On demand analysis – Zementis
About 2 years ago, Alan and me discussed with Dr.Andreas for the analysis service based on the “on demand” mode. We thought that this kind of service faced the following challenges:
1 how to ensure data safety?
2 how to express the modeling?
3 how to processing high volume data/performance?
Anyway, Zementis had the first step as a predict analysis service provider- their product was named as ADAPA. They took PMML as the describe language to define the models. They access the “on demand” via hosting on Amazon EC2. Currently, they offer these models samples:
Predictive Models
Clustering Models
Decision Trees
Naive Bayes Classifiers
Neural Network Models
Regression Models
Scorecards
Support Vector Machines
It is said it supports R, SAS…model platform very well. Hope this is a correct, new direction. I guess they only solve the 2nd question listed above, they still have a long way to go if they want to attract some real enterprise level applications/orders.
PS: It seems that SAP BO take Zementis as the BO explorer development contest now:
Updated: Thanks for Alex from Zementis giving us more details
Hi ZZ,
Thanks for commenting on ADAPA and Zementis. We do address all the questions you pose in your blog with ADAPA.
1) How to ensure data safety? All ADAPA instances are private and secure. You don’t share models between users and data upload and download is done via HTTPS. For more information on security, please follow the link below.
http://adapasupport.zementis.com/2009/01/adapa-predictive-analytics-scoring.html
2) How to express predictive models? Basically, predictive models should be expressed in PMML. PMML stands for Predictive Model Markup Language and can be exported by the top statistical packages. ADAPA reads models in PMML and makes them available for execution right away. For more, please see:
http://www.zementis.com/cloud-tour.htm
3) How to process high volume data/performance? ADAPA is extremely fast. We have conducted a study which shows that in average, ADAPA can execute more than 300 Million transactions per hour on a top instance on the Amazon Cloud. For more on performance, please see:
http://adapasupport.zementis.com/2009/06/how-to-score-300000000-customer-records.html
Hope this helps. Feel free to contact us for more info. There is also a lot of information about ADAPA and PMML in our support site:
http://adapasupport.zementis.com/
By ZZ
Hi ZZ,
Thanks for commenting on ADAPA and Zementis. We do address all the questions you pose in your blog with ADAPA.
1) How to ensure data safety? All ADAPA instances are private and secure. You don’t share models between users and data upload and download is done via HTTPS. For more information on security, please follow the link below.
http://adapasupport.zementis.com/2009/01/adapa-predictive-analytics-scoring.html
2) How to express predictive models? Basically, predictive models should be expressed in PMML. PMML stands for Predictive Model Markup Language and can be exported by the top statistical packages. ADAPA reads models in PMML and makes them available for execution right away. For more, please see:
http://www.zementis.com/cloud-tour.htm
3) How to process high volume data/performance? ADAPA is extremely fast. We have conducted a study which shows that in average, ADAPA can execute more than 300 Million transactions per hour on a top instance on the Amazon Cloud. For more on performance, please see:
http://adapasupport.zementis.com/2009/06/how-to-score-300000000-customer-records.html
Hope this helps. Feel free to contact us for more info. There is also a lot of information about ADAPA and PMML in our support site:
http://adapasupport.zementis.com/
Best,
Alex