def compute_percentiles()

in orbit/utils/predictions.py [0:0]

• 23 lines of code
• 8 McCabe index (conditional complexity)

def compute_percentiles(arrays_dict, percentiles):
    """Compute percentiles of dictionary of arrays.  Return results as dataframe.
    Parameters
    ----------
    arrays_dict : dict
        a dictionary where each key will be the output column label of a dataframe and
        value are a 2d array-like of shape (number of samples, prediction length) required to compute
        percentile(s)
    percentiles : list
        A sorted list of percentile(s) which will be used to specify percentiles needed to compute across various
        arrays
    Returns
    -------
    dict
        The percentiles across samples with columns for `50` aka median and all other percentiles
        specified in `percentiles`.
    """

    computed_dict = {}
    run_check = False
    prev_shape = None
    for k, v in arrays_dict.items():
        curr_shape = v.shape
        if len(curr_shape) != 2:
            raise ValueError(
                "Input arrays_dict requires 2 dimensions: (number of samples, prediction length)."
                " Please revise input. Your input:{}".format(curr_shape)
            )
        if run_check:
            if curr_shape[1] != prev_shape[1]:
                raise ValueError(
                    "Input arrays has different lengths in second dimension. Please revise input."
                )
        # run prediction consistency after first iteration
        run_check = True
        prev_shape = curr_shape
        # row = nth percentile, col = prediction length
        computed_array = np.percentile(v, percentiles, axis=0)
        columns = [k + "_" + str(p) if p != 50 else k for p in percentiles]
        for idx in range(computed_array.shape[0]):
            computed_dict[columns[idx]] = computed_array[idx]

    return computed_dict