René's URL Explorer Experiment

{"@context":"https://schema.org","@type":"DiscussionForumPosting","headline":"Inconsistent force predictions between `tensorpot` and `pyace` backends","articleBody":"Dear python-ace developers,\n\nI trained an ACE potential (`output_potential.yaml`) using the `tensorpot` backend and wanted to use the resulting potential with the ASE `pyace` calculator.\n\nWhen plotting **F_pred vs F_ref** on my test dataset, I noticed that the resulting force MAE and RMSE differ significantly from those reported in `log.txt` during training.\n\nTo ensure this was not an issue in my own Python code, I performed a no-fit run using:\n\n~~~bash\npacemaker input.yaml --no-fit --potential output_potential.yaml\n~~~\n\nThe `input.yaml` file is identical to the one used during training, except:\n- The backend is set to `pyace` instead of `tensorpot`\n- The filenames are set to `fitting_data_info.pckl.gzip` and `test_data_info.pckl.gzip` (both generated during training)\n\nThis run gave me the same forces as my own ASE calculator, confirming consistency between them. However, a subset of forces predicted by `pyace` differs noticeably from those predicted by `tensorpot`.\n\nI attached a minimal dataset for reproducing this issue:\n- `output_potential.yaml` – the trained potential  \n- `difference_data.pckl.gzip` – ASE Atoms, reference energies, forces, and positions  \n- `difference_data_pred_tensorpot.pckl.gzip` – predictions by `tensorpot`  \n- `difference_data_pred_pyace.pckl.gzip` – predictions by `pyace`  \n- `input.yaml` – configuration file used  \n- Two scatter plots of **F_tensorpot vs F_pyace**\n\nI am using:\n- python version: 3.10.13\n- pacemaker/pyace version: 0.2.7+172.gd1c213a\n- ace_evaluator version: 2023.11.25\n\nThe file:\n[example.tar.gz](https://github.com/user-attachments/files/23195936/example.tar.gz)\n\nPlease let me know if additional information is needed to reproduce this issue.\n\n---\n\n**Questions:**\n1. Is this difference between `tensorpot` and `pyace` expected, or does it indicate an issue?  \n2. Would it be better to train the potential directly with `backend=pyace` to avoid such inconsistencies?  \n3. If I run MD simulations with LAMMPS, is there any known issue regarding different force predictions?\n\nThank you very much for your help!\n\nBest,  \nSijia\n\n","author":{"url":"https://github.com/Alexandrina-Chen","@type":"Person","name":"Alexandrina-Chen"},"datePublished":"2025-10-28T19:10:23.000Z","interactionStatistic":{"@type":"InteractionCounter","interactionType":"https://schema.org/CommentAction","userInteractionCount":0},"url":"https://github.com/97/python-ace/issues/97"}