Question

动作为装箱的3D点集中的每个bin选择最靠近质心的位置。

尝试使用binned_statistic_dd中的scipy.stats我试图获取此3D点集的分箱平均值。但是，它一次只能计算其中一个维度的平均值。

import numpy as np
from scipy.stats import binned_statistic_dd

points = np.array([range(2000), range(2000), range(2000)]).T
# points[:,1] is y
y_means, bin_edges, bin_indices = binned_statistic_dd(points, points[:, 1], statistic=np.mean, bins=np.cbrt(1000))
y_means.shape # (10,10,10)

接下来将计算每个均值到所有原始点的距离并采用最接近的点。导致以下完整实施：

def grid_sample(df):
    x_mean, edges, binnumber = binned_statistic_dd(df[['x', 'y', 'z']].values, df.x.values, bins=np.cbrt(1000), statistic=np.mean)
    y_mean, edges, binnumber = binned_statistic_dd(df[['x', 'y', 'z']].values, df.y.values, bins=np.cbrt(1000), statistic=np.mean)
    z_mean, edges, binnumber = binned_statistic_dd(df[['x', 'y', 'z']].values, df.z.values, bins=np.cbrt(1000), statistic=np.mean)
    xyz_mean = pd.DataFrame({'x_mean':x_mean.flatten(), 'y_mean':y_mean.flatten(), 'z_mean':z_mean.flatten()}).dropna()

    def find_nearest(point, df=df):
        idx = (np.sum(np.abs(df[['x', 'y', 'z']].values-point.values)**2, axis=1)**.5).argmin()
        return df.iloc[idx]

    df = xyz_mean.apply(find_nearest, axis=1)
    return df

问题两个问题：

如何避免计算每个维度的分箱统计信息？
是否有更智能的closest-to-centroid功能？我在scipy.stats

Scipy.stats binned_statistic_dd用于每个bin最接近质心的3D

0 个答案: