BanditRegressor

Bandit-based model selection for regression.

Each model is associated with an arm. At each learn_one call, the policy decides which arm/model to pull. The reward is the performance of the model on the provided sample. The predict_one method uses the current best model.

Parameters

• models

  The models to select from.

• metric

  Type → metrics.base.RegressionMetric

  The metric that is used to measure the performance of each model.

• policy

  Type → bandit.base.Policy

  The bandit policy to use.

Attributes

• best_model

• models

Examples

from river import bandit
from river import datasets
from river import evaluate
from river import linear_model
from river import metrics
from river import model_selection
from river import optim
from river import preprocessing

models = [
    linear_model.LinearRegression(optimizer=optim.SGD(lr=lr))
    for lr in [0.0001, 0.001, 1e-05, 0.01]
]

dataset = datasets.TrumpApproval()
model = (
    preprocessing.StandardScaler() |
    model_selection.BanditRegressor(
        models,
        metric=metrics.MAE(),
        policy=bandit.EpsilonGreedy(
            epsilon=0.1,
            decay=0.001,
            burn_in=100,
            seed=42
        )
    )
)
metric = metrics.MAE()

evaluate.progressive_val_score(dataset, model, metric)

MAE: 3.13815

Here's another example using the UCB policy. The latter is more sensitive to the target scale, and usually works better when the target is rescaled.

models = [
    linear_model.LinearRegression(optimizer=optim.SGD(lr=lr))
    for lr in [0.0001, 0.001, 1e-05, 0.01]
]

model = (
    preprocessing.StandardScaler() |
    preprocessing.TargetStandardScaler(
        model_selection.BanditRegressor(
            models,
            metric=metrics.MAE(),
            policy=bandit.UCB(
                delta=1,
                burn_in=100
            )
        )
    )
)
metric = metrics.MAE()

evaluate.progressive_val_score(dataset, model, metric)

MAE: 0.875457

Methods

learn_one

Fits to a set of features x and a real-valued target y.

Parameters

• x
• y

Returns

self

predict_one

Predict the output of features x.

Parameters

• x

Returns

The prediction.