ML RDES SPARSDES: Difference between revisions
No edit summary |
No edit summary |
||
| Line 5: | Line 5: | ||
---- | ---- | ||
During [[Machine_learning_force_field:_Theory#Sparsification_of_angular_descriptors|angular-descriptor sparsification]] ({{TAG|ML_LSPARSDES}}=T), insignificant angular descriptors are removed based on a leverage scoring. The percentage of angular descriptors that are | During [[Machine_learning_force_field:_Theory#Sparsification_of_angular_descriptors|angular-descriptor sparsification]] ({{TAG|ML_LSPARSDES}}=T), insignificant angular descriptors are removed based on a leverage scoring. The percentage of angular descriptors that are kept is determined by the value of {{TAG|ML_RDES_SPARSDES}}, which must be chosen between <math>0 < r \leq 1</math>. In practice, we recommend scanning a range between 0.1 to 0.9. Removing angular descriptors increases the performance of a force field, but it decreases accuracy at the same time. One method of finding the optimal tradeoff between accuracy and performance is to do a Pareto front with run time on the x-axis and accuracy on the y-axis. | ||
== Related tags and articles == | == Related tags and articles == | ||
Revision as of 09:36, 28 November 2023
ML_RDES_SPARSDES = [real]
Default: ML_RDES_SPARSDES = 0.5
Description: Sets the ratio of descriptors removed during angular-descriptor sparsification.
During angular-descriptor sparsification (ML_LSPARSDES=T), insignificant angular descriptors are removed based on a leverage scoring. The percentage of angular descriptors that are kept is determined by the value of ML_RDES_SPARSDES, which must be chosen between . In practice, we recommend scanning a range between 0.1 to 0.9. Removing angular descriptors increases the performance of a force field, but it decreases accuracy at the same time. One method of finding the optimal tradeoff between accuracy and performance is to do a Pareto front with run time on the x-axis and accuracy on the y-axis.