Closed thegodone closed 6 years ago
Thanks for the comments Guillaume. Apple's version of Clang is a source of persistent frustration. I will add some more macros to deal with Sierra.
If you would like a pre-trained network for CHNO Place the three files in https://drive.google.com/drive/folders/14QioAgc-1IbKxhFMtQjXH4uFme2mekCP?usp=sharing into /TensorMol/networks.
You can then evaluate periodic or aperiodic forces following the examples in /samples/test_neb.py Ie: python samples/test_neb.py
Best- John
Dear John,
I follow instructions in the supporting file of your article. It's working fine.
I would like to use your model to explain GCMS fragmentation based on bond NRJ (my molecules are mostly CHNO).
I will like to use RDKit 3D generated molecules (or corina).
Did you check your model error on those type of "none 3D ab initio optimized" molecules ?
Thanks to sharing your model.
Best regard,
Guillaume
2017-11-11 16:45 GMT+01:00 John Parkhill notifications@github.com:
Thanks for the comments Guillaume. Apple's version of Clang is a source of persistent frustration. I will add some more macros to deal with Sierra.
If you would like a pre-trained network for CHNO Place the three files in https://drive.google.com/drive/folders/14QioAgc- 1IbKxhFMtQjXH4uFme2mekCP?usp=sharing into /TensorMol/networks.
You can then evaluate periodic or aperiodic forces following the examples in /samples/test_neb.py Ie: python samples/test_neb.py
Best- John
— You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub https://github.com/jparkhill/TensorMol/issues/8#issuecomment-343673726, or mute the thread https://github.com/notifications/unsubscribe-auth/ABIbYnlhsZApV9bT1Mfyn3VIh2wDHHaWks5s1cD9gaJpZM4QacI6 .
Guillaume- That's an interesting idea. The bond neural network was trained on Ab-Initio minimum geometries, because it's all we had available at the time. It won't do very well if the bond lengths are far from minimum. You easily could use the BP network I just provided to optimize molecules with high-throughput if the bond energies that you get from the old network with the geometries you feed in seems poor.
Since the last paper we've put together better datasets, that we could use to make bond energy predictions away from equilibrium. Kun simply hasn't had the time to train the bond network with the new data. Don't hesitate to keep in touch for assistance or collaboration.
Best- John
I receive to compile on sierra 10.12 by changing the SH.hpp file: replace those two lines:
include <tr1/array>
using namespace std::tr1; by this:
include
//using namespace std::tr1;
in this bloc of conditions:
ifdef clang
if __clang_major__ >= 7
include
using namespace std;
else
include
include
//using namespace std::tr1;
endif
else
include
include
using namespace std;
endif