Skip to content

Home

Jump to bottom
bart1 edited this page Jul 12, 2023 · 2 revisions

  • Omitting x, y coords to test edges: drop in accuracy after 40 epochs

  • Best model so far (accuracy 96%): runs/3d374601-MED_650.0-NN_50-MPPL_5000-UMIC_True-AETP_True-features_range-azimuth-elevation-z-DBZH-TH-VRADH-RHOHV-PHIDP-centered_x-centered_y-DBZH_isna-TH_isna-VRADH_isna-RHOHV_isna-PHIDP_isna/in-GC_16-RELU-GC_2-mean-out-Conv1+Conv2+Conv3/LR_0.005-BS_512-SEED132/1,[150,300],0.5

  • Adding indicator column always add accuracy (bart)

  • By varying graph size (num_nodes) between 1,10,50 I see larger graphs outperforming smaller ones: image

  • Varying the maximal_edge_distance (MED) is less conclusive for some single cases a very low value (3) seems to work slightly better but an other seed does not show the same result: image

Clone this wiki locally