Model is so slow to track the new trend

As Addonis i modified the hotgym prediction demo for my data (similiar the example, timestamp and energy consumption). I used the same parameters of the example and this is the output.
I don’t understand if the network works fine and if the trend of the prediction, anomaly score and anomaly likelihood is good or i have to change any parameters.
In this figure you see the actual value and prevision for 10000 records.


Here instead the anomaly score and the anomaly likelihood.

In other figure you a zoom of actual and prevision in the range 23 Jan e 26 Jan of the first figure.

In the other i don’t know if is correct that the model is so slow to track the new trend.

I used the htm.java library

Thank’s for your help…

1 Like

I moved this from #nupic to here.

Considering the model has only seen a few days of data, it does not surprise me that it takes a day or so to learn a new pattern.

Thank’s rhyolight!!! I needed a confirm… And what about you think about the same configuration of parameters used in this example?
n = 50
w = 21
resolution = 0.1
KEY.POTENTIAL RADIUS = 12
KEY.POTENTIAL PCT = 0.5
KEY.COLUMN DIMENSIONS = 2048
KEY.INPUT DIMENSIONS = 8