Larger absolute errors in case of steep slopes

Info iconThis preview shows page 1. Sign up to view the full content.

View Full Document Right Arrow Icon
This is the end of the preview. Sign up to access the rest of the document.

Unformatted text preview: of BART Inference 0.510 Validation RMSD 0.520 0.530 0.540 1 sample: 0.5459 100 samples: 0.5234 2000 samples: 0.5144 10000 samples: 0.5109 0 2000 4000 6000 Number of BART Samples 8000 10000 BART Feature Selec-on All 484 Features Ordered by Usage Top Ten Features Ordered by BART Usage Onset Delta Max Dressing Score ALSFRS Slope Last Systolic Blood Pressure Slope 0.0 0.5 1.0 1.5 2.0 2.5 3.0 3.5 Average usage Mean Weight Slope Last FVC Slope Last Weight Slope Last ALSFRS   Many pairwise slope features   Lab data excluded Min Turning Score Mean ALSFRS 0.0 0.5 1.0 1.5 2.0 Average usage 2.5 3.0 3.5 BART on Feature Subsets Validation RMSD 0.515 0.520 0.525 0.530 Effect of Adding Each Feature in Order of BART Usage max.dressing 1 feature: 0.5291 Onset.Delta last.slope.bp.systolic alsfrs.score.slope 3 features: 0.5246 mean.slope.weight last.slope.fvc.liters last.alsfrs.score 14 features: 0.5157 last.speech 6 features: 0.5190 last.handwriting 21 features: 0.5113 5 meansquares.speech 10 15 Features Added in Order of Usage 20 25 Model Comparison How do other models perform using our feature set? Model Model Our RMSD Our RMSD Compe;tor (Test) (Valida;on) RMSD Lasso Regression 0.5006 0.5287  ­ Random Forests 0.5052 0.5120 0.52 ­0.53 Boosted Trees 0.4940 0.5118  ­ BART 0.4860 0.5109     Addi;ve decision tree models especially effec-ve   Featuriza;on was a main differen-ator of compe-tors Contest Evalua-on … Pa-ent Pa-ent Pa-ent Pa-ent Pa-ent Baseline performance RMSD: Slow vs. Fast Progressors all slow 1 2 3 4 5 6 7 8 9 10 0.51 0.52 0.52 0.53 0.53 0.53 0.57 0.57 0.89 1.30 slow med fast 0.43 0.43 0.40 0.42 0.44 0.38 0.46 0.47 0.92 1.04 0.29 0.30 0.30 0.31 0.30 0.34 0.26 0.36 0.61 1.43 0.78 0.79 0.84 0.83 0.82 0.88 0.91 0.88 1.04 1.67 fast Different solvers predict slow or fast progressors more reliably. Larger (absolute) errors in case of steep slopes. Predcted slope...
View Full Document

Ask a homework question - tutors are online