One of the main improvements in the IoT AggreGate Platform 5.5 release was machine learning, which gave data processing professionals a chance to extract required knowledge while analyzing time series streams and large data sets.
And more recently, we have added several algorithms allowing you to create trainable units with an option of online learning.
Such trainable units have two important features. The first is ability to train on a data set that is too large to fit into memory. The second is ability to update with new data that wasn’t available during initial learning.
In other words, a trainable unit can be trained on available data, used for predictions, and get updated as soon as new data arrives, which will improve the quality of further predictions.
Importantly, the trainable unit in this case is not trained from scratch, but updated, modifying its existing state with new training data set.
The plus point is that it takes significantly less time to complete online learning than would be needed for learning from scratch.