`ani_quicktest.py` example requires changes to script

[bannanc]

Hello, I'm just getting started with ASE_ANI and am trying to make sure the simple ani_quicktest.py runs before I try anything else. I noticed what seem to be some python type problems that I'm hoping you could help with.

Once I got everything installed correctly and was able to run it (python ani_quicktest.py in the example folder) this is the output I got:

van Der Waals correction will be unavailable. Please install ased3
Traceback (most recent call last):
  File "ani_quicktest.py", line 24, in <module>
    ei = mol.get_potential_energy()
  File "/home/cbannan/.local/lib/python3.6/site-packages/ase/atoms.py", line 674, in get_potential_energy
    energy = self._calc.get_potential_energy(self)
  File "/home/cbannan/.local/lib/python3.6/site-packages/ase/calculators/calculator.py", line 618, in get_potential_energy
    energy = self.get_property('energy', atoms)
  File "/home/cbannan/.local/lib/python3.6/site-packages/ase/calculators/calculator.py", line 668, in get_property
    self.calculate(atoms, [name], system_changes)
  File "/home/cbannan/ANI/ASE_ANI/lib/ase_interface.py", line 689, in calculate
    self.nc.set_cell((self.atoms.get_cell()).astype(np.float32), pbc_inv)
AttributeError: 'Cell' object has no attribute 'astype'

After some digging into ase.Atoms and ase.cell.Cell, it became clear that the Cell object does not in fact have a method astype. Looking at other lines of code in ase_interface I noticed that on line 687 this call np.linalg.inv(self.atoms.get_cell())).astype(np.float32) works just fine which made me assume the astype needed an np.arrary instead of an ase.cell.Cell. I decided to try changing the Cell object to an array in line 689 with this change:

self.nc.set_cell(np.arrary(self.atoms.get_cell()).astype(np.float32), pbc_inv)

This appeared to "fix" things in that at least when I reran the script I got different out:

van Der Waals correction will be unavailable. Please install ased3
With Caitlin's change
Initial Energy:  -2078.5028228165684
Optimizing...
Traceback (most recent call last):
  File "ani_quicktest.py", line 31, in <module>
    dyn.run(fmax=0.001)
  File "/home/cbannan/.local/lib/python3.6/site-packages/ase/optimize/optimize.py", line 246, in run
    return Dynamics.run(self)
  File "/home/cbannan/.local/lib/python3.6/site-packages/ase/optimize/optimize.py", line 159, in run
    for converged in Dynamics.irun(self):
  File "/home/cbannan/.local/lib/python3.6/site-packages/ase/optimize/optimize.py", line 125, in irun
    self.atoms.get_forces()
  File "/home/cbannan/.local/lib/python3.6/site-packages/ase/atoms.py", line 728, in get_forces
    forces = self._calc.get_forces(self)
  File "/home/cbannan/.local/lib/python3.6/site-packages/ase/calculators/calculator.py", line 632, in get_forces
    return self.get_property('forces', atoms)
  File "/home/cbannan/.local/lib/python3.6/site-packages/ase/calculators/calculator.py", line 668, in get_property
    self.calculate(atoms, [name], system_changes)
  File "/home/cbannan/ANI/ASE_ANI/lib/ase_interface.py", line 703, in calculate
    self.nc.set_cell((self.atoms.get_cell()).astype(np.float32), pbc_inv)
AttributeError: 'Cell' object has no attribute 'astype'

So now line 703 seems to have the same problem 689 did. So I changed line 703 to:

self.nc.set_cell(np.array(self.atoms.get_cell()).astype(np.float32), pbc_inv)

Finally I was able to run the ani_quicktest.py without any errors and got the energies as:

• Initial Energy: -2078.5028228165684
• Final Energy: -2078.50426605258
  Is it possible these values don't agree with the values in the README because I'm missing the ased3 module?

Initially, I didn't install ased3 because I was under the impression it was optional and I wanted to make sure that I could run the script as is. However, now I've noticed that ased3 also doesn't have a license associated with it which makes me hesitant to try it. Do you know if this same issue will come up if I include ased3?

Thanks for your help in advance!

[isayev]

Hey, @bannanc thanks for giving ANI a try:) Two things.

1. It seems that recent changes in ASE broke the compatibility with our interface:/ We will check it out, but PR is welcome to fix it.
2. This is a mixed-precision GPU code, so there is always some stochasticity in the last few digits because of rounding. Here is a run on my machine:

Initial Energy: -2078.50282282132
Final Energy:-2078.5042660113986

So if 7 digits agree, you are good to go!

[bannanc]

Thanks @isayev I'm not going to bother with ased3 then for now.
I'm happy to put in a PR with the changes I mentioned above.

[UnixJunkie]

Hi from Tokyo @bannanc ,

How did you manage to make the test run?
I am still fighting with it; my latest trial script:
#45

Regards,
Francois.

[bannanc]

Hi @UnixJunkie I managed to get something working back in 2019 for some benchmarking. However, I haven't used it since that summer so I don't remember what I did exactly. Good luck!