Dataset of local linear GENE simulation parameters, outputs, and surrogate-model predictions

This dataset contains the data used to train and evaluate machine-learning surrogate models for local linear GENE simulations of the MAST-U pedestal. It includes the samples used for training, validation and testing of the surrogate models. Each row corresponds to a local linear GENE simulation at a given radial position and binormal wavenumber. The dataset contains local plasma and simulation parameters, the corresponding GENE outputs, and predictions from the trained surrogate models. The parameters included in the dataset are: sample_id: identifier of the sampled pedestal equilibrium. x0: radial position of the local GENE simulation. kymin: normalized binormal wavenumber. beta: local plasma beta. coll: collisionality. q0: safety factor. shat: magnetic shear. rhostar: normalized gyroradius. dpdx_pm: ratio of the negative equilibrium pressure gradient and the reference magnetic pressure. Bref: reference magnetic field. Tref: reference temperature. nref: reference density. Lref: reference length. omn_e, omt_e: normalized electron density and temperature gradients. omn_i, omt_i: normalized ion density and temperature gradients. omegatorref: reference toroidal rotation at the local flux surface. These quantities follow the normalization and sign conventions used by GENE and correspond to the quantities supplied to, or derived for, the local flux-tube simulations. Multiple rows can therefore correspond to the same sample_id, with different radial positions (x0) and wavenumbers (kymin). The true GENE outputs included in the dataset are the linear growth rate gamma, real frequency omega, and the transport ratios D_e/Chi_e and Chi_i/Chi_e. Predictions are provided from two surrogate models: pred_mhmlp_*: predictions from the multi-head multilayer perceptron (MH-MLP). pred_classreg_*: predictions from the classification-regression (Class-Reg) model. The MH-MLP predicts gamma, omega, D_e/Chi_e, and Chi_i/Chi_e. The Class-Reg model predicts omega, D_e/Chi_e, and Chi_i/Chi_e. The class_omega_name column contains the frequency-sign class predicted by the classifier component of the Class-Reg model. The dataset accompanies the study on machine-learning surrogate modelling of pedestal microinstabilities in MAST-U