Imputation - Timesnet vs NonStationaryTransformer

[omkar-334]

in the leaderboard, for time series imputation task, timesnet is reported as first and non-stationary transformer is reported as second.
However, when i run the scripts for ETTh1 without any changes, here are the results that i get -

imputation_ETTh1_mask_0.125_TimesNet_ETTh1_ftM_sl96_ll0_pl0_dm16_nh8_el2_dl1_df32_expand2_dc4_fc3_ebtimeF_dtTrue_Exp_0  
mse:0.0618780180811882, mae:0.16756796836853027

imputation_ETTh1_mask_0.25_TimesNet_ETTh1_ftM_sl96_ll0_pl0_dm16_nh8_el2_dl1_df32_expand2_dc4_fc3_ebtimeF_dtTrue_Exp_0  
mse:0.08032868057489395, mae:0.19080348312854767

imputation_ETTh1_mask_0.375_TimesNet_ETTh1_ftM_sl96_ll0_pl0_dm16_nh8_el2_dl1_df32_expand2_dc4_fc3_ebtimeF_dtTrue_Exp_0  
mse:0.09744171798229218, mae:0.2092771679162979

imputation_ETTh1_mask_0.5_TimesNet_ETTh1_ftM_sl96_ll0_pl0_dm16_nh8_el2_dl1_df32_expand2_dc4_fc3_ebtimeF_dtTrue_Exp_0  
mse:0.11592429876327515, mae:0.22686615586280823

imputation_ETTh1_mask_0.125_Nonstationary_Transformer_ETTh1_ftM_sl96_ll0_pl0_dm128_nh8_el2_dl1_df128_expand2_dc4_fc3_ebtimeF_dtTrue_Exp_0  
mse:0.04747990891337395, mae:0.14657214283943176

imputation_ETTh1_mask_0.25_Nonstationary_Transformer_ETTh1_ftM_sl96_ll0_pl0_dm128_nh8_el2_dl1_df128_expand2_dc4_fc3_ebtimeF_dtTrue_Exp_0  
mse:0.06437327712774277, mae:0.16930104792118073

imputation_ETTh1_mask_0.375_Nonstationary_Transformer_ETTh1_ftM_sl96_ll0_pl0_dm128_nh8_el2_dl1_df128_expand2_dc4_fc3_ebtimeF_dtTrue_Exp_0  
mse:0.08262055367231369, mae:0.1922348588705063

imputation_ETTh1_mask_0.5_Nonstationary_Transformer_ETTh1_ftM_sl96_ll0_pl0_dm128_nh8_el2_dl1_df128_expand2_dc4_fc3_ebtimeF_dtTrue_Exp_0  
mse:0.10656864196062088, mae:0.21720004081726074

This shows that non-stationary transformer beats timesnet in this task....
Any explanation for this?

[omkar-334]

@wuhaixu2016 These results contradict both the leaderboard rankings and the reported results in TimesNet paper. According to this, NS-Transformer beats Timesnet, but the paper shows otherwise...