facing nan for mu

[hahakid]

facing nan for mu at 2nd round I use local LiDAR data, and can generate samples as above figure, while using E-M to evaluate the mu, I always get Nan value.

[RansML]

gamma=gamma, grid=None, cell_resolution=cell_resolution, cell_max_min=cell_max_min

Do your parameters make sense? I'd start with something like cell_max_min=[-40,40,0,100] and cell_resolution=10.

Also, can you check how many hinge points you have? print(model.grid.shape)

If these are correct, then then I'd test several gamma values. e.g. =0.1, 10, 100.

Alternatively, if you want to use the same parameters as in the example code, you can rescale your sampled data to be within the same range as the example.

[hahakid]

here is the point cloud and hinge points. The shape is 5*5.

In the first round I can get values for mu and sig, and the second turns to Nan for both params.

Changing the gamma shows the same results. I will try sbhm.py instead.

[hahakid]

How do you preprocess the KITTI data? I try to send pointcloud[x,y,z] into the function get_filled_txy_3d(X, y, robot_pos=np.array([0,0,0]), max_range=80, unoccupied_points_per_meter=0.1, margin=0.05) as X, and can get a similiar file formate as the two frame kitti data.

[RansML]

.