:

Icon

Name

Description

icon_linear

 Linear Regression Uses Linear Regression to build a model whose target is numeric.

icon_lasso

 Lasso Regression A type of Linear Regression. Lasso is often used when there is a large number of features as it automatically does feature selection.

icon_ridge

 Ridge Regression A type of Linear Regression, Ridge reduces the model complexity by sending coefficients to 0.

icon_svr

 Support Vector Regression Uses Support Vectors to build a model whose target is numeric.

icon_rfr

 Random Forest Regression Uses a Random Forest to build a model whose target is numeric.

icon_dtreereg

 Decision Tree Regression Uses a Decision Tree to build a model whose target is numeric.

icon_gpr

 Gaussian Process Regression Uses a Gaussian Process to build a model whose target is numeric.

Classification Types

Supervised Classification ML types, which include Logistic Regression, learn and predict text or a category by fitting the data into a statistical model. Here are the available classification algorithms in Simlytiks:

Icon

Name

Description

icon_feat_imp

 Feature Importance Provides scores for how each input relates to the target.

icon_getschema

 Get Schema Displays the Schema of the dataset.

icon_dtree

 Decision Tree Classifier Uses a decision tree to build a model whose target is categorical.

icon_rfc

 Random Forest Classifier Uses a Random Forest to build a model whose target is categorical.

icon_gnb

 Gaussian Naive Bayes Uses a Gaussian Naive Bayes to build a model whose target is numeric.

icon_logistic

 Logistic Regression Uses Logistic Regression to build a model whose target is categorical.

Clustering Types

Unsupervised Clustering ML types learn and predict features/patterns in data via the classifying and grouping of data points. Here are the available clustering algorithms in Simlytiks:

Icon

Name

Description

icon_kmeans

 K-Means Groups data based on a pre- defined number of groups.

icon_meanshift

 Mean Shift Groups data by shifting data to the mode of a region of a pre-defined size.

20.2. Creating Machine Learning Models

In this section, we’ll go over the basics for creating Machine Learning models in Simlytiks.

Check out this video to see the whole process of creating and predicting a model in action using linear regression. Make sure to read on for a more in-depth explanation.



Adding Algorithms

Access all the Machine Learning algorithms by clicking the green ML button at the top. Then, drag-and-drop or click to add one to the page.

Figure 2: Access a Machine Learning Algorithm

Setting Up an Algorithm

For Regression Types such a Linear Regression, required inputs include our Target Column (1 - what we are predicting) and our Input Features (2 - what we are learning from).

Figure 3: Setting Up Regression Types

For Classification Types such as Decision Tree Classification, required inputs also include our Target Column (1 - what we are predicting) and our Input Features (2 - what we are learning from).

Figure 4: Setting Up Classification Types

For Clustering Types, such as K-Means, required inputs include our Input Features (1 - what we are learning from) and our clustering parameters (2).

Figure 5: Setting Up Clustering Types

Live Data Table

When adding our inputs, we’ll see them highlighted and updated in a table to the right, with the orange being our target and green our features, as shown in this video with Linear Regression:



Executing an Algorithm

Once we have our required inputs set, we can execute our algorithms one-by-one by clicking on “Execute” (1) or all at once by adding each algorithm to the “Enqueue” (2).

Figure 6: Execution Options

For the Enqueue option, after we’ve added algorithms to it, we’ll see a button appear next to the green ML button called “ML Queue” (1). We’ll click it to see the available ML algorithms to run. Here, we can view or delete any of them (2), run them (3), or clear them (4).

Figure 7: Execution Enqueue

With this option, we’ll get meta-model of all the algorithms in one.

Figure 8: Regression Meta-model

ML Features

An executed algorithm will have multiple tabs of options to explore.

Figure 9: Machine Learning Tabs

Charts

We can view our charts in the Home and Charts Tabs.

Figure 10: View Charts

Here, we can use the pointing hand icon to pick a point on the chart to predict.



Or, we can use the upload icon to predict points from a CSV file.



We can also predict under the Prediction tab which we will go over shortly.

Outputs

The Outputs tab shows us a summary of our executed data.



Predictions

Under the Predictions tabs, we can input values manually (1) or upload a CSV (2) of values of our input features to predict our target column:

Figure 11: Predict Values

Predictions will populate and accumulate in a table to the right.



The prediction table has options to change the chart type (1), clear the values (2) or view the predictions via a JSON response (3).

Figure 12: Prediction Table Options

Chart Types aside from a table include Parallel Line and 3D Scatter:



Options

Enter a name or description for our ML model under Options

Figure 13: Options

Logs

Review the Lucy execution logs under the Logs tab.

Figure 14: Logs

Export/Save

Export and save the ML model to reuse later.

Figure 15: Export/Save

20.3. ML Examples

In this section, we’ll go over some examples of how to predict data using specific datasets.

Predicting HIC Value

For this example, we’ll be predicting the HIC value of a DOE simulation by learning from its t-bumper and t-hood values.

Opening the Dataset

To get started, navigate to the Simlytiks application.

Figure 16: Access Simlytiks

We’ll be using the Simulation DOE Results sample dataset. Open it under New Dataset -> Sample Datasets.

Figure 3: Open Simulation DOE Results Sample Dataset


Upon opening, we’ll see an overview table of all the data.

Figure 17: Overview Table


Add a Visualization

Let’s create a visualization to gain a better understanding of the relationship between the HIC, t-bumper and t-hood values in our dataset. We’ll need to add a visualization by first creating a new page.

Figure 18: Create a New Page


Then, we’ll click the blue button at the top that says “Add Visualization” and drag-and-drop or click to add the Scatter Plot Visualizer onto the page.

Figure 19: Add Scatter Plot Visualization


Choose HIC for our X Data By; then choose t-bumper and t-hood for our Y Data By, and press accept.

Figure 20: Choose X and Y Data By


We can see that our t-hood values seem to increase when our HIC values increase, while the t-bumper values are a bit more scattered.

Figure 21: HIC Vs. T-Bumper and T-Hood


icon_linear Add Linear Regression ML Algorithm

Let’s create another page and add the linear regression ML feature from the list by clicking on or dragging-and-dropping the feature.

Figure 22: Add ML Feature to Page


Learn Input Features

Our Input Features will be the column(s) we would like to learn from.

We’ll choose t-bumper and t-hood for our learning columns.

Figure 23: Input Features


Predict Target Column

Our Target Column will be the column we would like to predict.

In order to generate the predictions correctly using a regression type algorithm, we’ll only be able to add a numeric Target Column to predict data. We’ll choose to predict HIC and check the regression line/surface option as it will produce a 2D and 3D graph for the predictions.

Figure 24: Target Column


Predictions Predictions

Click on “Execute” to see the Machine Learning predictions for the data.

Figure 25: Execute Machine Learning


In this example, we are given three graphs to explore: a table, a surface linear regression graph and a parallel line chart.

Figure 26: Prediction Charts


icon_feat_imp Add Feature Importance ML Algorithm

Again, we’ll create another page and add the Feature Importance ML feature from the list by clicking on or dragging-and-dropping the feature.

Figure 27: Add ML Feature to Page


Predict Target Column

Our Target Column will be the column we would like to predict.

With classification type algorithms, we can add any text or number column as the output will be categorical. As before, we’ll choose to predict HIC.

Figure 28: Target Column


Learn Input Features

Our Input Features will be the column(s) we would like to learn from.

We’ll choose t-bumper and t-hood for our learning columns.

Figure 29: Input Features


Predictions Predictions

Click on “Execute” to see the Machine Learning predictions for the data.

Figure 17: Execute Machine Learning


In this example, we are given two graphs to explore: two bar charts showing importance and correlation between t-hood and t-bumper in reference to HIC value.

Figure 30: Prediction Charts


icon_kmeans Add K-Means ML Algorithm

Let’s create yet another page and add the K-Means ML feature from the list by clicking on or dragging-and-dropping the feature.

Figure 31: Add ML Feature to Page


Clusters Number of Clusters

Since K-Means is an unsupervised clustering type, we won’t be inputing a Target Column but instead deciding on the amount of clusters for grouping the data points.

Let’s take a look back at our scatter plot from earlier to decide on the number:

Figure 32: HIC Vs. T-Bumper and T-Hood


We could split the above graph into three sections or three groups, with the middle group being the most populated; so, we’ll choose to have three clusters for our K-Means algorithm.

Figure 33: Number of Clusters


Learn Input Features

Our Input Features will be the column(s) we would like to learn from.

Not only will we choose t-bumper and t-hood, but we’ll also add HIC for our learning columns, since we are still exploring the predictions for HIC value.

Figure 34: Input Features


Predictions Predictions

Click on “Execute” to see the Machine Learning predictions for the data.

Figure 35: Execute Machine Learning


In this example, we are given a scatter plot with three differently colored clusters or groups, mapped as t-hood vs t-bumper but clustered based on the HIC value as well.

Figure 36: Prediction Chart


Predicting Iris Flower Class

For this example, we’ll be predicting Iris flower class by learning from sepal and petal widths and lengths.

Opening the Dataset

To get started, navigate to the Simlytiks application.

Figure 37: Access Simlytiks

We’ll be using the Iris dataset that can be found at UCI’s Machine Learning Repository or by following this link. Drag-and-drop the dataset file onto the orange drop zone located on the left side panel of the Simlytiks main page.

Figure 38: Open Iris Dataset


Upon opening, we’ll see an overview table of all the data.

Figure 39: Iris Overview Table


Add a Visualization

Let’s create a visualization to gain a better understanding of the relationship between Iris class and petal/sepal dimensions. We’ll need to add a visualization by first creating a new page.

Figure 40: Create a New Page


Then, we’ll click the blue button at the top that says “Add Visualization” and drag-and-drop or click to add the Parallel Line Chart Visualizer onto the page.

Figure 41: Add Parallel Line Chart Visualization


For the Columns, choose class, petal width, petal length, sepal width and sepal length in that order, then click Accept.

Figure 42: Choose Columns


We can see that for each class the widths and lengths of the petals and sepals follow a similar trajectory.

Figure 43: Class, Petal Width & Length, Sepal Width & Length


icon_dtree Add Decision Tree Classifier ML Algorithm

Let’s create another page and add the Decision Tree Classifier ML feature from the list by clicking on or dragging-and-dropping the feature.

Figure 44: Add ML Feature to Page


Learn Input Features

Our Input Features will be the column(s) we would like to learn from.

We’ll choose the petal width, petal length, sepal width and sepal length as our learning columns.

Figure 45: Input Features


Predict Target Column

Our Target Column will be the column we would like to predict.

We’ll choose to predict the Iris class.

Figure 46: Target Column


Predictions Predictions

Click on “Execute” to see the Machine Learning predictions for the data.

Figure 47: Execute Machine Learning


In this example, we are given two graphs to explore: a table and a decision tree.

Figure 48: Prediction Charts


The decision tree helps us decipher the class by following the attributes down the branches.



Predicting World Happiness Score

For this example, we’ll be predicting World Happiness Score by learning from gdp per capita, social support, health, freedom, generosity, government trust and dystopia residual.

Opening the Dataset

To get started, navigate to the Simlytiks application.

Figure 49: Access Simlytiks

We’ll be using the World Happiness 2020 dataset that can be found at the World Happiness Report website. Drag-and-drop the dataset file onto the orange drop zone located on the left side panel of the Simlytiks main page.

Figure 50: Open World Happiness Dataset


Upon opening, we’ll see an overview table of all the data.

Figure 51: World Happiness 2020 Overview Table


Add a Visualization

Let’s create a visualization to gain a better understanding of the relationship between World Happiness Score and all the score variable columns. We’ll need to add a visualization by first creating a new page.

Figure 52: Create a New Page


Then, we’ll click the blue button at the top that says “Add Visualization” and drag-and-drop or click to add the Ranking Table Visualizer onto the page.

Figure 53: Add Ranking Table Visualization


For the Columns, choose gdp per capita, social support, health, freedom, generosity, government trust and dystopia residual. For the Primary Key, choose happiness score; then click Accept.

Figure 54: Choose Columns and Primary Key


We can see that the higher ranked countries with the higher ranked score variables (ranking blocks) generally have the highest happiness score (Name).



icon_dtree Add Decision Tree Classifier ML Algorithm

Let’s create another page and add the Decision Tree Classifier ML feature from the list by clicking on or dragging-and-dropping the feature.

Figure 55: Add ML Feature to Page


Learn Input Features

Our Input Features will be the column(s) we would like to learn from.

We’ll choose gdp per capita, social support, health, freedom, generosity, government trust and dystopia residual as our learning columns.

Figure 56: Input Features


Predict Target Column

Our Target Column will be the column we would like to predict.

We’ll choose to happiness score as our target column.

Figure 57: Target Column


Predictions Predictions

Click on “Execute” to see the Machine Learning predictions for the data.

Figure 58: Execute Machine Learning


In this example, we are given a table of numeric values to explore. Our ridge regression model gives us a more precise correlation between the score variables to the happiness score (intercept) than would a normal linear regression model.

Figure 59: Prediction Chart

20.4. Tutorial Videos

Skip to this section to see some Machine Learning in action.

Live Data Table

When adding our inputs, we’ll see them highlighted and updated in a table to the right, with the orange being our target and green our features, as shown in this video with Linear Regression:



Predicting from the Chart

Here, we can use the pointing hand icon to pick a point on the chart to predict, or the upload icon to predict points from a CSV file.



ML Outputs Tab

The Outputs tab shows us a summary of our executed data.



Predictions Tab

Under the Predictions tabs, we can input values manually or upload a CSV of values of our input features to predict our target column. Predictions will populate and accumulate in a table to the right.



The prediction table has options to change the chart type which include, aside from a table, Parallel Line and 3D Scatter.



Iris Prediction

The decision tree helps us decipher the Iris class by following the attributes down the branches.



Happiness Score Prediction

We can see that the higher ranked countries with the higher ranked score variables (ranking blocks) generally have the highest happiness score (Name).