Don't use the 'S' flag for your string categories. Don't use any flag. NuPIC will know it's a category simply because the datatype is "string".
Flagging a data column as 'S' tells NuPIC to reset the current sequence each time that field changes. So every time a different value is seen in that field, the NuPIC model will reset learning so that it thinks it's seeing a new sequence. The same goes for the 'Reset' and 'Learning' flags. These are like hacks that allow you to send commands to the nupic model from within the data itself. When the values of these fields change in the data over time, the model is either reset or learning disabled/enabled. I'm not sure if the CategoryIdx is even exposed to the user. I think you can ignore it. --------- Matt Taylor OS Community Flag-Bearer Numenta On Tue, Sep 9, 2014 at 5:11 AM, Ghislain Gagne <[email protected]> wrote: > Hello! > > I'm working on a model to predict/detect anomalies of engine sensors. This > engine has different states, for example it can be in 'Reverse', 'Idle' or > 'Forward' while being in Gear-D1, -H2, F3, etc. My assumption is that sensor > values will have a different pattern depending on the state and gear of the > engine at a specific time. In the model’s input file, I'm using SequenceIdIDX > ('S' flag) for those fields (state, gear). Is this flag appropriate for what > i'm trying to do? > > Also, could someone please provide a definition of the following field flags > used during swarming? > > -SequenceIdIdx > -ResetIdx > -CategoryIdx > -LearningIdx > > Thanks, > > Ghislain
