I am not sure if there is a way to classify new data, I'll investigate further and update this post.
You may want to look at:
which gives you a robust way to implement Predictive Analytics.
is there any news about using built-in decision trees in Datameer. The zementis plug-in can execute within Datameer models, which have been trained somewhere else. So in the case of decision trees the logical flow would be:
1. Have data source intergarted with Datameer
2. Prepare data as training data (add known results) and use Smart-Analytics, train a decision-tree, get decision rule and find out that the problem is well suited for decision trees.
3. Export training data to another environment (e.g. R) and train again in this environment.
4. Move learned decision rules via PMML and Zementis plugin back to Datameer.
5. Execute zementis Plugin to classify data.
Do I understamd this right? Then it is hard to argue the value of "Smart Analytics", since you need the same learning algorithm in another envirinment anyway.
Your assessment of the difficulties of working with predictive analytics and Datameer is for the most part correct. Smart Analytics cannot integrate seamlessly into every workflow. There are many ways of addressing this, I feel this article could help describe more about how Datameer fits into a predictive modeling environment: https://datameer.zendesk.com/hc/en-us/articles/210188686.
The true power of the Smart Analytics module is not to replace tools that have been specifically created for data scientists, e.g. R or MLlib. Smart Analytics allows business users to use machine leaning algorithms to work on exploratory analyses and help form follow-on analyses. The learning curve of Smart Analytics is much lower than for R or MLlib.
You aren't the first person to suggest integrating Datameer more firmly into the predictive analytics environment, and I'm sure you won't be the last. But at the moment, our development team is concentrating on features that will make Datameer more powerful in general and easier to use.
Please sign in to leave a comment.