This is the source code for a solution scoring 0.85 on the Poker dataset.

Please see the following link for the notebook with descriptions: https://rawgit.com/Laurae2/Poker/master/Poker_Workbook.html

• Features used: 37
• Computation time (i7-4600U, 16GB RAM): approx. 1 hour for 0.805-0.805 Public/Private LB (you can achieve 0.8506-0.8503 Public/Private LB in less than 10 minutes using the feature engineering and XGBoost alone)
• Models used: Extreme Gradient Boosting, Random Forest, Gradient Boosted Machines, Deep Learning / MLP

N.B: Power 4 Magic weighted models coming from a blackbox. Your mileage may vary due to computation determinism and multithreading.

    h2o Random Forest     |    0.847355+0.001501    | 0.8473699 |  0.8468554

h2o Gradient Boosted Machines | 0.848027+0.001419 | 0.8485799 | 0.8485444 h2o Deep Learning / MLP | 0.843473+0.001568 | 0.8429358 | 0.8432577

     Best model (XGBoost) | 0.8506203 |  0.8503818
   Average of all models  | 0.8487509 |  0.8485156
   Power 2 of all models  | 0.8492364 |  0.8489594
   Power 4 of all models  | 0.8500361 |  0.8497416
   Power 8 of all models  | 0.8502813 |  0.8500659
  Power 16 of all models  | 0.8502647 |  0.8501988
 Power 12 weighted models | 0.8506915 |  0.8503820

Power 4 Magic weighted models | 0.8505044 | 0.8505464

Garbage collect information at the end (notice: approx. 2GB max usage, can go under 2GB if cleaning up all data everytime):

   |      used |   (Mb) | gc trigger |   (Mb) |  max used |   (Mb)

------ | --------- | ------ | ---------- | ------ | --------- | ------ Ncells | 1444124 | 77.2 | 4397788 | 234.9 | 5684620 | 303.6 Vcells | 180365034 | 1376.1 | 267283192 | 2039.3 | 267271001 | 2039.2