aggregation bootstrap bagging forest random boosting
play

- " " AGGREGATION " " BOOTSTRAP BAGGING - PowerPoint PPT Presentation

METHODS ENSEMBLE - " " AGGREGATION " " BOOTSTRAP BAGGING FOREST RANDOM BOOSTING ATAmlLIARSETc Efx ; ]=u X , . X . X. , X . . . . . . Var [ Xi ) ' - r - Efx . ) - n . - Efx ] - a - ? varlet


  1. METHODS ENSEMBLE - " " AGGREGATION " " BOOTSTRAP BAGGING • FOREST RANDOM • BOOSTING •

  2. ATAmlLIARSETc Efx ; ]=u X , . X . X. , X . . . . . . Var [ Xi ) ' - r - Efx . ) - n . - Efx ] - a - ? varlet . " = . . .

  3. . . " " . . } X y , ! I Fine I , : D : ^ / " " ! . . . - ¥71 's .IS :c , ÷÷÷÷÷ #HgI*q= " ⇒ . . 7 - ya ' i : / tar . ) : ' ' ' ' MODEL i BAGGED • , MODEL ENSEMBLE AN Y X • I :c 's 5 D* 5 6 B- REPLICATES BESAMPLES BOOTSTRAP BOOTSTRAP

  4. ↳ B = & Ifi # Db ) # BESA - PLES D- to THAT → Naoise . I - th POINT DATA contain b - I - noo÷ E FI ( xi ) III. ( Xi ) = prenotion oops i -4dB # RE SAMPLE 's Do WHICH X but CONTAIN [ ( yi - II. ( x :)) ' = 0013 BMSE f# c- =/ | , .org D # a pa , , ,u , yea 1/5 LESS COM POTATO - TRAINING TIME → once Bootstrap → • way CV c- THAN 5- Foa 's us

  5. SIMPLE f) Fro - II SILLY EXAMPLE ppm II l Is KNN - - +1--6 II. Air it . :* , s II. 1*41=47--5 " " ' ' =f÷÷÷ n' T I Bout z , From IT 6 i z ' i¥H÷÷÷ it :c " " .÷¥÷÷ :* : . x ) :{ ÷÷÷ I B ' . ooo .nse=÷iiInui= µ ⇒

  6. ↳ . ÷÷÷ ⇒ varlet . . . N ✓ REPLICATES Bootstrap CORRELATED ARE - CORRELATION ? THIS REDUCE HOW TO PROCEDURE ! Fc TANG ADD RANDOMNESS to - IEIIEII " trees - with s . " ← .

  7. ↳ ↳ TONED OFTEN f) FOREST RANDOM SPLIT - EACH SAMPLE AT m try PREDICTORS TO # → REPLICATES BOOTSTRAP PARAMETERS ↳ IS THAT TREES , TUNING # n tree \ → n tree - 500 w/ FAXED usually - \ TEES AMPLE w/ is BOOT STRAP mtry =p 1 : ntree b BF For in y > . model tree gauge , a D , , , , ,, ⇐ , , - y , p , , , SPLIT EACH { AT " " P } • { × " ' × " " " Mt - J F " " - " VA " " " " SAMPLE a " mt - y UAREIABCES THESE From SPLIT BEST FIND II • y " µ→ ⇐ s " t ! ! " ° " " - TREE Grow . 1 w.tn size CLASSIFICATION • 5 OBSERVATION MOST NODES AT ALL REGRESSION ← ntnee £ppG)= n ! I £ ( x ) REGRESSION FOR → RETURN b. =L CLASSIFICATION PAGE tow SEE NEXT

  8. FIRES 5 RANDOM CLASSIFICATION - t.cn#eiEii:cxsntree y " " MAJORITY ' ' ¥ b.FI ( = k ) VOTE ( x ) ramson Forest x ] P' in * [ y use , - KIX = - Jk , ,a= ( x ) - Thtkaoentree - = " ' II : EEE . ( x ) l l - D= - II. [ * Hx n - pie . . .cn - b - . PROBABIC 'T -1 ESTIMATED TREE D - th From THEN TO CLASSIFY PROBABILITY LARGEST

  9. F- zest TINI - 6 RANDO - A - p ) { I , 2,3 , n try C- - - . . - ( FEATURES # Try ( or si w i - - n DE - J m try I " " = Smale - f "÷ . . . } " ' nts ' " " ' ' ' " ' - y ne .im a . . . DEFAULT BY . ( L 1 mtg =p i BIG

  10. STILL Discussion NEEDS - B - X THE OF out RIGHT WORKS IT RF → STRENGTHS OF TREES ? RANDOMIZED EXTREMELY ! Boost .no • gbn - xgbaost

Recommend


More recommend