Yes. That's where I got my initial try. But, When I change the descriptor for the different element, the amp does not start to train.For example, I keep the descriptor for 'H' unchanged as default and delete some vector for 'C', after the amp interprete the fingerprints it hang there,and the error file, it shows:Step Time Loss (SSD) EnergyRMSE MaxResid===== =================== ============ ============ ============Traceback (most recent call last):File "/gpfs/runtime/opt/python/2.7.
3/lib/python2.7/runpy.py", line 162, in _run_module_as_main"__main__", fname, loader, pkg_name)File "/gpfs/runtime/opt/python/2.7. 3/lib/python2.7/runpy.py", line 72, in _run_codeexec code in run_globalsFile "/gpfs_home/xc13/amp/amp/ model/__main__.py", line 113, in <module>lossprime=lossprime)File "/users/xc13/amp/amp/model/__ init__.py", line 517, in get_lossself._send_data_to_fortran()File "/users/xc13/amp/amp/model/__ init__.py", line 475, in _send_data_to_fortranself.d)File "/users/xc13/amp/amp/model/__ init__.py", line 1083, in send_data_to_fortran_fmodules.fingerprint_props. raveled_fingerprints = raveled_fingerprintsValueError: setting an array element with a sequence.--On Mon, Jul 31, 2017 at 10:25 AM, El Khatib Rodriguez, Muammar <[log in to unmask]> wrote:I haven't stumbled upon the problem you described on 1. But, regardingOn Mon, Jul 31, 2017 at 9:56 AM, Chen, Xi <[log in to unmask]> wrote:
> Hello all,
> I am using Amp Gaussian training for a potential surface for a gas phase
> combination reaction (with 'C' and 'H' two elements).
> However, when I try to test different parameters in the gaussian descriptor
> and visualize, I came through some problems.
> 1. For some reason, I can only plot the node information ('NodePlot'
> fucntion) for 'H' element, when the code start to plot "C" element, it
> always gave me the following error while calling 'violinplot' function:
> line 90, in _raise_linalgerror_singular
> raise LinAlgError("Singular matrix")
r: Singular matrix
> I also came through the same problem when I using the 'FingerprintPlot'
> function. Tried less images (using 2 images), got the same error.
> I am wondering have you ever came through this problem and know why this
> happens and fix it?
> 2. Gaussian descriptor can be specified by:
> calc = Amp(descriptor = Gaussian(Gs = Gs)).(Gs is the dictionary)
> For the customized Gs, could we have a different number of Gaussian feature
> subvectors for different elements? More broadly, I am wondering is there any
> rules that I should stick to when I manipulate the descriptor?
your second question, you may take a look at this page in Amp's
There you will find an example on how you may add more elements to
your fingerprint vectors.
Muammar W El Khatib Rodriguez
Postdoctoral Research Associate
Brown University School of Engineering
184 Hope Street
Providence, RI, 02912, USA
Xi CHENGraduate StudentDepartment of ChemistryBrown UniversityProvidence