Skip to content

Latest commit

 

History

History
30 lines (21 loc) · 934 Bytes

04_tune_model.md

File metadata and controls

30 lines (21 loc) · 934 Bytes

Model tuning

The default models included with ampir are both SVMs with radial kernel and have two tuning parameters, sigma and C (cost). Optimal values for these tuning parameters were obtained by using model tuning functions available in caret. Since this is computationally intensive it was performed on an HPC system. The R script scripts/tune_model.R provides a convenient wrapper for the required functions and allows it to be run easily on an HPC system. See the scripts directory for details of how this was run.

Inspecting the results for the mature peptide model reveals the variation of model performance with the tuning parameters.

Best tune:

##    sigma C
## 57  0.06 1

And for the precursor model:

Best tune:

##     sigma C
## 104   0.1 4