Frequently-Asked Questions

Here are some answers to frequently-asked questions about Logi Predict:

  1. What's the best way to get started with predictive analytics?
  2. Who should be on a predictive analytics planning team?
  3. What are the biggest mistakes organizations typically make?
  4. Is Logi Predict available as a standalone product?
  5. Does Logi offer advice about data quality?
  6. Does Logi provide a Professional Services offering for data cleaning?
  7. Can other applications initiate predictions and use the output from Logi Predict?
  8. Can we get a time series forecast, e.g. forecast sales, over multiple quarters?
  9. What types of prediction models does Logi Predict offer?
  10. What database servers does Logi Predict work with?
  11. Can prediction runs be scheduled to run automatically?
  12. Can Logi Predict be used with Java-based web servers?
  13. Does Logi Predict require a separate license?
  14. Does Logi Predict use R?
  15. Is data access configuration in Logi Predict secured?
  16. Does Logi Predict work with NoSQL DBs (MongoDB, ElasticSearch, etc.)?
  17. Does it integrate unstructured data for training, e.g. social media sentiment?
  18. Is Logi Predict fairly "plug-and-play", like the SSRM module?
  19. Can resulting data be saved into a database or plugged into charts?
  20. Does a large data set/decision tree affect the accuracy of the results?
  21. How long does it take to run models; how many iterations are needed?
  22. Do you have to first join and flatten all data?

1. What's the best way to get started with predictive analytics?
This easiest way to get started is to identify a problem that directly ties into your existing application. For example, suppose you have a customer churn application that reports how many and which customers churned. The natural way to enhance that application is add the ability to predict how many customers will churn and to identify them. You would get started by using historical churn data to train predictive models to identify who will churn, then use those models to predict who will churn in the future. You could embed that insight into your application as a prototype, which will immediately get then attention of management to move forward.

Back to top

2. Who should be on a predictive analytics planning team?
Planning members should include a Business Liason person, who can help identify the problem with the best ROI, and a Software Developer to create the predictive model and to embed the predictive insights into an application.

Back to top

3. What are the biggest mistakes organizations typically make when launching a predictive analytics initiative?
Here are three mistakes often encountered:

  • Focusing on the "cool" thing to do, instead of what is important for the business.
  • Showing predictive insight in one application and making the user go to a different application to take action.
  • Iterating on the usage of the current predictive insight and expanding on additional ones, turning your application into a data-led application.

Back to top

4. Is Logi Predict available as a standalone product?
No. Logi Predict is currently an add-on to Logi Info, it is not available as a standalone product. You must also purchase Logi Info.

Back to top

5. Does Logi offer advice about data quality?
Maybe. In general, data quality is an issue to be addressed by you, although Logi Predict does handle common problems. Logi may provide guidance about data cleansing in some cases.

Back to top

6. Does Logi provide a Professional Services offering for data cleaning?
No. There are no plans to offer such a service at this time.

Back to top

7. Can other application initiate predictions and then use the output from Logi Predict?
Yes. There are two ways this can be done: for batch predictions, you can connect to the database table that receives the output from Logi Predict and, starting with v2.1, for individual predictions, you can use the Logi Predict REST API to get the result as JSON data. The latter lets you embed real-time predictive functionality in other applications. For more information, see Logi Predict API.

Back to top

8. Can we get a time series forecast, e.g. forecast sales, over multiple quarters?
Yes. Logi Info itself already provides elements that can perform a linear or non-linear time series forecast, which is typically done by extrapolating known values of a variable without consideration of other variables that can impact the prediction. Logi Predict, however, uses more sophisticated algorithms to factor in the impact of multiple variables on a prediction.

Back to top

9. What types of prediction models does Logi Predict offer?
Logi Predict 3.0 allows you to create Classification Models, Forecast Models, Cluster Models, Time Series Models, and Outliers Models.

Back to top

10. What database servers does Logi Predict work with?
Logi Predict works with these database servers:

  • Amazon Redshift
  • Microsoft SQL Server
  • MySQL
  • Oracle
  • PostgreSQL v9.1+

Back to top

11. Can prediction runs be scheduled to run automatically?
Yes. The Logi Scheduler (part of Logi Info and included with Logi Predict) can be used to schedule prediction plan execution. The ability to create and modify the scheduling of runs is a special permission Administrators can grant to all or just to designated users.

Back to top

12. Can Logi Predict be used with Java-based web servers?
Yes. Logi Predict, in a Java version, can be installed on a computer using a Install Logi Predict - Windows or a Install Logi Predict - Linux, for use with Java-based web servers.

Back to top

13. Does Logi Predict require a separate license?
No, but it does require a Logi Info license file, a copy of the same file you use to license Logi Info.

Back to top

14. Does Logi Predict use R?
Yes. Logi Predict makes use of the R statistical analysis environment and its libraries.

Back to top

15. Is data access configuration in Logi Predict secured?
Yes. Administrators can control access to the data connections and schema, granting it to all or just to designated users.

Back to top

16. Does Logi Predict work with NoSQL DBs (MongoDB, ElasticSearch, etc.)?
Yes, it generally works well with tabular data sources that can be represented as SQL data.

Back to top

17. Does Logi Predict integrate unstructured data for training, such as social media sentiment?
Yes, social media content can be used. For example, our retail customers typically want to include social media information such as "liked" pages, tweets, "liked" genres of music, sports interest, etc. This type of data is something that Logi Predict utilizes very well. However, Logi Predict doesn't directly consume free-form text, though is an area of interest for the future. Metric outputs derived from free-form text can currently be used.

Back to top

18. Is Logi Predict fairly "plug-and-play", like the SSRM module?
Yes, Logi Predict is meant to be used by developers, product managers, and application teams to create predictions and embed them into their applications. The Logi Predict application is ready to use, out-of-the-box, with minimal configuration, like the Info Go application distributed with SSRM. There's no need to write any R code or to be a data scientist to use it.

Back to top

19. Can resulting data be saved into a database or plugged into charts?
Yes, all predictions created can be either consumed in real-time through an API, or saved into a database table and used to generate charts and dashboards showing trends.

Back to top

20. When using a classification model, does a large data set/decision tree affect the accuracy of the results?
Not necessarily. First, you'd use the Column Importance feature to identify the right set of columns to use. Then run multiple statistical samples and possibly create an "ensemble" (average) sample of the predictions from multiple models. For example, one can create a model for January, a model for February, and a model for March, and then create predictions with all the models and use the weighted average of their predictions.

Back to top

21. How long does it take to run the models and how many iterations are needed to get the best fit?
The time it takes to run a model depends on the size of the data and the number of columns used. Typically, customers create statistical samples of the data and create "ensemble" models. For example, one can create a model for January, a model for February, and a model for March, run predictions with all the models and then use the weighted average of the predictions. If a model is trained just once a quarter, the amount of time it takes is usually not an important consideration. Most customers are more concerned about how long running a prediction plan will take.

Back to top

22. Do you have to first join and flatten all the data?
You can join data on-the-fly, by joining tables in the Logi Predict data manager. You do not have to independently flatten and save data in a separate table.

Back to top