rank,id,method,kind,mae_heat_all,rmse_heat_all,r2_heat_all,mae_dir_gap,mae_dir_gap_nonzero,rmse_dir_gap,r2_dir_gap,mae_heat_all_nonzero,n_evaluated,parameters,training_data,date,paper,code,weights
1,meidnet-2k,MEIDNet (published model),model,0.40447624858304404,0.5389006864628664,0.47540607269846924,2.7917899843503173,1.6761308891094284,3.051066040757689,-31.399695258970574,0.4043139405576926,3785.0,702405,"Perov-5, all 18,928 materials",2026-10-03,https://doi.org/10.1038/s41524-026-02153-3,https://github.com/ABnano/MEIDNet,https://huggingface.co/Babu09/MEIDNet/blob/main/dual_autoencoder_clip_earlyfusion_propertyaware_2k.pth
2,baseline-composition-knn,Composition k-NN,baseline,0.5106261558784676,0.67409273437046,0.17918539075266848,0.16282166446499338,2.0438410596026486,0.5677910568862923,-0.12205670370125676,0.5105317124735729,3785.0,,"Perov-5 train (11,356)",2026-10-03,,https://github.com/ABnano/MEIDNet,
3,meidnet-earlyfusion,MEIDNet (first early-fusion model),model,0.5219494838015406,0.6984068976178183,0.11890482075956676,13.468713194168071,11.997500254144732,13.491628684634343,-632.528955744602,0.5216916534392829,3785.0,702405,"Perov-5, all 18,928 materials",2026-10-03,https://doi.org/10.1038/s41524-026-02153-3,https://github.com/ABnano/MEIDNet,https://huggingface.co/Babu09/MEIDNet/blob/main/dual_autoencoder_clip_earlyfusion.pth
4,baseline-train-mean,Training mean,baseline,0.5661949544069839,0.7441009623341133,-0.00016008318594540327,0.17263679417138986,2.135410460730273,0.536024930927169,-1.770296343450717e-05,0.5659521226170967,3785.0,,"Perov-5 train (11,356)",2026-10-03,,https://github.com/ABnano/MEIDNet,
5,meidnet-propertyaware,MEIDNet (shorter training),model,1.0086242815386826,1.3909827389896252,-2.4950167930714287,3.2924426712440216,1.9782721325261703,3.391715482506435,-39.03838367248946,1.008619448166443,3785.0,702405,"Perov-5, all 18,928 materials",2026-10-03,https://doi.org/10.1038/s41524-026-02153-3,https://github.com/ABnano/MEIDNet,https://huggingface.co/Babu09/MEIDNet/blob/main/dual_autoencoder_clip_earlyfusion_propertyaware.pth