Question

考虑到簇超出值范围的限制，如何确定周期性数据的簇的最小值和最大值（在0到24之间）？

查看蓝色群集，我想将值22和2确定为群集的边界。哪种算法可以解决这个问题？

Answer 1

我找到了解决问题的方法。假设数据采用以下格式：

#!/usr/bin/env python3

import numpy as np

data = np.array([0, 1, 2, 12, 13, 14, 15, 21, 22, 23])
labels = np.array([0, 0, 0, 1, 1, 1, 1, 0, 0, 0])
bounds = get_cluster_bounds(data, labels)
print(bounds) # {0: array([21,  2]), 1: array([12, 15])}

您可以在此处找到该功能：

#!/usr/bin/env python3

import numpy as np


def get_cluster_bounds(data: np.ndarray, labels: np.ndarray) -> dict:
    """
    There are five ways in which the points of the cluster can be cyclically
    considered. The points to be determined are marked with an arrow.

    In the first case, the cluster data is distributed beyond the edge of
    the cycle:
         ↓B           ↓A
    |#####____________#####|

    In the second case, the data lies exactly at the beginning of the value
    range, but without exceeding it.
    ↓A        ↓B
    |##########____________|

    In the third case, the data lies exactly at the end of the value
    range, but without exceeding it.
                 ↓A       ↓B
    |____________##########|

    In the fourth, the data lies within the value range
    without touching a border.
            ↓A       ↓B
    |_______##########_____|

    In the fifth and simplest case, the data lies in the entire area without
    another label existing.
     ↓A                   ↓B
    |######################|

    Args:
        data:      (n, 1) numpy array containing all data points.
        labels:    (n, 1) numpy array containing all data labels.

    Returns:
        bounds:   A dictionary whose key is the index of the cluster and
                  whose value specifies the start and end point of the
                  cluster.
    """

    # Sort the data in ascending order.
    shuffle = data.argsort()
    data = data[shuffle]
    labels = labels[shuffle]

    # Get the number of unique clusters.
    labels_unique = np.unique(labels)
    num_clusters = labels_unique.size

    bounds = {}

    for c_index in range(num_clusters):
        mask = labels == c_index
        # Case 1 or 5
        if mask[0] and mask[-1]:
            # Case 5
            if np.all(mask):
                start = data[0]
                end = data[-1]
            # Case 1
            else:
                edges = np.where(np.invert(mask))[0]
                start = data[edges[-1] + 1]
                end = data[edges[0] - 1]

        # Case 2
        elif mask[0] and not mask[-1]:
            edges = np.where(np.invert(mask))[0]
            start = data[0]
            end = data[edges[0] - 1]

        # Case 3
        elif not mask[0] and mask[-1]:
            edges = np.where(np.invert(mask))[0]
            start = data[edges[-1] + 1]
            end = data[-1]

        # Case 4
        elif not mask[0] and not mask[-1]:
            edges = np.where(mask)[0]
            start = data[edges[0]]
            end = data[edges[-1]]

        else:
            raise ValueError('This should not happen.')

        bounds[c_index] = np.array([start, end])

    return bounds

查找循环数据集群的最小值和最大值

1 个答案: