A couple of months ago, I posted a blog Using Oracle ADW machine learning models in Oracle Analytics in which I described how can an Oracle Machine Learning Model from Oracle database can be used in Oracle Analytics.
In previous release of Oracle Analytics, Register ML Model option was introduced. This gives users in Oracle Analytics option to register and use a machine learning model that was prepared in Oracle Autonomous Database.
So, let’s begin with a new ML model registration.
Registering ML Model with Oracle Analytics
In the top bar, next to the Create button, there is an option, Register ML Model, available from the Page Menu.
After connection is selected, ML model needs to be selected. We will work with GLM_HOUSING machine learning model:
The registration process is really straight forward. However, when registering a new ML Model with Oracle Analytics, that ML Model in previous releases acted as a black box. There were no statistics available to the user, hence no one could really tell if the model is performing well, or not.
Related Model Views
All data scientists, who are familiar with Oracle Machine Learning in Oracle database, on the other hand know that Oracle database actually stores quite a lot of information about created machine learning data models. This information is stored in views which names begin with DM$V.
The list of the relevant model views can now be seen and analysed in Oracle Analytics.
Let’s examine first which are relevant model views. Simply choose Inspect from the Actions Menu and navigate to Related section in the ML model:
Of course, you can check all of these model information and statistics in your database. You can find more detailed explanation on Model Detail Views in Oracle Data Mining User’s Guide.
For example, let’s take a look at another view, DM$VGGLM_HOUSING, which describes model statistics:
But of course, better way is to view it directly in Oracle Analytics. Unfortunately, there is no “one-click” solution available at the moment, however, it is not too complicated to import model views as new data sets.
And we can now also use this in a project where, for example, we can compere statistics from different models and consequently decide which to choose in our analysis and prediction:
I am not sure, but I would also expect some other views in that list above. It is quite possible as this blog is being written based on Oracle Analytics 5.9 Early Release, so possibly not everything is in its place.
Anyway, let me point out another important model detail view, DM$VDGLM_HOUSING.
Model views that begin with DM$VD contain the final mode information for Linear Regression. Please find all the details in Oracle Data Mining User’s Guide, but let me just show you what are the features and statistics relevant to the GLM_HOUSING model, which in now fully described with features and their combinations:
Conclusion
Machine Learning models that have been registered with Oracle Analytics, didn’t provide any details about their performance. Hence, users were not sure if they are deploying good or bad machine learning model. Now, with the latest updates in Oracle Analytics 5.9, this has changed. However, you have to know a little bit of machine learning in order to understand the information that is stored in machine learning model views. In case, users can have some more insights about the model they are deploying.