dilkins / tensoap Goto Github PK

When executing the following line in example 2. Zundel Cation:
sagpr_get_kernel -z 2 -ps PS1.npy -ps0 PS0.npy -s PS1_natoms.npy -o kernel1

I get this error

Traceback (most recent call last):
  File "/home/harpera/work/code/TENSOAP/bin/sagpr_get_kernel", line 109, in <module>
    main()
  File "/home/harpera/work/code/TENSOAP/bin/sagpr_get_kernel", line 103, in main
    [power,scalefac,power0,zt,use_hermiticity] = parsing.set_variable_values_kernel(args)
  File "/work/code/TENSOAP/soapfast/utils/parsing.py", line 339, in set_variable_values_kernel
    scale.append(np.load(scaling[i]))
  File "/work/code/miniconda3/envs/tensoap/lib/python3.9/site-packages/numpy/lib/npyio.py", line 417, in load
    fid = stack.enter_context(open(os_fspath(file), "rb"))
FileNotFoundError: [Errno 2] No such file or directory: 'PS1_natoms.npy'

I tried changing the input to:
sagpr_get_kernel -z 2 -ps PS1.npy -ps0 PS0.npy -s PS0_natoms.npy -o kernel1
which calculated the kernel, but I'm not sure if this is just a typo error or an error above in the tutorial.

This error usually happens when I ask sagpr_parallel_get_PS to try to split a trajectory into too many parallel calculations, thereby ending up with an empty slice which throws a spanner in the works. But I have an automated workflow that sometimes spits out very short trajectories, so it would be nice if the parallel PS script could handle these cases gracefully (basically, if -nrun < len(traj), only run the first len(traj) calculations in parallel and don't try to run more).

For reference, a typical error backtrace is below:

...skipping...
IOError: [Errno 2] No such file or directory: 'cubic_tetra_transition/PS0v_seg015.npy'
Traceback (most recent call last):
  File "/home/veit/SOAPFAST//bin/sagpr_get_PS", line 276, in <module>
    main()
  File "/home/veit/SOAPFAST//bin/sagpr_get_PS", line 270, in main
    get_power_spectrum(lam,frames,nmax=nmax,lmax=lmax,rc=rcut,sg=sig,ncut=ncut,periodic=periodic,outfile=outfile,cw=cweight,subset=subset,initial=initial,sparsefile=sparsefile,sparse_options=sparse_options,cen=cen,spec=spec,atomic=atomic,all_radial=all_radial,useall=useall,xyz_slice=xyz_slice,get_imag=get_imag)
    natmax    = len(max(all_names, key=len))
ValueError: max() arg is an empty sequence
Traceback (most recent call last):
  File "/home/veit/SOAPFAST//bin/sagpr_get_PS", line 276, in <module>
    main()
  File "/home/veit/SOAPFAST//bin/sagpr_get_PS", line 270, in main
    get_power_spectrum(lam,frames,nmax=nmax,lmax=lmax,rc=rcut,sg=sig,ncut=ncut,periodic=periodic,outfile=outfile,cw=cweight,subset=subset,initial=initial,sparsefile=sparsefile,sparse_options=sparse_options,cen=cen,spec=spec,atomic=atomic,all_radial=all_radial,useall=useall,xyz_slice=xyz_slice,get_imag=get_imag)
    natmax    = len(max(all_names, key=len))
ValueError: max() arg is an empty sequence
Traceback (most recent call last):
  File "/home/veit/SOAPFAST//bin/sagpr_get_PS", line 276, in <module>
    main()
  File "/home/veit/SOAPFAST//bin/sagpr_get_PS", line 270, in main
    get_power_spectrum(lam,frames,nmax=nmax,lmax=lmax,rc=rcut,sg=sig,ncut=ncut,periodic=periodic,outfile=outfile,cw=cweight,subset=subset,initial=initial,sparsefile=sparsefile,sparse_options=sparse_options,cen=cen,spec=spec,atomic=atomic,all_radial=all_radial,useall=useall,xyz_slice=xyz_slice,get_imag=get_imag)
    natmax    = len(max(all_names, key=len))
ValueError: max() arg is an empty sequence

(this happens when using -nrun 4 on a trajectory of length 1)

Line 73 of get_power_spectrum.py determines the maximum atomic density (number of atoms / cell volume) in the series of frames in order to estimate a safe maximum number of possible nearest neighbours in periodic structures.

However, for slabs with a large vacuum region this fails, as the maximum atomic density is artificially lowered by the vacuum region, making it hard to define a "safe" upper bound for n. This leads to segmentation faults in initsoap.pyx, due to out-of-bounds access to arrays with an upper bound of nnmax on some axis (e.g. "length", "efact", "harmonic").

I'm afraid it's not obvious to me what the best solution to this is, given that the size of a vacuum region is largely arbitrary - possibly simply a 'slab' flag which can be set at runtime which sets nnmax = 100?

This may simply be me not understanding the code properly, but it appears to me that "nat" when "ncen" should be used in many places throughout "compute_power_spectrum" in PS_utils.py and the subroutines called from there.

For example, comparing lines 173 and 182-184: the array omegaconj is created using nat[i] as the first dimension, but the first index iterated over runs to ncen[i].

The simplest solution is calling compute_power_spectrum with nat=ncen, which does not produce any errors, I can see even when ncen is a subset of nat. However, this leads of course to very confusing code...

Finally, the output power spectrum has an entry full of zeros for every atom which is not a centre, potentially significantly increasing the size of the files produced. This does make it easier to map the power spectrum onto its corresponding atom though.

When I was runnnig example 12. Application of LODE to the prediction of binding energies , I got this error. What should I do? I just pasete the code in example 12 .I have run example 1 correctly.

When calculating nonlinear tensor kernels, if an L=0 power spectrum is not provided the code doesn't "fail gracefully" but instead crashes. e.g. in sagpr_get_kernel -z 2 -ps PS1.npy -s PS1_natoms.npy -o K1, with -ps0 PS0.npy not specified, this crashes rather than warning the user that no L=0 power spectrum has been given.

The script cartesian_to_spherical.py should be modified so that atomic properties can also be converted.