Errors in sample_entropy and higuchi_fd on a small vector

[raphaelvallat]

Errors in sample_entropy and higuchi_fd on a small vector

[raphaelvallat]

Hi,

I think that it doesn't really make sense to calculate the entropy or fractal dimension on such small arrays. That said, I haven't formally tested what should be the minimum sample size for these functions. It would be helpful to add a more explicit error message when the sample size is too low.

Thanks,
Raphael

[FirefoxMetzger]

I just ran into this as well when benchmarking the old and the new kernel. The old kernel works fine from 6 samples onwards for order=2. My guess is that the old kernel accessed the samples in blocks of a certain length (perhaps 2*(order+1)?) and thus needs at least one window/block worth of data to operate.

The new kernel doesn't have this limitation and will work happily with any array length. That said, @raphaelvallat comment still stands: sample_entropy is not reliable on small sample sizes so results should be taken with a grain of salt.