Suppose I have a time series such as:
[1, 1, 1, 1, 0, 1, 1, 1, 1, 0, 0, 0, 0, 0, 0, 1, 0, 0, 0 , 1, 1, 1, 1]
and I know there is some noise in the signal. I want to remove the noise as best I can and still output a binary signal. The above example would turn into something like:
[1, 1, 1, 1, 1, 1, 1, 1, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 , 1, 1, 1, 1]
I have implemented a naive rule-based approach where I iterate through the values and have some minimum amount of 1s or 0s I need to "swap" the signal.
It seems like there must be a better way to do it. A lot of the results from googling around give non-binary output. Is there some scipy function I could leverage for this?
There are two similar functions that can help you: scipy.signal.argrelmin and scipy.signal.argrelmax. There are search for local min/max in discrete arrays. You should pass your array and neighbours search radius as order. Your problem can be solved by their combination:
>>> a = np.asarray([1, 1, 1, 1, 0, 1, 1, 1, 1, 0, 0, 0, 0, 0, 0, 1, 0, 0, 0 , 1, 1, 1, 1], int)
>>> signal.argrelmin(a, order=3)
(array([4], dtype=int32),)
>>> signal.argrelmax(a, order=3)
(array([15], dtype=int32),)
Then you can just replace these elements.