Numpy module

ssp.numpy.ssp.ssp(x1, x2, batched=False)[source]

Surface Similarity Parameter (SSP) without any fuzz, cf. https://doi.org/10.1007/s10665-016-9849-7. This implementation does not rely in any FFT call and thus complies with n-D data.

Parameters:

  • x1 (ndarray) – First input array for comparison

  • x2 (ndarray) – Second input array for comparison

  • batched (bool, optional) – If this is set to True, the result is not reduced along the first dimension. This option allows to use the SSP for comparing batches of data.

Returns:

ssp – Surface Similarity Parameter between arrays x1 and x2.

Return type:

ndarray

Notes

The result is always within [0, 1] with

  • 0 indicating perfect agreement, and

  • 1 indicating perfect disagreement.

Perfect disagreement in terms of SSP means either

  • a non-zero signal is compared against a zero signal, or

  • a signal is compared against the inverse itself, e.g., ssp(x, -x).

Examples

>>> from ssp.numpy import ssp
>>> import numpy as np
>>> np.random.seed(0)  # for deterministic results
>>> x1 = np.random.random((2, 32))
>>> x2 = np.random.random((2, 32))
>>> ssp(x1, x2)  # some value between 0 and 1
np.float64(0.35119514129237195)
>>> ssp(x1, x1)
np.float64(0.0)
>>> ssp(x1, -x1)
np.float64(1.0)
>>> ssp(x1, np.zeros_like(x1))
np.float64(1.0)

Use batched=True to get a unique result for each signal in x1 and x2

>>> from ssp.numpy import ssp
>>> import numpy as np
>>> np.random.seed(0)  # for deterministic results
>>> x1 = np.random.random((2, 32))
>>> x2 = np.random.random((2, 32))
>>> ssp(x1, x2, batched=True)
array([0.34864963, 0.35827101])
>>> ssp(x1, x1, batched=True)
array([0., 0.])
>>> ssp(x1, -x1, batched=True)
array([1., 1.])
>>> ssp(x1, np.zeros_like(x1), batched=True)
array([1., 1.])