A spherical template, which functioned well due to the homogeneity of pyrenoid composition and the roughly spherical shape of RuBisCO at the tomogram resolution, made it possible to discard the rotational search, dramatically increasing the localization speed and enabling iterative fine-tuning of parameter values.
Most of the eight unsupervised methods evaluated here can be tuned by selection of parameter values.
We do not tune parameters again for each model and instead use the optimal setting of parameter values from our 10-fold CV experiments.
This limitation was overcome by having obtained good estimates of parameter values prior to this step, derived by the empirical-preparatory manual tuning procedure.
Based on the fine tuning of their parameter values through a Design of Experiment, all algorithms are compared in order to highlight their strengths and weaknesses.
The values can be determined by evaluating the performance over a grid of tuning parameter values using the boostrap or cross-validation.
Although trial and error also is used in this study, our findings provide a general guideline for the selection of tuning parameter values.
The time usage for cross-validation relies on the number of folds and number of potential tuning parameter values under investigation.
Besides, a constraint condition for tuning the parameter values of the B-spline base and an involved smooth factor is presented for restricting the nonlinear regularization process.
To further reduce the computation time of POCRE and POCRE0, some strategies (discussed later) can be designed to search for a much smaller number of candidate tuning parameter values during cross-validation, and the number of folds can be reduced when a large number of individuals is available.
A parallel hybrid framework that combines gene expression programming (GEP) as the evolutionary problem-solving methodology and alternative meta-heuristics for tuning parameter values of the parallel GEP runs is presented.
