Gradient boosting

Ensemblemetoder i Python

Román de las Heras

Data Scientist, Appodeal

Introduktion till gradient boosting

function_objective.png

  1. Initialmodell (svag estimator): $y\sim f_1(X)$
  2. Ny modell anpassas till residualer: $y-f_1(X)\sim f_2(X)$
  3. Ny additiv modell: $y\sim f_1(X)+f_2(X)$
  4. Upprepa $n$ gånger eller tills felet är tillräckligt litet
  5. Slutlig additiv modell: $$y\sim f_1(X)+f_2(X)+ ... +f_n(x)=\sum_{i=1}^{n} f_i(X)$$
Ensemblemetoder i Python

Ekvivalens med gradientnedstigning

function_residuals.png

Gradientnedstigning:

function_loss.png

function_gradient.png

Residualer = negativ gradient

function_gradient_negative.png

Ensemblemetoder i Python

Gradient boosting-klassificerare

Gradient Boosting-klassificerare

from sklearn.ensemble import GradientBoostingClassifier

clf_gbm = GradientBoostingClassifier(
   n_estimators=100,
   learning_rate=0.1,
   max_depth=3,
   min_samples_split,
   min_samples_leaf,
   max_features
)
  • n_estimators
    • Standard: 100
  • learning_rate
    • Standard: 0.1
  • max_depth
    • Standard: 3
  • min_samples_split
  • min_samples_leaf
  • max_features
Ensemblemetoder i Python

Gradient boosting-regressor

Gradient Boosting-regressor

from sklearn.ensemble import GradientBoostingRegressor

reg_gbm = GradientBoostingRegressor(
   n_estimators=100,
   learning_rate=0.1,
   max_depth=3,
   min_samples_split,
   min_samples_leaf,
   max_features
)
Ensemblemetoder i Python

Nu kör vi en övning!

Ensemblemetoder i Python

Preparing Video For Download...