Skip to content

GitLab

Last edited by Jeong-Yoon Lee
Page history

Home

INTERNAL LEADERBOARD

Level 2 Ensemble Models

Model Public LB MCC Score CV MCC Score CV AUC Score Threshold Comment
avg_esb.esb3.sub.csv 0.48895 0.494301 0.929332 0.326768 average of esb.esb3

Level 1 Ensemble Models

Model Public LB MCC Score CV MCC Score CV AUC Score Threshold Comment
keras_20_2_128_0.5_512_5_esb15 - 0.494306 0.926863 0.267374 keras with esb15
keras_20_2_128_0.5_512_5_esb13 0.48830 0.494092 0.923443 0.247576 keras with esb13
xg_10000_6_0.05_1_0.7_0.5_1_100_0.5_esb16 - 0.493853 0.928585 0.415859 xgb with esb16
xg_10000_6_0.05_1_0.7_0.5_1_100_0.5_esb13 0.48756 0.493007 0.928596 0.435657 xgb with esb13
xg_10000_6_0.05_1_0.7_0.5_1_100_0.5_esb11 0.47429 0.484894 0.927990 0.425758 xgb with esb11
xg_10000_6_0.05_1_0.7_0.5_1_100_0.5_esb9 0.41406 0.415531 0.907144 0.306970 xgb with esb9
keras_20_2_128_0.5_512_5_esb9 0.40453 0.416551 0.905934 0.217879 keras with esb9
xg_10000_6_0.05_1_0.7_0.5_1_100_0.5_esb8 0.40410 0.416778 0.907292 0.425758 xgb with esb8

Single Models

Model Public LB MCC Score CV MCC Score CV AUC Score Threshold Comment
xg_10000_6_0.05_1_0.8_0.5_1_100_0.5_xi 0.46912 0.471735 0.925452 0.346566 xgb with feature xi
xg_10000_6_0.05_1_0.8_0.5_1_100_0.5_h2 0.40621 0.414588 0.905808 0.316869 xgb with feature h2
xg_10000_6_0.05_1_0.7_0.5_1_100_m8 0.40318 0.412482 0.904498 0.277273 xgb with feature m8
xg_10000_6_0.05_1_0.7_0.5_1_100_m10 0.40102 0.414860 0.907744 0.386162 xgb with feature m10
xg_10000_6_0.05_1_0.7_0.5_1_100_m12 0.40055 0.412622 0.905791 0.277273 xgb with feature m12
xg_10000_6_0.05_1_0.7_0.5_1_100_m13 0.39887 0.410871 0.905665 0.267374 xgb with feature m13
xg_10000_6_0.05_1_0.7_0.5_1_100_m11 0.39838 0.410234 0.904060 0.267374 xgb with feature m11
xg_10000_6_0.05_1_0.7_0.5_1_100_m19 0.39804 0.411858 0.904613 0.415859 xgb with feature m19
xg_10000_6_0.05_1_0.7_0.5_1_100_m14 0.39719 0.417231 0.908390 0.306970 xgb with feature m14
xg_10000_6_0.05_1_0.7_0.5_1_100_m11_cat_cnt 0.39696 0.413397 0.904728 0.356465 xgb with feature m11_cat_cnt
xg_10000_6_0.05_1_0.7_0.5_1_100_m18 0.39236 0.409568 0.905975 0.356465 xgb with feature m18
xg_10000_6_0.05_1_0.7_0.5_1_100_m9 0.38976 0.407061 0.902651 0.326768 xgb with feature m9
xg_10000_6_0.05_1_0.7_0.5_1_100_m15 0.39510 0.411835 0.905861 0.326768 xgb with feature m15
xg_10000_6_0.05_1_0.7_0.5_1_100_m16 0.36598 0.431463 0.914549 0.376263 xgb with feature m15
xg_10000_6_0.05_1_0.7_0.5_1_100_m17 0.27355 0.287223 0.792510 0.188182 xgb with feature m17
xg_10000_6_0.05_1_0.8_0.5_1_100_0.00581_m5 - 0.260144 0.736417 0.247576 xgb w prior w feature m5
xg_10000_6_0.05_1_0.7_0.5_1_100_m3 0.25978 0.262398 0.735327 0.158485 xgb with feature m3
xg_10000_6_0.05_1_0.7_0.5_1_100_m2 0.25847 0.268000 0.741536 0.210000 xgb with feature m2 (no retrain)
xg_10000_6_0.05_1_0.7_0.5_1_100_m4 0.25814 0.267268 0.739925 0.190000 xgb with feature m4
xg_10000_6_0.05_1_0.7_0.5_1_100_m6 0.25715 0.265159 0.733994 0.168384 xgb with feature m6
xg_10000_6_0.05_1_0.7_0.5_1_100_m5 0.25702 0.265147 0.733484 0.178283 xgb with feature m5
xg_10000_6_0.05_1_0.7_0.5_1_100_0.5_j1 0.25658 0.266439 0.737122 0.227778 xgb with feature j1
xg_10000_6_0.05_1_0.8_0.5_1_100_m2 0.25569 0.266177 0.741136 0.190000 xgb with feature m2
xg_10000_6_0.05_1_0.7_0.5_1_100_m7 0.25334 0.266705 0.737535 0.168384 xgb with feature m7
xg_10000_6_0.05_1_0.8_0.5_1_100_m1 0.25306 0.261948 0.735805 0.170000 xgb with feature m1
xg_10000_6_0.05_1_0.8_0.5_1_100_m1p 0.23213 0.298276 0.767727 0.227778 xgb with feature m1p
xg_10000_6_0.05_1_0.8_0.5_1_100_h1 0.20810 0.207277 0.732212 0.110000 xgb with feature h1
xg_10000_6_0.05_1_0.8_0.5_1_100_0.00581_h1 - 0.205662 0.732547 0.099091 xgb w prior w feature h1
xg_10000_6_0.05_1_0.8_0.5_1_100_h1 - 0.195504 0.731770 0.070000 xgb with feature h1

Feature Ideas

Other Ideas

A very cool viz get more understanding of data set

Meeting 10/21:

  • Different delta for ids (Mert)
  • Magic features per line station (Mert)
  • hd5 implementation (Jeong)
  • Time of day, day of week features with Downtime/weekend distinction (Jeong)
  • Product features (Erkut)
  • All new features will have their own makefile e.g. Makefile.ls (linestation)

Meeting 10/14:

  • Different delta for ids (Mert)

  • encode error rate of each line station combinations (52 features) (Mert)
    • encode time-varying error rate 52 x 10 ...
  • Encode error rate of each product (~8000 features) (Erkut)
    • time-varying version
  • Extend time overall (h2) to time per station / line (Hang)
  • Combine train and test some stats?
  • Change makefile structure to separate group of features