trjhtr

Given the following artificially generated data:

t_steps = 30

data = np.array([
 np.arange(t_steps) * .05,
 np.arange(t_steps) * .1,
 np.arange(t_steps) * .2,
 np.arange(t_steps) * .3

])

I find the time-step the each line of data has passed a threshold. If it does not pass the given threshold, I assign a time-step of -1:

react_tms = 
thresh = 3.5

for dat in data:
 whr = np.where(dat > thresh)
 if len(whr[0]) == 0:
 react_tms.append(-1)
 else:
 react_tms.append(whr[0][0])

This gives:

[-1, -1, 18, 12]

Is there some way to do this without the for-loop? Even before the for-loop is removed, should I be using something other than np.where to find the threshold crossing?

In principle you can use numpy.argmax for this. The only problem is that if no value is above the threshold, the maximum is False, so it returns 0 as the index of the maximum. We therefore need to subtract 1 for those cases:

above_threshold = data > thresh
react_tms = np.argmax(above_threshold, axis=1)
react_tms = react_tms - (~np.any(data > thresh, axis=1)).astype(float)
print(react_tms)
# array([ -1., -1., 18., 12.])

Whether or not that is really more readable, I am not sure. It is, however slightly faster than using numpy.where (and probably also faster than a list comprehension): https://stackoverflow.com/q/16243955/4042267

In the end this does not really matter: