Logi Predict Setup and Use

Logi Predict is a Logi application that provides a complete, self-service, predictive analytics experience.

Other documentation about installing, configuring, and using the Logi Predict application are available on DevNet, on the Logi Predict Resources page.

About the Logi Predict Application

Logi Predict is a web application that's part of the Logi Predict add-on module. It leverages the power of the Logi Info platform and adds predictive technology, allowing you to make accurate predictions, without the assistance of the data scientists normally associated with this task.

It allows you to:

  • Create statistical "models" and train them, using historical or transactional data
  • Create "prediction plans", which use the models to analyze current data
  • Visualize the predictions generated when a prediction plan is executed
  • Use the Predicted Column element in Logi Info to get predictions from data
  • Manage collections of models and prediction plans
  • Schedule prediction plans for automatic generation

and more.

If you haven't already, you should read About Logi Predict to get a basic understanding of the terms and techniques used in Logi Predict. 

The Predictive Process

The two-step process used by Logi Predict is shown in the following diagram:
 

In Step 1, historical or transactional data is processed using predictive algorithms to create a "prediction model", a process referred to as training a model. The model contains information about the patterns and other statistical indicators discovered during its analysis of the data. Logi Predict offers four types of models: Classification, Clustering, Forecast, and Outliers, which are discussed later on. A model only includes the results of its analysis of the historical data, not the historical data itself. Once created and trained, models are stored and can be re-used.

In Step 2, the model is applied to the "new" data (the input table), the data about which we want predictions, by creating a "prediction plan". The model applies the patterns and indicators it learned in Step 1 to the new data. The resulting predictions are either written back into the input table in new columns, or are written into a completely new output table. If an output table is used, it will have new rows inserted into it with values provided for all columns with matching names from the input table, in addition to the prediction values.

This step can be scheduled to run repeatedly against new or changing data over time.

Back to top

The Home Page

When you first launch Logi Predict for this first time, you'll see this initial page:

This is your Home page. The icons on the left side are menu options for different features in the application. The "Home" icon at the top of the menu will always return you to this page.

 Does your application have a different title, different colors, or different font sizes than the example shown here? The developer who configured your Logi Predict application may have customized it to reflect your organization's branding, to be consistent with other applications, or for other reasons. The basic functionality of the application remains the same, however.

 If you've been given special permissions, you may see some side menu items for special features. These features are discussed later in the Using the Schedule Manager section.

Back to top

Creating and Training a Prediction Model

As mentioned earlier, the first step in the predictive process is to create and train a model. 

Start the process by clicking the Prediction Models menu item, shown above. 

Then, on the Prediction Models page, click New Model to start creating a new model. 

Select the model type to be used, as shown above, and click Create. Unsure about which type to choose? Review About Logi Predict .

Instructions for creating and training each model type are contained in these separate topics:

After you've created and trained one or more models, return here to continue.

Back to top

Managing Prediction Models

After you've created a model, it will appear in a table when you visit the Prediction Models page: 

In the example shown above, table columns are keyed to the numbers in these descriptions:

  1. Model Name  - Click this link to review, edit, or retrain the model.
  2. Status - Indicates whether model training has been completed.
  3. Last Update Time- The timestamp for the latest edit or training.
  4. Prediction Plans - Click New Plan to create a new plan using this model. Also provides a link to any prediction plans that use this model.
  5. Actions - Click this icon to delete the model.

You can click the Model Name, Status, and Last Update Time column headers to sort the table on that column.

Back to top

Creating a Prediction Plan

To apply the information learned by training a model and generate predictions, you need to create a prediction plan. 

Start this process by clicking New Plan in either the model training results table or the prediction models table.

When the new plan page is displayed, you can give the plan a name by clicking the gear icon, as shown above. Then use the following steps to create and execute the plan.

Step 1. Select Data - Select the "source" data from which you want to generate predictions. This has to be data with a similar schema, but should not be the historical data you used to train your model. If you have multiple data sources to choose from, additional controls will appear. Select the columns to be used, and apply filters or formula columns as desired. The selected data will appear in a 10-row table.

You can expand and collapse the data selector controls using the  icons.


 Note that the data columns you select must include those that were used to train the model. An error message similar to the one shown above will be displayed if required columns have been left out.



Step 2. Select Prediction Save Options - The results of executing the prediction plan, the predictions, are written out to a table column. This can be to an existing column in the table of your choice, or to an entirely new table the process will create. The options are shown above.

For the first scenario, select a table from the available data and additional input controls will appear:

  1. Select the existing column that will receive the prediction results.
  2. Select the existing column that will receive the timestamp of the plan execution.
  3. (Outliers Model only) Specify whether to write out all rows, or just those with outlier values.

Or, in the second scenario, click Create a new table... and a panel will appear for the new results table details:

  1. Specify a name for the new table. It will be created in the same database as the source data.
  2. Specify the new table schema. It can include all of the source data columns, or just three columns: a unique key column specified here, the predicted values column, and a timestamp column. The latter schema gives you the ability to join the prediction results to other data, without duplicating all the source data, saving time and storage space.
  3. Specify the name of the column that will receive the prediction results.
  4. Specify the name of the column that will receive the plan execution timestamp.
  5. (Classification Model only) Specify whether to add columns that will include the prediction probability percentages.
     

Step 3. Execute Prediction Plan - Click Predict Now... to execute the plan. A "wait" icon will be displayed while the plan runs and processing is asynchronous, so you can exit Logi Predict and return later to check on progress.

 If you have appropriate permissions, you'll also see an Add Schedule... button. Scheduling plan execution is discussed in the Using the Schedule Manager section.

 Previous prediction values in existing tables will be overwritten without warning.  

 

When plan execution finishes, an entry is added to a results table at the bottom of the page, as shown above. The table entry displays summary information and will vary in appearance depending on the model type used. The table entries include some useful links:

Click the View link to open the results in an Analysis Grid, as shown above. You can further manipulate and export the data from here, if desired.


 

If you hover your mouse cursor over the distribution chart, you can see the relevant values, as shown above. Click the "Trashcan" icon to delete the results table entry.

 Deleting an entry in the results table does not delete prediction plan execution result data added to a data table. This data may be in use by other prediction plans or applications.

The data in your database table, including the prediction value, can now be used in whatever manner you find useful.

Back to top

Managing Prediction Plans

After you've created a plan, it will appear in a table when you visit the Prediction Plans page:  

Click the Prediction Plans menu item, shown above. 

In the example shown above, table columns are keyed to the numbers in these descriptions:

  1. Prediction Plan Name - Click this link to review, edit, or rerun the plan.
  2. Model Name - Identifies the prediction model associated with this prediction plan. Click the name link to edit the model.
  3. Last Update Time - The timestamp for the latest plan edit.
  4. Last Run Status - The status resulting from the last plan execution.
  5. Last Run Time - The timestamp and duration of the latest plan execution.
  6. Actions - Click this icon to delete the plan.

You can click all of the column headers, except Actions, to sort the table on that column.

Back to top

Using the Data Manager

If your system administrator has granted you the proper permission, you may be able to use the Data Manager tool to manage the data that all Logi Predict users can access for predictions. 

With the proper permission, you'll see the Data Manager option, shown above, in the menu. Click it to go to the Data Manager page.

When the Data Manager page opens in Logi Predict, it looks like the example shown above. Detailed instructions for using it to create and manage data connections and metadata are available in Using the Web Metadata Builder.

Back to top

Using the Schedule Manager

If you've been granted the proper permission and the Scheduler feature is enabled, you'll see several features related to the scheduling of prediction plan execution.

For example, when you're working with a prediction plan, the last step will include an Add Schedule... button. Click it to display the Prediction Schedule pop-up panel:


 

Depending on the schedule interval selected (Once, Minutes, Hourly, Daily, Weekly, or Monthly), the panel controls will change to allow you to configure the desired schedule. Click Save to create the scheduled occurrence.

Once you create a scheduled occurrence, the Add Schedule... button will change to an Edit Schedule... button.

You'll also be able to use the Schedule Manager tool to manage all scheduled prediction plan executions.

You'll see the Schedule Manager option, shown above, in the menu. Click it to go to the Schedule Manager page. 

In the example shown above, table columns are keyed to the numbers in these descriptions:

  1. Prediction Plan Name - Click this link to review, edit, or rerun the plan.
  2. Model Name - Identifies the prediction model associated with this prediction plan. Click the name link to edit the model.
  3. Description - Displays the scheduled interval. Click the Clock icon to edit the schedule.
  4. Status - The status resulting from the last scheduled plan execution.
  5. Last Run Time - The timestamp of the latest plan execution.
  6. Next Run Time - The timestamp for the next scheduled plan execution.
  7. Action - Click this icon to delete the scheduled plan execution.

You can click all of the column headers, except Description and Actions, to sort the table on that column.

Back to top