issue about the example

[nikdon]

issue about the example

[xanderdunn]

I'm having the same issue. To provide more detail:

Running this:

from pyentrp import entropy as ent
import numpy as np

x = np.random.rand(3000)
ts = [1, 4, 5, 1, 7, 3, 1, 2, 5, 8, 9, 7, 3, 7, 9, 5, 4, 3]

for values in [ts, x]:
    std_ts = np.std(values)
    sample_entropy = ent.sample_entropy(values, 5, 0.2 * std_ts)
    multiscale_entropy = ent.multiscale_entropy(values, sample_length=5)
    permutation_entropy = ent.permutation_entropy(values, order=5, normalize=True)
    print(sample_entropy)
    print(multiscale_entropy)
    print(permutation_entropy)

Produces this output:

/usr/local/anaconda3/envs/plutus/lib/python3.7/site-packages/pyentrp/entropy.py:163: RuntimeWarning: invalid value encountered in true_divide
  sampen =  - np.log(Ntemp[1:] / Ntemp[:-1])
/usr/local/anaconda3/envs/plutus/lib/python3.7/site-packages/pyentrp/entropy.py:163: RuntimeWarning: divide by zero encountered in log
  sampen =  - np.log(Ntemp[1:] / Ntemp[:-1])
[2.95099638 2.07944154        inf        nan        nan]
[nan nan nan nan nan nan nan nan nan nan nan nan nan nan nan nan nan nan]
0.5512400796500767
[2.18821295 2.18706536 2.1974189  2.19977194 2.11596861]
[3.11351531 2.07944154 2.7080502  ...        nan        nan        nan]
0.9960386507116764

It appears to be specific to sample lengths greater than 2. No warnings, inf, or nan are produced when the sample length is 2 or 1, but the problem arrises when the sample length is 3 or greater.

Multiscale is built on top of sample, so naturally it suffers the same issue. Permutation entropy appears to be fine.

Is it possibly a function of the length of the input series? The sample_entropy sample_length=5 does not produce any warnings, nans, or infs on the longer sequence of 3000 random values, although multiscale still produces nans even on the 3000-length input.

pyentrp 0.6. numpy 1.19.4.

[nikdon]

This error is related to division by 0. Most probably it is because of the input: at some scale this operation appears and that's why there are nans/inf in the result