I would like to identify state transitions in a data set. An example is: "At which index does the sin(x) fall below 0.5?"
The only way I could think of is like this:
a = np.arange(0,10,.01)
b= np.sin(a)
c = np.roll(b,1)
c[:1] = 0
print(a[(.5 > b) & (.5< c)])
[2.62 8.91]
Can I do that without an additional array c? How can I detect cases like "At which index does sin(x) fall below 0.5 and stay there for 7 samples?" Would i need 7 additional arrays?
For 7 steps, you can just build an intermediate array:
mask = (b < 0.5)
window = 7
# count the number of times the value is below thresh in the window
below_thresh = np.sum([mask[i:len(mask)-window+i] for i in range(window)], axis=0)
mask1 = below_thresh == window
a[window + 1:][mask1[1:] & (~mask1[:-1])]
# out
# array([2.62, 8.91])
For higher number of steps, you can build the below_thresh faster with as_strided.
mask[1:] & (~mask[:-1]) is True and shifted by window + 1. So something np.where(mask[1:] & (~mask[:-1])) + window + 1?