brainconn.utils.matrix.threshold_proportional

threshold_proportional(W, p, copy=True)[source]

This function “thresholds” the connectivity matrix by preserving a proportion p (0<p<1) of the strongest weights. All other weights, and all weights on the main diagonal (self-self connections) are set to 0.

If copy is not set, this function will modify W in place.

Parameters:
  • W (numpy.ndarray) – weighted connectivity matrix
  • p (float) – proportional weight threshold (0<p<1)
  • copy (bool) – if True, returns a copy of the matrix. Otherwise, modifies the matrix in place. Default value=True.
Returns:

W – thresholded connectivity matrix
Return type:

numpy.ndarray

Notes

The proportion of elements set to 0 is a fraction of all elements in the matrix, whether or not they are already 0. That is, this function has the following behavior:

>> x = np.random.random((10,10)) >> x_25 = threshold_proportional(x, .25) >> np.size(np.where(x_25)) #note this double counts each nonzero element 46 >> x_125 = threshold_proportional(x, .125) >> np.size(np.where(x_125)) 22 >> x_test = threshold_proportional(x_25, .5) >> np.size(np.where(x_test)) 46

That is, the 50% thresholding of x_25 does nothing because >=50% of the elements in x_25 are aleady <=0. This behavior is the same as in BCT. Be careful with matrices that are both signed and sparse.