Use Case:

Predictive Machine Learning
Models of Earthquake Intensity and Location

This image is of a polyture dataflow for earthquake prediction using machine learning

Summary:

Task: Use Polyture’s AutoML to Make Earthquake Magnitude and Location Forecasts

Data From: Kaggle

Polyture Version: 0.13.10

Contact: team@polyture.com

Steps Taken:

  • Uploaded the dataset into the data warehouse
  • Ran Data Transform (Remove Unnecessary Columns)
  • Created a Train Test split
  • Utilized three AutoML nodes to create: 
      • Latitude Prediction Model
      • Longitude Prediction Model
      • Magnitude Prediction Model
  • Deployed the three models using the test data
  • Merged Latitude and Longitude predictions together with Column Append
  • Created Visualizations and Exports for the predictions

Walk Through:

Polyture data warehouse

Upload and Warehousing

The first step is to import the data into the Data Warehouse

Polyture allows you to upload multiple data types, connect to multiple APIs, and provides you with a unique view of all data sources.

 

Demonstrates the creation of a Polyture data source node and the quick insights panel

Import Data / Select Data Type

We then grab a CSV Data Source Node from the Import panel, and select the correct file.

Notice that Polyture already provides “Quick Insights”, such as an instant sentiment analysis histogram, in a panel on the right hand side of the import options.

Polyture column filter node

Perform Column Filter

Next we use a Column Filter node and attach it to the CSV data source.

In the options panel on the bottom right, we select the columns we wish to include in the data-set we will run AutoML on. 

In this case, everything but Date and Time

Polyture train test split node for automated machine learning

Perform Train / Test Data Split

To get a more accurate prediction, we are going to split the data into train and test tables.

We achieve this by connecting a “Train Test Split” node.

In the Settings panel, we select for data randomization, and split percentage. In this case, we will do an 80/20 train/split.

 

Send Data to AutoML

We now take the training data from the split node and attach it to three separate “Send to AutoML” nodes.

We then give each node a name within the settings panel.

 

Create AutoML Experiments | Apply Parameters

Navigate to the main navigation panel on the left hand side, navigate to “AutoML” and select “New Experiment”.

A panel will pop up, and we will give the experiment a name. In this instance, “Longitude”.

We then click “Settings” and select the data source, as well as the predict column. Polyture’s AutoML will automatically detect the AutoML Type (Classification / Regression). 

This process is repeated twice more for latitude and longitude.

 

Deploy the AutoML Nodes

Drag and connect three of the “Deploy AutoML Model” nodes to the Test data from the “Train Test Split” node.

Within the settings panel of each Deploy AutoML Model node, select the AutoML model you wish to deploy.

 

Join Latitude and Longitude

Latitude and Longitude will generate separate prediction columns. As a result, they will need to be joined back into one table. 

To do this, we will drag in a “Column Append” node and connect the outputs from both the Latitude and Longitude deployed models.

 

Insert & Connect Graph Nodes

Drag in and connect a “Dashboard Graph” node to the Latitude and Longitude Column Append.

Drag another “Dashboard Graph” node in and connect it to the Deployed Magnitude AutoML output.

 

Open Graph Editor Panel

Click on the graph nodes to reveal the settings panel. 

Click on the Graph Editor button to enter the editor.

For Latitude and Longitude, we will select Scatter Plot with the X axis being Longitude and the Y axis being Latitude.

For Magnitude we will select a Histogram. In the histogram configuration, we will select the predict column to generate the graph.

 

This image is of a polyture dataflow for earthquake prediction using machine learning

Export Predictive Data

Now that we have generated the graphs, we want to export all of our predictive data, in the context of the test data.

To do so, we drag in a “Column Filter” node and connect it to the test data output from the “Train Test Split” node. In the settings panel, we filter out the latitude, longitude and magnitude columns.

Finally, we drag in a “Column Append” node, and connect the Latitude & Longitude Column Append, the Test Data Column Filter, and the Magnitude Deployed AutoML output data.

We then drag in an “Export Data” node and connect it to the recently created “Column Append” node. From the export settings panel, we select CSV and give the file the name of “Prediction-Export.csv”

 

Resize and Space the Graphs in Dashboard View

For the best view of the created graphs, we open up the “Dashboard” view.

In the Dashboard view, we can resize the graphs to achieve the best possible visual presentation.

 

Results:

This exercise resulted in two unique visualizations:

  • A histogram graph that predicts future magnitude distribution of earthquakes
  • A scatter plot graph showing likely locations of future earthquakes by latitude and longitude.

Polyture’s AutoML was able to generate this data without any input from the user other than identifying the predictive column.