caput.weighted_median

Weighted Median Functions

caput.weighted_median.moving_weighted_median()

Compute moving weighted median for 1 and 2 dimensional arrays.

Parameters

  • data (array_like) – The data to move the window over. Can have 1 or 2 dimensions. The data type should be float64 or something that can be converted to float64.

  • weights (array_like) – The weights for the data. Can have 1 or 2 dimensions. The data type should be float64 or something that can be converted to float64.

  • size (int or tuple of int) – Size of the window. All values must be uneven.

  • method (str) – Either ‘split’, ‘lower’ or ‘higher’. If multiple values sastisfy the conditions to be the weighted median of a window, this decides what is returned: split: The average of all candidate values is returned. lower: The lowest of all candidate values is returned. higher: The highest of all candidate values is returned.

Returns

An array containing the weighted median values. The size is the same as the given data and weights, the data type is float64.

Return type

class:’numpy.ndarray’

Raises

  • ValueError – If the value of the window size was not odd.

  • RuntimeError – If there was an internal error in the C++ implementation.

  • NotImplementedError – If the data has more than two dimensions.

caput.weighted_median.quantile()

Calculate the weighted quantile of a set of data.

The weighted quantile is always calculated along the last axis.

The weights must be postive or zero for the calculation to make sense. This is not checked within this routine, and so you must sanitize the input before calling.

In the case that all elements have zero weight, a standard uniformly weighted quantile calculation is performed. If there is one, and only one non-zero weighted element that is always returned. For two or more non-zero weighted elements, we can proceed as expected.

If the quantile is “split”, i.e. it lies exactly on the boundary between two elements, we use the bounding non-zero weighted elements to calculate the quantile according to the chosen method.

Examples of special cases:

>>> quantile([1.0, 2.0, 3.0, 4.0], [1, 1, 0, 2], 0.5)
3.0
>>> quantile([1.0, 2.0, 3.0, 4.0], [1, 1, 0, 2], 0.5)
3.0
>>> quantile([1.0, 2.0, 4.0, 3.0], [0, 0, 0, 0], 0.5)
2.5
Parameters

  • A (array_like) – The array of data. Only 32 and 64 bit integers and floats are supported.

  • W (array_like) – The array of weights. Only 32 and 64 bit integers and floats are supported.

  • q (float) – The quantile as a number from zero to one.

  • method ({"lower", "higher", "split"}:) – Method to use if the requested quantile is exactly between two elements.

Returns

r – The calculated quantile. This has the same shape as A with the last axis removed. If A was one dimensional the value returned is a scalar.

Return type

np.ndarray or scalar

Raises

  • ValueError – If the array shapes do not match, or the quantile value is invalid.

  • TypeError – If the array types are not supported.

caput.weighted_median.weighted_median()

Calculate the weighted median of a set of data.

The weighted median is always calculated along the last axis.

See quantile for more information on the behaviour for some special cases.

Parameters

  • A (np.ndarray) – The array of data.

  • W (np.ndarray) – The array of weights.

  • method ({"lower", "higher", "split"}:) – Method to use if the requested quantile is exactly between two elements.

Returns

r – The calculated median. This has the same shape as A with the last axis removed. If A was one dimensional the value returned is a scalar.

Return type

np.ndarray or scalar