3 Replies Latest reply on Jan 2, 2017 2:23 PM by Intel Corporation

    101 Curie Neural Network Lib - Can not change global context

    marcuso

      Hi,

       

      I call the setGlobalContext with any number, and it remains at 127.

       

      Here is the code :

       

      void setGContext(int gc)

      {

         CuriePME.setGlobalContext(gc);

         uint16_t newGC = CuriePME.getGlobalContext();

         if (newGC != gc)

         {

            Serial.print("Failed to ");

         }

         Serial.print("set Global Context to:");

         Serial.println(gc);

         Serial.print("Global context is Currently:");

         Serial.println(newGC);

      }

       

       

      Output

       

      Set Global context to 1

      Failed to set Global Context to:1

      Global context is Currently:127

       

       

      It doesnt matter if I am in KNN or RBF mode, it always stays at 127.

       

       

      Is there supposed to be 127 different contexts for each neuron ? There-by giving me the ability to have 17 different patterns ?

       

      If so does this mean I could have a context for each OCR character I'm trying to match, as well as having one context with a category for every OCR character I'm trying to match.

       

      So for example I could have context 5 for my general what is this character ( OCR image 0 - zero through to 9 - nice, so 10 digits learned in this context number 5. Then I could have (I'm not saying this is a good idea or even desirable - I'm just asking if its possible - so please nobody say this is a bad idea), a context for each digit, so context 10 is for only the number 0, 11 is only for the number 1 etc. I then test the image in the general context (5), if it fails to recognise, then iterate through each of the contexts from 10 - 19, and see if that matches the OCR image ?