Hello, i would like to ask if the behavior that i am observing with htm.core implementation is correct or not.
See two pictures,
On picture 1, there is one predictive cell.
Next timestep has become the column active and there is one winner cell, but it is not the one that was previously predicted.
Is that possible from theoretical point of view?
Note: there can be problem somewhere in my visualisation, or on other places
According to this paper it shouldn’t if i am not wrong.
It should be the one that was predictive that becomes the sole active cell in the next tine step. The visualization is awesome!!
I got it, my mistake. The problem is that the prediction cells that are shown are predictions for current timestamp.
So basically for the timestep 1 the lower cell was predicted but wasn’t activated because column was not active. At timestep 2, the upper cell was the winner cell because there was no predictions at all so the minicolumn was bursting and that one was chosen.
Now seems to work correctly.
I will definitely add visualisation of bursting columns, and probably its good idea to distinguish winner cells from active cells.
