Skip to content
cogmission edited this page Nov 24, 2014 · 20 revisions

As of today, a great amount of functionality can be derived from the use of the current code. Today the code consists of a complete SpatialPooler and TemporalMemory, which is all that's required in order to do basic inferencing.

However, the SpatialPooler (the basic component entry point into the CLA model) requires Sparse Data Representations(SDR's). Here is some information regarding this topic:

We say "basic component", because typically the entry point would be a containing entity called a Region.

In order to produce SDR's, an Encoder is required. There are several Python encoders; some for general use and some for more specialized use cases. If you are anxious to get started using the Java version of the CLA algorithms, then at least a rudimentary implementation of these would be required. (Fear not! The Java version of this functionality is being developed as we speak!) This information is merely provided for those who absolutely cannot wait :-)

UPDATE !!: 6 encoders are now completed: ScalarEncoder, CategoryEncoder, CoordinateEncoder, GeospatialCoordinateEncoder, PassThruEncoder, and SparsePassThruEncoder!

The output of the SpatialPooler is fed into the TemporalPooler which then outputs SDR's representing predicted input of the next inferencing cycle. In order to relate this "prediction" to a result in the context of the input, in some cases a Classifier is used.

While it is true that some missing components will limit the usability of this code in its present stage of development, there is still some basic utility which can be derived; and rest assured, more is on its way in the upcoming weeks!

For more information about upcoming development, see: the Roadmap

Next: Architecture