Feature "kurto_z" RuntimeWarning: invalid value encountered in double_scalars

[lbogaardt]

Feature "kurto_z" gives a RuntimeWarning: invalid value encountered in double_scalars for some tiles. I think such warning should be logged somewhere (for example in the comment section of the .ply file).

python /home/ubuntu/feature_scripts/computefea_wtargets_cell.py /data/local/eecolidar/modules/python/laserchicken/ /data/local/eecolidar/rclone/tmp/ahn3_256x256_2km_norm/tile_76_79.LAZ /data/local/eecolidar_webdav/01_Work/ALS/Netherlands/ahn3_targets_100m/tile_76_79_target.laz 50 /home/ubuntu/tile_76_79.ply
------ Import is started ------
Number of points: 59068115
Number of points in target: 400
------ Computing neighborhood is started ------
Cylinder size in Bytes: 13697343969.7
Memory size in Bytes: 33737809920
Start tree creation
Done with env tree creation
Done with target tree creation
build kd-tree: 46.377516 sec
Computed neighborhoods list length at iteration 0 is: 400
------ Feature calculation is started ------
Feature "eigenv_3"
Number of targets: 400, number of features: 3
took 2.74 seconds
Feature "point_density"
Number of targets: 400, number of features: 1
took 0.71 seconds
Feature "pulse_penetration_ratio"
Number of targets: 400, number of features: 2
took 24.42 seconds
Feature "perc_90"
Number of targets: 400, number of features: 10
took 4.30 seconds
Feature "sigma_z"
Number of targets: 400, number of features: 1
took 2.82 seconds
Feature "kurto_z"
Number of targets: 400, number of features: 10
/data/local/eecolidar/modules/python/laserchicken/laserchicken/feature_extractor/height_statistics_feature_extractor.py:30: RuntimeWarning: invalid value encountered in double_scalars
coeff_var_z = np.std(z) / np.mean(z)
took 1.44 seconds
feature calc: 36.425914 sec

[romulogoncalves]

That could mean mistake no how things are read or corrupted data. Could you share the details on how things were run? Like script, target and src point cloud?

[lbogaardt]

The script and input data is included in my issue above.

I think the problem comes when all 'z' is zero. Then std(z)=0 and mean(z)=0 so the coeff_var_z = np.std(z) / np.mean(z) calculation gives inf. It probably should give back zero. So coeff_var_z = if mean(z) == 0 then 0 else np.std(z) / np.mean(z);