Elipsa User Guide

App Navigation

Site navigation menus exist on both the left have side and upper right hand portion of the page.  The left side offers a collapsable navigation bar while the icon in the upper right provides a drop-down to navigate to various pages as detailed below.

My Models

The My Models page allows the user to see a list of models that they have created and saved in the system along with key statistics of the individual model. Clicking on the model brings up the Model Detail Popup. For models where the create or improve model task is still running, the Status will appear under the model name.  Saved models will appear with the given name provided.  Non-saved models will appear with their temp ID. All temp models are active in the system for the day and are deleted in an overnight cleanup process.

Model Detail Popup

Clicking the model will launch the popup showing details of the completed model.  Below is a list of model attributes that are displayed:

  • Model Type: The type of model that the user selected in the model creation process (predicting a value, predicting an event, grouping like items, finding outliers)

  • Model Name: The name of the model selected by the Elipsa engine that best fits your data

  • Model Details: Results from the various models run against your data

  • Predictors: The list of predictors or attributes used by the model to make predictions

  • Accuracy: The % of items that the model predicted correctly on new data

  • Minimum Accuracy: The minimum class is the event outcome in your data that happens least frequently.  This measure is the percentage of items that the model predicted correctly of this minimum class to show how well the model performs on uneven data

  • Error: This is a measure of how confident the system was on items that it predicted incorrectly.  In other words, when it is wrong, how wrong was it.

  • Backtesting: A graphical representation of the true positives, true negatives, false positives, and false negatives of predictions on new data

  • Data Importance: A graphical representation showing which attributes/predictors are most important in making predictions

Save and Use Model (button): Save models for future use and to utilize the model for predictions on new data

Improve Model (button): Queue up process to run Elipsa model tuning to test different parameters and settings to improve the predictive power of the model

Scenario Analyzer (button): For saved models, navigate to the Scenario Analyzer page to make predictions on new data

Create Model

The Create Model page is a step by step process for the user to be able to create new predictive models with their data. The Create Model process can take several minutes depending on the size of the dataset that you are using for training. While the process is running, you can see the status updates of the progress or move to another section of the application while the create process works in the background.

Step 1 (Select Model Type):

Select the type of predictive model that you would like to apply to your data: Value (coming soon), Event, Grouping(coming soon), Outliers (coming soon)

Step 2 (Select Target):
  • Drop or browse to a file to upload into the system for analysis

  • Date Column: If your data set is time-series, specify the column name of the date in your data

  • Target Event Column: Specify the column in your dataset that represents the event outcome that you are looking to predict.

  • Event Name: Provide a name for the event that you are predicting.  This descriptive name is used to identify the column throughout the application

Step 3 (Select Predictors):

Select the columns from your file that you will use as attributes/predictors to predict the outcome in your target event set in step 2.

The left side will contain all columns from your uploaded file.  You can click to select/deselect individual predictors or select/unselect all.

The right arrow button in the middle, when clicked, will add the selected predictors to the list on the right.  Items can be removed one at a time from the list on the right by clicking the red X in front of the predictor name.  Alternatively, all items can be removed from the list by clicking the clear button in the top right.

The Create Model button will begin the automated machine learning creation of the model that best fits your data

Step 4 (Results):

As the model creation task is running, Step 4 will continuously update to show you the status and key statistics of the model creation process

  • Model Type: The type of model that the user selected in the model creation process (predicting a value, predicting an event, grouping like items, finding outliers)

  • Model Name: The name of the model selected by the Elipsa engine that best fits your data

  • Model Details: Results from the various models run against your data

  • Predictors: The list of predictors or attributes used by the model to make predictions

  • Accuracy: The % of items that the model predicted correctly on new data

  • Minimum Accuracy: The minimum class is the event outcome in your data that happens least frequently.  This measure is the percentage of items that the model predicted correctly of this minimum class to show how well the model performs on uneven data

  • Error: This is a measure of how confident the system was on items that it predicted incorrectly.  In other words, when it is wrong, how wrong was it.

  • Backtesting: A graphical representation of the true positives, true negatives, false positives, and false negatives of predictions on new data

  • Data Importance: A graphical representation showing which attributes/predictors are most important in making predictions

Save and Use Model (button): Save models for future use and to utilize the model for predictions on new data

Improve Model (button): Queue up process to run Elipsa model tuning to test different parameters and settings to improve the predictive power of the model

Status: The status message indicating how far along the process is

Save and Use Model

Clicking the Save and Use Model button will produce a popup asking for a Name and Description. Enter a name that will be used for this model throughout the application.  The description will provide you with the ability to provide more details on what the model is solving.

Improve Model

Once the model is created, the user has the option to allow the system to try and improve the model.  The Improve Model button initiates a process in the background to run through a series of automated data science experiments working to improve the results of the predictive model.  

The process will run in the background and could take up to an hour depending on the size of the training data.  The status of the process will be reported on the My Models page.

Scenario Analyzer

The Scenario Analyzer page allows the user to upload new unseen data to run predictions off of.

The Model Selection dropdown will list all models that you have saved in the system.  The user can drag or browse to new data to upload into the system to run predictions off of.  The Run Scenario button will utilize the columns from the uploaded file and the model in the dropdown the predict for the target event of the model.

The columns in the uploaded file need to match the names of the columns used to create the original model or an error will be returned.

Change Password

To change your password, click on the menu icon on the top right and go to Profile.  On the left-hand side of your profile you will see the option to set a new password.

 
 
 
 
 

Template Files

Below are example files to use when testing the application if you do not have your own files to utilize.

Zip files contain a file for training models as well as a second file starting with 'predict' to be used for testing against your model

Heart Failure Clinical Records

Dataset to predict heart failure of a patient

Prediction Column: death_event

Credit Card Default

Dataset to detect whether a customer will default on his or her credit card payment

Prediction Column: default payment

Bank Purchase

Dataset to predict whether a customer will purchase a specific product

Prediction Column: Purchase

Mine Safety

Dataset to predict the probability of a safety issue in a mine

Prediction Column: death_event

Insurance Fraud

Dataset to predict the likelihood of an insurance claim being fraudulent

Prediction Column: Fraud Reported

Shopper Intention

Dataset to predict whether a customer will make an online purchase

Prediction Column: Revenue

Real Estate Sales

Dataset to predict real estate sales over a certain price

Prediction Column: Over225

Traffic Behavior

Dataset to predict whether traffic will slow by a certain %

Prediction Column: Slow75

Truck Failure

Dataset to predict failure of Automatic Pressure System on Scania trucks

Prediction Column: Class

 
SOCIAL
  • Twitter
  • Black LinkedIn Icon