是否有任何类型的包包含通过删除样本来平滑或均匀分布数组的功能？

如果这可以帮助将来的任何人，这就是我想出的：

def reject_outliers(x_t, y_t, m):
    mean = np.mean(y_t)
    std = np.std(y_t)
    x_t, y_t = zip(*[[x, y] for x, y in zip(x_t, y_t) if abs(y - mean) < (m * std)])
    return list(x_t), np.array(y_t)


def even_out_distribution(x_t, y_t, n_sections, reduction=0.5, reduce_min=.5, m=2):
    x_t, y_t = reject_outliers(x_t, y_t, m)
    linspace = np.linspace(np.min(y_t), np.max(y_t), n_sections + 1)
    sections = [[] for i in range(n_sections)]
    for x, y in zip(x_t, y_t):
        where = max(np.searchsorted(linspace, y) - 1, 0)
        sections[where].append([x, y])
    sections = [sec for sec in sections if sec != []]

    min_section = min([len(i) for i in sections])  # np.mean([len(i) for i in sections]) * reduce_min  # todo: in replace of min([len(i) for i in sections])
    print([len(i) for i in sections])
    new_sections = []
    for section in sections:
        this_section = list(section)
        if len(section) > min_section:
            to_remove = (len(section) - min_section) * reduction
            for i in range(int(to_remove)):
                this_section.pop(random.randrange(len(this_section)))

        new_sections.append(this_section)
    print([len(i) for i in new_sections])
    output = [inner for outer in new_sections for inner in outer]
    x_t, y_t = zip(*output)

    return list(x_t), np.array(y_t)

我这样做是为了如果它从 中删除一个样本y_train，它会从你的输入列表中删除它，或者x_train.

用法：x, y = even_out_distribution(x_train, y_train, n_sections=10, m=2, reduction=.8)现在我的分布图现在看起来好多了！