I used your package to extract features from a timeseries. I managed to reduce the classification to just 10 features. Among them was the SY_RangeEvolve totnuq feature. After visually checking validation data after classification I noticed the strange behavior of this feature.
Suppose I have two signals
Signal (y): [10, 0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10]
Reversed Signal (y rev ): [10, 9, 8, 7, 6, 5, 4, 3, 2, 1, 0, 10]
The reversed signal is the normal signal but just reversed.
The totnuq of the first signal is 2 and the totnuq of the reversed signal is 11. It seems strange to me that a measure for stationarity depends so much on this.
Can't you add an option to do range_evolve once on a normal segment and then on a reversed segment. And then select the maximum of the two?
I used your package to extract features from a timeseries. I managed to reduce the classification to just 10 features. Among them was the SY_RangeEvolve totnuq feature. After visually checking validation data after classification I noticed the strange behavior of this feature.
Suppose I have two signals
Signal (y): [10, 0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10]
Reversed Signal (y rev ): [10, 9, 8, 7, 6, 5, 4, 3, 2, 1, 0, 10]
The reversed signal is the normal signal but just reversed.
The totnuq of the first signal is 2 and the totnuq of the reversed signal is 11. It seems strange to me that a measure for stationarity depends so much on this.
Can't you add an option to do range_evolve once on a normal segment and then on a reversed segment. And then select the maximum of the two?