Trained Neural Network Weights for model selection on Cosmic Microwave Background maps

This dataset contains the trained Neural Network weights and associated data produced during the model selection training on Cosmic Microwave Background (CMB) temperature and polarisation maps (T, Q, U), as part of the analysis in:https://github.com/SkyExplain/SkyNeuralNets The objective is to train a hybrid PCA–MLP neural network architecture on CMB maps (T, Q, U) to perform model selection between the standard cosmological model, ΛCDM, and a model featuring an oscillatory template in the primordial power spectrum, motivated by early Universe physics. The feature is parametrised as: P_R(k) = P_{R,0}(k) [1 + A_lin sin(ω_lin k/k* + φ)] where the feature amplitude A_lin and feature frequency ω_lin are varied across the dataset. The input CMB maps used for training were simulated using CAMB and healpy at Planck resolution, with the official Planck galactic mask and Planck-like noise applied. The full simulation pipeline is available at: https://github.com/SkyExplain/SkySimulation, and the maps are available at: 10.5281/zenodo.19445834 The interpretability analysis of the network's predictions was performed using SHAP (SHapley Additive exPlanations) via https://github.com/SkyExplain/SkyInterpret. The resulting SHAP attribution maps are available at: 10.5281/zenodo.19445676