Calculate the prediction raw error (score) based on the predicted cells and active cells

[StefanJin]

I just looked the code in htm_prediction_model and see the raw score is calculated based on the predicted columns and the active columns.

Due to the whitepaper and other published papers, I have no knowledge about how to determine whether a column is in prediction state or not, but I know the mechanism to determine the predicted cells. Thus, I’m confused if there exists a method to calculate the score based on the active cell and predicted cell. If yes, where is it? in which file?

Thank your

Best regard

[subutai]

@StefanJin

Thus, I’m confused if there exists a method to calculate the score based on the active cell and predicted cell. If yes, where is it? in which file?

That specific output value is computed in the TMRegion’s code:

numenta/nupic-legacy/blob/master/src/nupic/regions/tm_region.py#L542-L544

The underlying method is here:

numenta/nupic-legacy/blob/master/src/nupic/algorithms/anomaly.py#L30-L38

[StefanJin]

Thank you for your reply.

I have read the introduction about the prediction cells Vs. confidence as below.

From my point, the predicted columns is a column with non-zero confidence, that means, if a column has a cell whose confidence is greater than a threshold, then the column is in a predictive state. Am I right?

Best regards
Stefan