I'm attempting to define a simple exponential and regardless of the numbers I put into the argument of the exponential I get the following:
FloatingPointError: underflow encountered in exp
The array I'm trying to define is as follows:
time = np.arange(length)
window = np.exp(-(time-512)**2/1000.0)
where length = 4096. I'm rather inexperienced with Python and saw a similar question was answered on here somewhere but I didn't really understand it. I should also note that this code ran before without a problem.
Break down what you're computing for the final element:
temp1 = 4096 - 512
temp2 = temp1**2
temp3 = -temp2
temp4 = temp3 / 1000.0
temp4 is -12845.056
Now, what happens when you try to take the natural anti-log of that number? What is the range of float values you're allowed in numpy?
I believe you'll find that you just tried to make it too small. If this is an intermediate result in a computation, I suggest that you hold the number as a natural log until you can make it larger again. Alternately, research some packages to support extreme values, or multiply all your values by a scalar large enough to preserve accuracy.