/TorchANN

Use torch to train NN potential

Primary LanguageC++MIT LicenseMIT

This repo is no longer under maintaining. It has been replaced by a C++ version which is currently under development.

TorchANN

Use torch to train NN potential

Features

  • Supports DeePMD types of symmetry coordinates.
  • Trains on multi-gpus.

Request:

  1. python version >= 3
  2. pyTorch (necessary: Torch v1.x; optional: Torchvision)
  3. ninja
  4. C++ compiler which supports c11 standard (icpc 2018 and gcc 4.9.4 tested) (Note: The C++ code compiled by g++-7 is much slower and much more memory-consuming than g++4.9. Still do not know why...)

Below are not requested at present but on the road map.

  1. horovod(Optional, future)
  2. mpi4py(Optional, future)
  3. An MPI library (Intel MPI 2018 and OpenMPI 4.0 tested)(Optional, future)

Prepare training data:

box.raw type.raw coord.raw force.raw energy.raw PARAMS.json
Property Unit
Time ps
Length Å
Energy eV
Force eV/Å
Pressure Bar

box.raw contains box of each frame:

 7.710900  0.000000  0.000000  0.000000  8.904000  0.000000  0.000000  0.000000 26.296000 
 7.710900  0.000000  0.000000  0.000000  8.904000  0.000000  0.000000  0.000000 26.296000 
 7.710900  0.000000  0.000000  0.000000  8.904000  0.000000  0.000000  0.000000 26.296000  
 7.667900  0.000000  0.000000 -11.501850 19.921789  0.000000  0.000000  0.000000 21.260799 
 7.667900  0.000000  0.000000 -11.501850 19.921789  0.000000  0.000000  0.000000 21.260799 
 7.667900  0.000000  0.000000 -11.501850 19.921789  0.000000  0.000000  0.000000 21.260799
 ...

Each line contains nine numbers: a1, a2, a3; b1, b2, b3; c1, c2, c3. Take the first line as an exampe:

 7.710900  0.000000  0.000000  0.000000  8.904000  0.000000  0.000000  0.000000 26.296000

means a box of:

vec_a = (7.710900, 0.000000, 0.000000)
vec_b = (0.000000, 8.904000, 0.000000)
vec_c = (0.000000, 0.000000,26.296000)

type.raw contains type of each atom in one frame:

0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 
0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 
0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 
0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1 1 1 1 1 1 1 1 1 1 1 1 
0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1 1 1 1 1 1 1 1 1 1 1 1 
0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1 1 1 1 1 1 1 1 1 1 1 1

Here 0 and 1 represent two different types of atoms. The numbers function as labels of different element, therefore the values could be arbitary.

coord.raw stores the coordinates of atoms in one frame:

 2.939720  6.538850  2.954150  1.850690  6.321660  0.736290  4.276000  8.778080  2.631220  4.612180  7.558540  4.802170  0.551250  8.550070  0.733660  0.655030  6.664390  2.937170  3.113080  0.419850  5.534930  3.123380  8.568410  0.743650  0.778900  0.453260  5.211400  4.408090  6.334930  0.735760  1.849100  0.016200  2.721580  1.857600  6.967150  5.332680  4.405930  1.883460  0.735760  3.121450  4.118100  0.742600  2.949370  2.374500  2.578150  0.194690  2.338610  3.065110  3.308660  5.091720  4.742190  4.441900  4.275650  2.786670  4.261720  3.017710  5.301710  1.778140  4.519680  2.751790  1.900780  2.518370  4.560860  0.914160  4.823890  4.846900  0.555570  4.098240  0.733130  1.845300  1.871980  0.735760  6.991240  6.321660  0.736290  5.691800  8.550070  0.733660  5.741350  6.848900  2.586650  7.341340  7.182960  5.095860  6.951990  0.137590  3.165190  0.294910  5.856450  7.814560  5.704330  2.236160  2.999860  7.062350  4.585110  2.871480  6.699420  2.917990  5.042990  5.583340  4.966420  5.114320  5.696120  4.098240  0.733130  6.985840  1.871980  0.735760  4.022520  7.317680  7.556660  6.571000  7.409080  7.691360  3.024180  0.713790  7.936110  4.204900  8.777400  7.712050  4.180610  2.885990  7.801380  5.499700  0.609370  7.984420  6.806580  8.902100  7.918360  6.806870  2.871500  7.769460  2.977010  2.095440  7.859100  2.948980  5.082750  7.984330  4.150370  4.408660  7.961870  5.495450  2.058400  7.841980  5.390350  5.021750  7.704490  6.816330  4.345000  7.692350  2.911550  6.479290  7.755130  5.447710  6.499890  7.516550  1.665270  7.231910  7.824960  0.419990  0.751450  7.927650  1.528080  8.637960  8.123200  1.703230  2.980040  8.090160  0.454800  2.287580  7.972300  1.807610  4.392440  8.106240 
 3.120500  6.994950  2.949000  1.850690  6.321660  0.736290  4.195500  0.137570  3.207200  4.359900  7.345810  5.352240  0.551250  8.550070  0.733660  0.161560  7.125580  3.070050  3.496400  0.684930  5.298330  3.123380  8.568410  0.743650  0.651680  0.820960  4.879630  4.408090  6.334930  0.735760  1.777200  0.238560  2.589990  1.910160  7.465450  5.420560  4.405930  1.883460  0.735760  3.121450  4.118100  0.742600  3.216250  2.589190  3.767250  0.516070  2.320130  2.712070  2.682760  5.240930  5.003450  4.106400  4.743610  2.826160  4.325630  3.457960  5.460560  1.746680  4.366620  2.877730  1.389450  3.361190  5.324870  0.066800  5.167380  5.431660  0.555570  4.098240  0.733130  1.845300  1.871980  0.735760  6.991240  6.321660  0.736290  5.691800  8.550070  0.733660  5.623340  7.049830  3.068470  6.839820  7.282070  5.414690  6.448320  0.099900  3.361960  0.596430  5.570910  7.522860  5.529760  2.331660  2.991950  7.202050  4.575950  2.920380  6.762590  2.744450  5.309250  5.540470  5.169300  5.068410  5.696120  4.098240  0.733130  6.985840  1.871980  0.735760  4.679570  7.015460  7.830100  7.105320  6.868540  7.626400  3.232950  0.216650  7.622630  4.574330  8.473800  7.798490  4.502490  2.408260  7.686960  5.808430  0.459780  7.765280  7.073010  8.438140  7.512430  7.079220  2.716440  7.780310  3.210720  1.677380  7.711180  3.204360  4.859500  8.017850  4.496510  3.935950  7.855960  5.835410  1.844390  7.614220  5.701180  4.715300  7.780400  6.851490  4.166810  7.765460  3.057940  6.288290  7.714330  5.809890  6.177390  7.881030  1.886420  7.124600  7.360930  0.547770  0.443140  7.540130  1.820000  8.495150  7.339720  1.912260  2.545990  7.607010  0.606200  1.864010  7.730840  2.047170  4.021280  7.747450 
 3.112030  7.005310  2.958420  1.850690  6.321660  0.736290  4.189430  0.128210  3.194720  4.357020  7.343710  5.359230  0.551250  8.550070  0.733660  0.160430  7.126910  3.086360  3.499650  0.679880  5.304050  3.123380  8.568410  0.743650  0.640140  0.834520  4.879190  4.408090  6.334930  0.735760  1.785940  0.238440  2.596630  1.895200  7.473770  5.416660  4.405930  1.883460  0.735760  3.121450  4.118100  0.742600  3.205240  2.606860  3.791300  0.516330  2.322430  2.715500  2.677730  5.251680  5.009540  4.101450  4.751660  2.824440  4.341590  3.471110  5.455930  1.747990  4.374300  2.875750  1.383420  3.370110  5.336510  0.066550  5.169330  5.418750  0.555570  4.098240  0.733130  1.845300  1.871980  0.735760  6.991240  6.321660  0.736290  5.691800  8.550070  0.733660  5.618360  7.051110  3.069130  6.842410  7.290190  5.422980  6.467830  0.098520  3.363100  0.611390  5.577910  7.522050  5.535000  2.348870  2.988000  7.195520  4.571400  2.917060  6.762370  2.742990  5.303560  5.528970  5.146850  5.070090  5.696120  4.098240  0.733130  6.985840  1.871980  0.735760  4.677550  7.015900  7.832900  7.130750  6.857830  7.594210  3.199580  0.202120  7.625500  4.572570  8.443950  7.840930  4.510280  2.369360  7.694390  5.813740  0.444550  7.757600  7.078400  8.421720  7.533060  7.091530  2.705350  7.770630  3.191240  1.681540  7.718300  3.233580  4.841050  7.995790  4.481970  3.915460  7.869820  5.815950  1.851240  7.576770  5.698690  4.722620  7.757330  6.879610  4.164200  7.754710  3.058440  6.288870  7.685320  5.797690  6.189200  7.832260  1.882780  7.107990  7.338960  0.525580  0.449370  7.528100  1.854690  8.507680  7.343310  1.954980  2.546380  7.588540  0.592690  1.857680  7.716280  2.056580  4.034160  7.787930 
 1.282800  2.212310  0.039760 -8.144420 19.347740  4.085240  6.215390  0.757420  1.897690  3.841620  2.212120  0.038910 -10.750040 19.272430  3.795100  1.103990  0.574690  1.689960 -1.272530  2.213110  0.036990  2.237010  0.064850  6.143520  3.735900  0.970690  2.226780  0.001760  4.428220  0.027430  1.882580  1.743410  3.902960 -0.093660  2.609750  2.402980  2.561270  4.428810  0.039120  5.490550  0.553690  4.235460  2.989180  3.294010  2.286810 -2.546050  4.428610  0.028060 -0.011330  1.730800  5.377530  5.376480  2.921220  1.833860 -1.271150  6.635350  0.028910  1.240240  4.054270  4.357950 -2.588110  5.438170  1.959160  1.283760  6.639930  0.036780  4.011410  4.654690  4.378530  0.143110  5.240130  2.399690 -3.828310  6.640530  0.044220 -1.597310  3.576900  4.136180  2.475790  5.723850  2.142770 -2.551110  8.852840  0.039760  0.284490  6.478610  4.392990 -3.646020  7.961560  2.274960  0.007590  8.852840  0.038910  2.681140  7.060980  4.030470 -1.067040  7.859500  2.231120 -5.106550  8.853840  0.036990 -2.096530  6.922060  4.126020  1.448760  8.301170  1.929250 -3.832150 11.068750  0.027430 -5.048350  9.394130  5.193820 -5.223310  9.948230  2.906820 -1.272640 11.069340  0.039120 -2.544470  9.557530  4.570610 -2.395560 10.660440  2.501560 -6.379960 11.069140  0.028060  0.526200  8.691680  4.205300 -0.299950 10.555280  2.270450 -5.105060 13.275880  0.028910 -5.682980 12.034100  4.572380 -6.127530 12.270750  2.124080 -2.550150 13.280460  0.036780 -3.199250 12.132430  4.255530 -4.033210 13.526430  2.371890 -7.662230 13.281060  0.044220 -0.608960 11.823160  4.577450 -1.622170 12.987430  2.335600 -6.385020 15.493370  0.039760 -6.075000 16.122790  6.224120 -5.810030 16.161570  2.301890 -3.826320 15.493370  0.038910 -6.892850 13.892490  6.422450 -4.842590 14.905860  4.223450 -8.940470 15.494370  0.036990 -7.352800 13.806640  3.893940 -2.241940 14.801190  4.087140 -7.666170 17.709470  0.027430 -5.112170 17.253040  4.365730 -8.351570 15.335220  2.393300 -5.106670 17.710070  0.039120 -3.938600 16.631870  6.654630 -3.551740 15.925770  2.191850 -10.213990 17.709870  0.028060 -8.542710 15.633880  6.129220 -2.371150 17.300640  4.789790 -8.939080 19.916610  0.028910 -7.364190 17.125720  3.924720 -9.548590 17.661200  2.565380 -6.384180 19.921190  0.036780 -4.191650 18.890540  5.880700 -7.126000 18.479830  2.164090 -11.496140 19.921590  0.044220 -7.801300 17.994070  6.020740 -4.499310 18.262000  2.262530 -3.650700 13.958980  6.649480 -4.877950 14.312180  6.939930 -1.296120 13.068070  6.502380 -2.249570 13.906730  6.584870  0.084850 10.835580  6.731200 -0.166780 12.147960  6.408330  0.441670  8.388820  6.044280  0.663110  9.657400  6.417660 -1.409550  6.299840  5.961230 -0.322650  7.146510  6.197500 -0.571820  3.610700  5.835240 -1.068860  4.868430  6.011910 
 1.282800  2.212310  0.039760 -8.148690 19.344870  4.070540  6.225990  0.756000  1.905160  3.841620  2.212120  0.038910 -10.747680 19.264990  3.791540  1.102660  0.556370  1.692410 -1.272530  2.213110  0.036990  2.250140  0.057370  6.140490  3.745820  0.971250  2.229880  0.001760  4.428220  0.027430  1.898430  1.737300  3.897810 -0.083840  2.602610  2.400730  2.561270  4.428810  0.039120  5.487000  0.538500  4.236090  2.992080  3.294560  2.276170 -2.546050  4.428610  0.028060 -0.010970  1.742640  5.377260  5.377950  2.925160  1.820120 -1.271150  6.635350  0.028910  1.238590  4.037450  4.347850 -2.585940  5.438230  1.939110  1.283760  6.639930  0.036780  4.012150  4.672020  4.370880  0.151550  5.245380  2.397580 -3.828310  6.640530  0.044220 -1.604380  3.580240  4.141270  2.471720  5.732290  2.143790 -2.551110  8.852840  0.039760  0.288490  6.477280  4.379320 -3.634390  7.953540  2.268980  0.007590  8.852840  0.038910  2.689060  7.059320  4.015330 -1.064390  7.861530  2.225750 -5.106550  8.853840  0.036990 -2.089110  6.921490  4.128250  1.443540  8.298490  1.939770 -3.832150 11.068750  0.027430 -5.050900  9.391510  5.208850 -5.229770  9.951400  2.873830 -1.272640 11.069340  0.039120 -2.555880  9.552630  4.586930 -2.400290 10.658570  2.509890 -6.379960 11.069140  0.028060  0.540580  8.688430  4.193650 -0.284530 10.547230  2.264770 -5.105060 13.275880  0.028910 -5.685160 12.048420  4.583060 -6.134660 12.275460  2.125300 -2.550150 13.280460  0.036780 -3.195580 12.135820  4.247110 -4.030020 13.536130  2.380980 -7.662230 13.281060  0.044220 -0.612330 11.811200  4.557520 -1.623020 12.993830  2.339420 -6.385020 15.493370  0.039760 -6.062170 16.127660  6.214140 -5.803260 16.166880  2.310440 -3.826320 15.493370  0.038910 -6.906190 13.884100  6.411420 -4.831400 14.908190  4.230400 -8.940470 15.494370  0.036990 -7.345330 13.797330  3.893770 -2.216480 14.790410  4.093970 -7.666170 17.709470  0.027430 -5.111920 17.252190  4.371340 -8.349100 15.331590  2.411990 -5.106670 17.710070  0.039120 -3.935930 16.646930  6.667140 -3.558240 15.916780  2.187180 -10.213990 17.709870  0.028060 -8.554650 15.641850  6.121360 -2.372310 17.306310  4.797400 -8.939080 19.916610  0.028910 -7.349680 17.129930  3.935300 -9.538110 17.652070  2.577920 -6.384180 19.921190  0.036780 -4.194950 18.897160  5.893140 -7.143260 18.488880  2.164200 -11.496140 19.921590  0.044220 -7.803960 18.005790  6.012380 -4.509860 18.277330  2.272730 -3.645530 13.984950  6.631980 -4.857030 14.309520  6.972420 -1.287960 13.066950  6.466180 -2.275770 13.882910  6.566400  0.078050 10.822940  6.745460 -0.158960 12.147450  6.370870  0.469150  8.408800  6.060470  0.670800  9.630560  6.429520 -1.427280  6.276960  5.974540 -0.293150  7.132910  6.206960 -0.546530  3.599010  5.807840 -1.074520  4.877300  6.014470 
 1.282800  2.212310  0.039760 -8.230690 19.019460  4.191020  6.688540  1.229780  2.693880  3.841620  2.212120  0.038910 -11.205490 19.600690  3.884580 -9.725040 19.901900  2.167700 -1.272530  2.213110  0.036990  1.796650  0.427850  5.898650  4.361010  1.171280  2.140280  0.001760  4.428220  0.027430  2.812280  1.870350  4.104780  0.670480  2.594600  2.072830  2.561270  4.428810  0.039120 -6.024260 19.145710  4.453130  2.900640  3.249350  2.133160 -2.546050  4.428610  0.028060 -0.058550  1.971560  4.988910  5.294730  3.477730  2.036330 -1.271150  6.635350  0.028910  1.592160  4.147660  4.050620 -2.010610  5.996470  2.375830  1.283760  6.639930  0.036780  3.655260  4.968680  3.821450  0.002330  5.093610  2.337150 -3.828310  6.640530  0.044220 -1.709770  3.989360  4.486400  2.658990  5.746690  1.786200 -2.551110  8.852840  0.039760  1.141990  6.351940  3.909630 -3.866510  8.105690  2.098420  0.007590  8.852840  0.038910  2.814710  8.078220  4.143360 -1.042550  8.262650  2.423140 -5.106550  8.853840  0.036990 -1.836460  7.012820  4.681370  1.499740  7.877290  2.236170 -3.832150 11.068750  0.027430 -5.320420  9.393190  6.032720 -5.007720 11.017230  2.172680 -1.272640 11.069340  0.039120 -3.036320  9.463990  4.399410 -2.653480 10.103650  2.156260 -6.379960 11.069140  0.028060  1.043820  9.575860  3.994380  0.773510  9.888170  1.845220 -5.105060 13.275880  0.028910 -5.345270 12.049380  4.636270 -6.640890 12.660660  1.901410 -2.550150 13.280460  0.036780 -2.984460 12.045600  4.173510 -4.168720 13.361730  2.048480 -7.662230 13.281060  0.044220 -0.859590 10.849170  3.080010 -1.259130 13.093070  1.960140 -6.385020 15.493370  0.039760 -6.027580 15.926010  6.397670 -5.446360 15.904470  2.212440 -3.826320 15.493370  0.038910 -7.238500 13.656720  6.047380 -4.497110 14.788910  4.229900 -8.940470 15.494370  0.036990 -6.620310 14.491940  3.800490 -1.422150 13.824490  4.323800 -7.666170 17.709470  0.027430 -4.670140 17.361140  4.668140 -7.995920 15.300010  1.818440 -5.106670 17.710070  0.039120 -3.558320 17.076160  7.221470 -3.056660 15.786700  2.848690 -10.213990 17.709870  0.028060 -8.384040 15.846320  6.269750 -9.586970 17.827510  5.289330 -8.939080 19.916610  0.028910 -7.412600 16.583120  4.004300 -9.236230 17.628890  2.403090 -6.384180 19.921190  0.036780 -4.553130 19.688670  6.813930 -6.950430 18.736620  2.207580 -11.496140 19.921590  0.044220 -7.202440 18.237440  6.204060 -4.652170 18.279890  2.274290 -3.751600 14.174990  7.055620 -4.431590 15.326950  6.839970 -1.615490 12.858380  6.287650 -2.906780 13.362640  6.211320 -0.035980 11.057980  7.199670 -0.695890 12.186310  6.715270  0.458630  8.570790  6.287430  0.118630  9.813900  7.038840  0.128780  6.210620  5.416850  0.146080  7.391040  6.223920  0.046440  3.540910  5.710960 -0.150760  4.794550  5.663420

The coordinates of atoms of one frame must be put in a single line. The number of coordinates in one line should be an integral multiple of 3.

force.raw stores all the forces acting one each atom:

 1.578933  1.234426 -0.215541 -0.071002  0.040477  0.067627  0.643874  0.817782  1.804038  0.140418 -0.602209 -0.370993 -0.314510 -0.026842  0.110945 -1.154725  0.746525  0.061570  0.393245  0.405126 -0.445063 -0.546614 -0.273987 -1.365228 -0.566814 -0.021014 -0.522211 -0.414802 -0.242908 -0.201699 -0.459542  0.251030  1.150876  1.202525  0.732509  0.011343  0.379316 -0.138404  0.048728  0.034608  0.015164 -0.299721  1.103928 -0.432312  0.801473  1.601354 -0.239050 -0.940189 -0.540637  0.017116  0.299377  0.260585  0.707602  0.428521 -0.315947 -1.127156 -0.358458  0.425075 -0.857528 -0.386716  0.349725  0.288721  1.581775 -1.103007 -0.194441  0.916294 -0.158656 -0.179477 -0.187845 -0.928670 -0.368195 -0.928392  0.490933 -0.205810 -0.385709  0.481216  0.184793 -0.151729  0.092444 -0.272774  1.704720 -1.663963 -0.161329 -0.126070 -0.116684 -0.755480  0.146262  0.810661 -0.273865 -0.229822 -1.939151 -0.414491 -0.357924 -0.980662  0.231375  0.181797  0.773947 -0.348447  0.497017  1.003129  1.214704 -0.319780  0.205269 -0.038593  0.012205 -0.036654 -0.091480  0.297419  5.058554 -1.428782  0.775975  2.200140 -1.459068 -0.361377 -2.407013 -4.248050 -0.665428 -0.575754  0.853238  1.125683  2.065092 -0.547921  0.604318  0.083711  1.084524 -2.235678 -1.765208 -2.241402 -0.987774  0.358767  1.280954  0.795373 -1.306879  3.183898  1.687930  1.458121  1.197451 -0.389000  2.504864 -3.851273 -1.450699 -0.239958  2.453266  0.149779  3.342770  1.194183 -0.583308 -4.377134 -0.359696 -0.217666 -0.603367  3.708960  0.266577 -2.502022  3.753318  1.391766 -1.291656  0.534799  1.061597 -1.012516 -2.187646  0.958831  3.057021  3.344461 -2.691689  0.885689 -1.152736 -1.099464 -1.586589 -3.697602 -0.149795 -3.991565 -1.113915 -0.300073 
-0.796242 -0.523549  0.185499 -0.062476  0.019478  0.585654 -0.591477  0.782278 -0.723076  0.183543 -0.499285 -0.767653 -0.555822 -0.230631 -0.231444  1.854631 -0.441515  0.069979 -0.588252  0.556004  0.476570  0.469344 -0.162928 -0.066125  0.198139 -0.048659 -0.572105  0.097700  0.009616  0.517998  0.009922 -0.051887  1.559134 -0.557552  0.454264 -1.518897 -0.082722 -0.034253  0.498334 -0.085702 -0.279068 -0.059516 -0.981297 -0.981662 -1.475982 -0.357170 -0.214046  0.987730  0.626894  0.711129 -0.658134  1.121814  0.019328  0.409214 -0.502892 -1.571833  1.228355  0.718453  0.601519  0.079543  0.866637 -1.768025 -0.105128  0.351306 -0.123173 -2.307382 -0.006363 -0.038363 -0.018239  0.331385  0.175154 -0.016156 -0.106709 -0.010445  0.523260 -0.008946  0.042684  0.581756 -1.779715 -2.622593 -0.294725  0.219354  1.115243 -0.199485  1.988320  2.411973  0.634199  2.206710 -2.132733  1.468926  0.235352  0.412879  0.096827 -1.604230  0.370091  0.337585 -0.394607 -0.005059 -0.781608  0.183999  1.502932 -1.548655 -0.052225  0.037002  0.586766 -0.398427 -0.171683  0.089223 -4.439925  0.074448 -0.151024 -3.185291  3.377960 -0.056827 -0.924372  1.904382 -0.366016  0.027128  0.868506 -0.678699 -1.005852  3.768328  0.612910  0.740465 -5.918768 -1.243535 -1.042473  1.583030  1.580949 -1.443538 -1.733673 -1.333074 -0.135535  0.807772 -0.870710 -0.070968 -2.905022 -2.537167 -2.087069  1.650645 -0.115878 -0.045157  0.935655  1.580355 -5.196712  2.886700  1.324522  8.778878 -4.780138 -0.463419  4.555694  0.856648  0.924607 -1.251237  2.181023 -0.873816  0.899663 -5.153917  1.620108  1.283406 -3.122272 -0.723248  2.007581  3.229544  1.833105  1.958434  0.507524  1.489690 -0.090309  1.400328 -0.771157 -1.484663  0.181305 -0.362305 
-0.817296 -0.652533  0.163111 -0.059358  0.021358  0.590996 -0.345903  0.946611 -0.608899  0.187809 -0.534664 -0.822442 -0.527495 -0.215686 -0.186113  1.866464 -0.505091  0.051047 -0.561904  0.537500  0.343798  0.461420 -0.163774 -0.069875  0.262378 -0.079401 -0.545043  0.095998  0.015035  0.521154 -0.086009 -0.015147  1.492100 -0.512326  0.477535 -1.619429 -0.084064 -0.031708  0.506260 -0.087614 -0.285133 -0.063650 -0.953855 -1.064325 -1.558479 -0.333947 -0.224804  0.952749  0.637595  0.669916 -0.672518  1.135042  0.025966  0.416070 -0.794508 -1.999859  1.338709  0.643183  0.604309  0.090510  0.926629 -1.843785 -0.101314  0.245574 -0.027807 -2.213285 -0.002366 -0.038619 -0.019653  0.327081  0.169383  0.003558 -0.105657 -0.002898  0.533662 -0.003105  0.042039  0.580295 -1.721280 -2.549653 -0.268653  0.176477  1.025700 -0.268717  1.705056  2.430831  0.584684  2.350041 -2.680709  1.431433  0.171413  0.325353  0.119261 -1.563371  0.380252  0.335705 -0.417948  0.004642 -0.760573  0.613002  2.070197 -1.619411 -0.056211  0.037681  0.586662 -0.391148 -0.174358  0.099096 -4.692153 -0.118222 -0.250013 -4.212673  3.728529  0.178739  0.385628  2.236060 -0.047788 -0.319671  2.177878 -1.249592 -2.265353  4.918514  0.670961  0.348248 -5.486467 -1.039963 -1.042491  1.937924  1.086461 -1.938951 -1.102159 -1.398615  0.673948  0.055069 -0.933922 -0.646088 -2.427474 -2.150145 -0.868974  2.108375 -0.530599  1.369426 -0.612049  1.934326 -3.801038  1.547223  1.209834  6.584486 -3.840466 -0.433028  4.842382  0.371692  1.116928 -0.244437  1.303990 -0.598807  1.177354 -4.024399  1.807698  2.188856 -3.463539 -0.580440  0.470605  1.966497  1.582409  0.240570  0.623809  1.842315  0.822148  1.654484 -0.756428 -1.423327 -0.356307 -0.839817 
 0.152379  0.613277 -0.058308 -0.897083  0.856596  0.105503  0.334027 -1.334169  1.272313 -0.305254 -0.016641  0.040809 -0.190677  1.866219  0.452211  1.577329  0.555882  3.786668  0.976581 -0.143425 -1.733374 -0.625626 -0.102191 -0.725882 -0.217117 -0.169766 -0.033862  0.077544 -0.079065  0.340263  0.548624 -0.665756 -0.393646  1.528669  0.590361  0.378691 -0.088882 -0.092194  0.146344  0.455685 -0.175066 -0.175418 -0.437806  0.271951  0.296939 -0.088569 -0.470948 -1.743660 -0.495757  0.268458  0.143309 -1.527926  1.233384  1.980152  0.029917  0.016319  0.394716 -0.066215  0.153766 -0.304629  0.087430  1.010728  1.948993 -0.134672 -0.103083  0.357260 -0.944634  0.487211 -0.978761 -0.556438 -0.208427 -0.572614 -0.030393  0.031472  0.432354  0.310598 -0.069330 -0.505895  0.221800 -0.069324 -0.248828 -0.001616  0.137280  0.546532  0.455726 -1.291152 -0.230206  0.424279  0.198545 -0.276129 -0.218396  0.095543  0.162350  0.090377  0.084460  0.529220  0.189629  0.364027 -0.472699  0.678549  0.346571 -0.457276 -1.227344 -0.321052  0.384644 -1.280193 -2.023672  0.137649 -0.009053 -0.002567  0.735383  0.233382  0.303002  0.256989  1.041804  2.081307  0.384627 -0.170678  0.044708  0.321619 -0.321648 -0.247037 -0.354585 -2.675377 -0.115219 -0.562595  0.011935 -0.246157  0.284450 -0.151777  1.710730 -2.048125  2.762064 -0.429745  0.015279  0.000372  0.078057  0.370645  0.154256 -0.098233 -0.467515 -0.212572  0.297391  0.210511 -0.112933  0.056877  0.366115 -0.279290 -0.258712 -0.273469 -0.040857 -0.236936 -0.324059  0.012884  0.061984  0.668089 -0.681765 -1.016948 -1.654924  0.533397  0.043479  0.042771 -0.120976 -0.158869  0.196715 -1.020499 -0.393880 -0.060540 -1.284750 -0.350858 -0.334624 -0.183288 -0.299278 -1.021460  1.479499 -0.492139  0.217691 -0.087148 -0.728417  0.088284 -0.132695  0.010846  0.261590  0.806519  0.900883 -0.500462 -0.282761 -0.076658  0.065013 -0.177393 -0.353062 -0.133133  1.034123  0.819740  0.242660  0.271950  0.645691 -0.732696 -0.213595 -0.163600 -0.069917  1.189144 -0.301030 -0.355754  1.264173  0.039560  1.255962  0.005125 -0.037874  0.480094 -0.846259  0.112613 -0.386861  0.037387 -0.804543  1.171021  0.702172 -0.311162 -0.296819 -0.919822 -1.998962  0.550415  0.680693  0.082859 -1.359929 -0.524721 -0.463025 -0.381010  0.061446  0.460643  0.027697  0.768915  0.577939 -1.081449 -1.666534 -1.170111 -2.583686  0.078198 -0.141880  0.658917  0.006952  0.283289 -0.414111 -0.343341  1.106108  0.707474  0.490315  0.051820 -1.398858  4.174293 -2.809781  0.434863 -2.846698  1.269675 -0.093278  1.113734 -1.170431 -0.893177 -2.134888  0.979196  1.176145 -2.866124 -4.408338  1.408871 -1.002744  2.428141  1.073363  2.314712 -1.319112 -0.227067  1.413341  3.614960 -0.107487 -1.032920  3.096624  0.452622  0.882969 -2.389687 -0.867614 
 0.148434  0.560317 -0.002643 -0.962767  0.901991  0.217425  0.253034 -1.276108  1.197737 -0.317957 -0.033304  0.018652 -0.203359  1.906536  0.542085  1.629674  0.617806  3.782880  1.028273 -0.203842 -1.798066 -0.661463 -0.049313 -0.716495 -0.218329 -0.110371 -0.070843  0.078392 -0.076469  0.340712  0.504924 -0.685201 -0.368023  1.457771  0.635587  0.387506 -0.095616 -0.077867  0.143608  0.465730 -0.168294 -0.148920 -0.446678  0.284188  0.324667 -0.091093 -0.531053 -1.956608 -0.416204  0.113098  0.084601 -1.460945  1.189382  2.157036  0.038805  0.023725  0.379155 -0.107769  0.183357 -0.259092  0.084841  1.103940  2.155343 -0.133863 -0.095671  0.355996 -0.943658  0.490075 -0.973163 -0.608664 -0.273022 -0.637340 -0.045199  0.040454  0.406581  0.312693 -0.084242 -0.571660  0.265827 -0.151986 -0.340197  0.002765  0.142045  0.545663  0.389979 -1.248499  0.018609  0.402847  0.229377 -0.283161 -0.204401  0.093912  0.166354  0.090181  0.143224  0.653965  0.171843  0.354307 -0.431491  0.638721  0.326090 -0.393412 -1.194811 -0.313412  0.413993 -1.283902 -2.167510 -0.063521 -0.007779 -0.002719  0.734992  0.232842  0.317139  0.069125  1.155151  2.179063  0.643258 -0.174513  0.047693  0.330801 -0.308224 -0.222389 -0.411132 -2.361148 -0.145916 -0.571141  0.015015 -0.239913  0.284675 -0.236238  1.627564 -1.751486  2.461089 -0.430147  0.018588  0.003590  0.077950  0.384099  0.169668 -0.140922 -0.460442 -0.163754  0.320132  0.200693 -0.111988  0.052243  0.367971 -0.322245 -0.270692 -0.242254 -0.089797 -0.318377 -0.363851  0.011561  0.061780  0.666189 -0.706631 -1.074586 -1.719392  0.533679 -0.004045  0.000007 -0.113715 -0.149957  0.229229 -1.150642 -0.529354  0.027852 -1.373712 -0.349133 -0.382305 -0.187139 -0.305412 -1.083712  1.593133 -0.255165  0.250448 -0.083714 -0.718757  0.098109 -0.120454  0.009011  0.316124  0.808920  0.930697 -0.514884 -0.400588 -0.063353  0.084756 -0.169907 -0.361978 -0.138097  1.177720  0.897157  0.225010  0.329784  0.705031 -0.868253 -0.197963 -0.159943 -0.031489  1.199750 -0.314343 -0.327856  1.383445  0.145203  1.321161  0.006057 -0.040387  0.493440 -0.682947  0.105347 -0.367825  0.022549 -0.847727  1.204964  0.713054 -0.300531 -0.307642 -0.962138 -1.998432  0.438472  0.622121  0.025868 -1.414106 -0.495825 -0.437166 -0.331091  0.073267  0.478818 -0.054649  0.826504  0.456991 -1.134286 -1.705667 -1.158421 -2.652069  0.104426 -0.185475  0.734322 -0.021889  0.181019 -0.485735 -0.471106  0.761333  0.799455 -0.292442  0.322586 -1.414326  3.276247 -2.058144  0.515312 -1.256701  0.674120 -0.232270  1.198644 -0.886044 -1.129732 -1.862400  0.056626  1.464781 -3.444370 -6.911232  0.577585 -0.897681  5.285204  1.742128  3.051212 -0.148005 -0.180469  0.935678  3.150278 -0.457562 -1.421818  4.025359  0.692159  1.318267 -3.645178 -1.112869 
 0.604129 -0.446541 -1.510948 -0.255516  2.185871 -1.438261 -1.089363 -0.502001  0.017567 -0.050338  0.102618 -0.046915  0.240426 -0.241478  1.638313  0.169258  1.056953  0.138703  0.063813 -0.053133  0.484351 -0.285654 -0.561936 -0.500775 -0.536307 -0.004420  0.462283  0.004941 -0.159551  0.129301 -0.119408 -0.498052 -0.157996 -0.437948  0.928794  1.482709 -0.204415 -0.889466 -1.355421  1.327296  1.622196 -0.493756  0.319188  0.083339 -0.072884 -0.108188  0.653705 -0.736878  0.226267 -2.814005 -1.066833  0.680590 -0.413807  1.391451  0.043025  0.104937  0.489517 -1.092236 -2.045251  0.330711 -1.689034  0.354184  0.052432 -0.401946  0.084886 -0.086006  1.688789  0.528252 -0.161052  0.837081 -0.849152 -0.647552  0.601468  0.272110 -0.465326  0.026296 -0.011105 -0.759388 -0.390289  0.473680  2.217674  0.114215 -0.146571  0.150502  0.714382  1.075012 -0.940209  1.178650 -0.608960 -0.577866 -0.619387 -0.795773 -0.969086  0.806992 -0.952452  0.804657  0.014734 -0.215467  0.134403  0.015499  0.123398  0.363599  0.196808 -0.022194 -0.657214 -0.532535 -1.156623 -1.087497  0.261708 -0.023998  0.250519 -0.021728  0.319106 -0.163740  1.637334 -0.059672  0.967353 -0.008216  0.046431  0.499771  0.159876  0.319661  0.122850 -1.373661  0.617772 -1.233912  0.231958  0.540885 -1.142038 -0.175369  0.260146  1.406688 -1.080864  1.018100 -0.551300 -0.391932  0.232666 -1.028707 -0.410523 -0.023198 -0.514709  0.665723  0.162449  1.784089 -0.470173  0.088776 -0.303003 -0.293751 -0.121912 -0.256853  0.828042  0.382136  1.508288 -0.249278  0.531968 -1.732841  0.777462  1.454174  1.433103 -1.265958 -0.477175  1.384789  0.214531 -0.052006 -0.264750 -1.433891  0.630055 -0.930988 -0.028915  0.227892 -0.132433 -0.038210  0.021518  0.585136  0.533226  0.520241 -0.367643  0.908587 -0.654266  1.377168 -2.024649  0.491440 -3.149453 -1.314839 -1.765886 -0.386164  0.055244  0.792766 -0.249723 -0.031570 -0.103004  0.443306  0.946256 -0.874651 -0.078292  1.571202  0.203816  3.845335 -0.080324 -0.218581 -0.160027 -0.233690 -0.090114 -1.159172  1.069336  0.269720 -0.874665 -0.179478 -0.097465  0.543894 -0.436304  0.446623 -0.461944 -2.004718 -1.431187  1.273544  0.414113  0.090644 -0.515694 -0.461012  1.103661  0.099350  0.163302 -0.579699 -0.180375  0.139206 -0.014869  0.220322  0.373043 -0.665017 -1.134409 -0.240903 -0.669387 -0.515301  0.028914  0.015201  0.536385 -0.274368 -0.683104  0.507355  0.730804 -0.080083  0.094654  1.306294  0.823134 -3.614189  2.390342 -2.110590  1.592846 -6.552589  3.671409 -1.376002 -0.369516  0.514739  1.217670 -1.539311  0.524965 -1.409590  6.522577 -4.989664  3.275281  1.228616  9.317260  3.428469  1.016093 -2.258229 -2.242847 -2.182574  0.118627  3.094190 -1.634932 -8.286061 -2.385627 -0.394981  0.902253  0.417109  0.922469  3.393731  0.073624

The format of force.raw is the same as coord.raw.

energy.raw contains energy of each frame:

-321.48014356
-320.22585212
-320.18770483
-379.73225706
-379.13588315
-374.89802095

PARAMS.json contains parameters for pre-processing and training

{
    "cutoff_1": 7.700,
    "cutoff_2": 8.000,
    "cutoff_3": 0.000,
    "cutoff_max": 8.000,
    "sym_coord_type": 2,
    "batch_size": 1,
    "stop_epoch": 1,
    "num_filter_layer": 3,
    ......
}

How to run the training

STEP 1: Make C executable for data pre-processing

cd ./Torch-ANN/c
#Adapt the makefile for your computer
#For example, change the -g flag to -O2
make partial

STEP 2: Pre-process input data

cd ../tests/test_2 #The input data for test is under this directory
#Modify the PARAMS.json for your dataset
#The [root path of TorchANN]/c/GENERATE_P will not calculate descriptors while [root path of TorchANN]/c/GENERATE calculates all the things before training.
../../c/GENERATE_P > log

The C code will convert the raw data into symmetry coordinates needed by training and some bin files will be stored in the ./ directory:

ALL_PARAMS.json,COORD.BIN, ENERGY.BIN, FORCE.BIN, N_ATOMS.BIN, SYM_COORD.BIN, TYPE.BIN, ...

The file all_frame_info.bin.temp could be deleted and should not affect the training process in the current version.

STEP 3: Run python script to train

export CUDA_VISIBLE_DEVICES=0,1 #For example, I want to use GPU0 and GPU1
#Current tests show that the no_mpi version works the most stable and efficiently.
../../python/no_mpi/train_noclassify_nompi.py

During running, checkpoint will be saved to ./freeze_model.pytorch.ckpt. After the training finished, the model will be saved as ./freeze_model.pytorch. You could copy the checkpoint or saved model to ./freeze_model.pytorch.ckpt.cont to continue if the training stops unexpectedly.

STEP 4: Run test script to evaluate the model

#Make sure that ./freeze_model.pytorch exists
../../python/no_mpi/test_noclassify_nompi.py

If you want to test your own data, you need to do STEP 2 AGAIN in your own data's folder, then run the test script.

Parameters in ALL_PARAMS.json

In the current version the read_parameters() function has not been fully completed. Some of the parameters involved in the data pre-processing procedure need to be modified through the source code. Remember to rebuild the C code after modifying any .c file

  • cutoff_1, cutoff_2, cutoff_3, cutoff_max
    • For DeePMD-type symmetry coordinates, cutoff_1 and cutoff_2 (or cutoff_max) correspond to rcs and rc in its paper.
  • filter_neuron, fitting_neuron, axis_neuron
    • filter_neuron and fitting_neuron are two arrays describing the filter network and fitting network in DeePMD's paper.
    • axis_neuron is the number of columns of the G^(i2) matrix in DeePMD's paper.
  • start_pref_e, limit_pref_e, start_pref_f, limit_pref_f
    • The total loss is calculated by: loss_tot = pref_e * loss_e + pref_f * loss_f. These four parameters determines the amount of contribution of energy E and force F to the total loss function loss_tot.
  • stop_epoch, batch_size
    • stop_epoch determines how many epoches of optimization it will run.
    • For train_noclassify_nompi.py, all the frames are mixed together. Some tests show that if frames with different numbers of atoms are in the same batch, the convergence may become slow, therefore setting batch_size to 1 may be an option. However, too large batch_size can also lead to terrible convergence. You should do your own tests.
  • start_lr, decay_epoch, decay_rate
    • start_lr is the initial learning rate. Learning rate will change its value every decay_epoch epoches by multiplying decay_rate. If decay_rate is larger than 1, then a warning will show up.
    • start_lr will be multiplied by sqrt(NUM OF PROCESSES) during the training in the MPI enabled version(Currently not available).
    • start_lr will remain untouched in the no_mpi version no matter how many GPUs are used.
  • check_step, output_epoch, save_epoch
    • Force of one random frame will be output every check_step batches.
    • Loss of force and energy per atom will be output every output_epoch epoches.
    • Every save_epoch epoches, a checkpoint will be stored to ./freeze_model.pytorch.ckpt
    • The optimizer used in the training is Adam optimizer. Modify train_noclassify_nompi.py to change the optimizer.