Comments (2)
Hi,
There has been a lot of changes to the defaults parser, to have more robust hypers but that are quite different to the one used in the paper. In particular the default weights in the loss.
This input should be able to reproduce it on the develop branch.
python ./mace-main/scripts/run_train.py \
--name="MACE_3bpa" \
--train_file="train_300K.xyz" \
--valid_fraction=0.1 \
--test_file="test.xyz" \
--energy_weight=27.0 \
--forces_weight=1000.0 \
--config_type_weights='{"Default":1.0}' \
--E0s='{1: -13.587222780835477, 6: -1029.4889999855063, 7: -1484.9814568572233, 8: -2041.9816003861047}' \
--model="ScaleShiftMACE" \
--interaction_first="RealAgnosticResidualInteractionBlock" \
--interaction="RealAgnosticResidualInteractionBlock" \
--num_interactions=2 \
--max_ell=3 \
--hidden_irreps='256x0e + 256x1o + 256x2e' \
--num_cutoff_basis=5 \
--correlation=3 \
--r_max=5.0 \
--scaling='rms_forces_scaling' \
--batch_size=5 \
--max_num_epochs=2000 \
--patience=256 \
--weight_decay=5e-7 \
--ema \
--ema_decay=0.99 \
--amsgrad \
--default_dtype="float32"\
--clip_grad=None \
--device=cuda \
--seed=3 \
On the question of added functionalities, they were not used in the code and added later to provide more robust defaults. It is on continuous development. I have linked here a zip file with the original code used for the paper, and the original input file.
The dataset can be found in the following repo : https://github.com/davkovacs/BOTNet-datasets .
For the last question on the eval script, did you make sure to use the .model
file? As opposed to the .pt
Please let me know if you have any other questions. In case it is needed, I have attached the copy of the code and the input script that I have used for the paper, if we stumble into compatibility problems.
Looking forward to hear from you.
Ilyes
reproduce_mace.zip
from mace.
Awesome, this worked. Thanks a lot.
from mace.
Related Issues (20)
- What are the node energies? HOT 2
- Training flag not working HOT 2
- Dipole fitting on multi-GPU
- Is 0.694 timesteps/s a normal speed?
- select pretrained configs for multihead replay with or without replacement? HOT 2
- adding cmd argument for specifying wandb directory HOT 2
- CUDA Out of Memory with GCMC
- Distributed Training: invalid device ordinal
- Missing dependancies in `multi-head-interface` HOT 1
- Stress / num_atoms in loss? HOT 3
- stress errors reported during fitting and at end should be consistent HOT 4
- during-training valid loss is wrong HOT 11
- np.random.choice for selecting pre-fit subset for multihead fine tune doesn't accept a list of Atoms
- multi-head-finetune validation batches not deterministic HOT 3
- installation HOT 6
- How to Convert VASP Output Files to MACE XYZ Format for Training and Testing Sets
- How to use multi-training without slurm system? HOT 11
- Problem in Compiling Lammps for GPU version HOT 3
- Error in online data loading for large datasets HOT 1
- Does MACE satisfy E(3) equivariance or just SE(3) equivariance?
Recommend Projects
-
React
A declarative, efficient, and flexible JavaScript library for building user interfaces.
-
Vue.js
🖖 Vue.js is a progressive, incrementally-adoptable JavaScript framework for building UI on the web.
-
Typescript
TypeScript is a superset of JavaScript that compiles to clean JavaScript output.
-
TensorFlow
An Open Source Machine Learning Framework for Everyone
-
Django
The Web framework for perfectionists with deadlines.
-
Laravel
A PHP framework for web artisans
-
D3
Bring data to life with SVG, Canvas and HTML. 📊📈🎉
-
Recommend Topics
-
javascript
JavaScript (JS) is a lightweight interpreted programming language with first-class functions.
-
web
Some thing interesting about web. New door for the world.
-
server
A server is a program made to process requests and deliver data to clients.
-
Machine learning
Machine learning is a way of modeling and interpreting data that allows a piece of software to respond intelligently.
-
Visualization
Some thing interesting about visualization, use data art
-
Game
Some thing interesting about game, make everyone happy.
Recommend Org
-
Facebook
We are working to build community through open source technology. NB: members must have two-factor auth.
-
Microsoft
Open source projects and samples from Microsoft.
-
Google
Google ❤️ Open Source for everyone.
-
Alibaba
Alibaba Open Source for everyone
-
D3
Data-Driven Documents codes.
-
Tencent
China tencent open source team.
from mace.