ML EPS LOW: Difference between revisions
No edit summary |
No edit summary |
||
| Line 5: | Line 5: | ||
This value sets the threshold for the eigenvalues that contribute to the leverage scoring used in the CUR algorithm for the rank compression ("sparsification") of the local environments (for details see appendix E of reference {{cite|jinnouchi2:arx:2019}}). Small eigenvalues and those columns (local configurations) that are strongly connected | This value sets the threshold for the eigenvalues that contribute to the leverage scoring used in the CUR algorithm for the rank compression ("sparsification") of the local environments (for details see appendix E of reference {{cite|jinnouchi2:arx:2019}}). Small eigenvalues and those columns (local configurations) that are strongly connected | ||
with these small eigenvalues are removed by the sparsification routines. The default value is fairly well balanced. However, if extensive training is performed, we recommend to reduce the threshold to | with these small eigenvalues are removed by the sparsification routines. The default value is fairly well balanced. However, if extensive training is performed, we recommend to reduce the threshold to | ||
1E-12. Unnecessary local environments can be removed in a post processing step ( | 1E-12. Unnecessary local environments can be removed in a post processing step (a single additional learning step with {{TAG|ML_FF_ISTART}}=1 using new parameters), after the on the fly learning | ||
has been finished. | has been finished. | ||
Revision as of 06:00, 2 September 2020
ML_FF_EPS_LOW = [real]
Default: ML_FF_EPS_LOW = 1E-10
Description: Threshold for the CUR algorithm used in the sparsification of the machine learning force fields.
This value sets the threshold for the eigenvalues that contribute to the leverage scoring used in the CUR algorithm for the rank compression ("sparsification") of the local environments (for details see appendix E of reference [1]). Small eigenvalues and those columns (local configurations) that are strongly connected with these small eigenvalues are removed by the sparsification routines. The default value is fairly well balanced. However, if extensive training is performed, we recommend to reduce the threshold to 1E-12. Unnecessary local environments can be removed in a post processing step (a single additional learning step with ML_FF_ISTART=1 using new parameters), after the on the fly learning has been finished.
Related Tags and Sections
ML_FF_LMLFF, ML_FF_MB_MB, ML_FF_LBASIS_DISCARD, ML_FF_LCONF_DISCARD
References