#!/usr/bin/env python3
# Copyright (c) Meta Platforms, Inc. and affiliates.
#
# This source code is licensed under the MIT license found in the
# LICENSE file in the root directory of this source tree.
from __future__ import annotations
from typing import Any, List, Optional, Tuple
import numpy as np
import torch
from ax.core.search_space import SearchSpaceDigest
from ax.core.types import TCandidateMetadata
from ax.models.torch.utils import _datasets_to_legacy_inputs
from ax.models.torch_base import TorchModel
from ax.utils.common.docutils import copy_doc
from botorch.utils.datasets import SupervisedDataset
from sklearn.ensemble import RandomForestRegressor
from sklearn.tree import DecisionTreeRegressor
from torch import Tensor
[docs]class RandomForest(TorchModel):
"""A Random Forest model.
Uses a parametric bootstrap to handle uncertainty in Y.
Can be used to fit data, make predictions, and do cross validation; however
gen is not implemented and so this model cannot generate new points.
Args:
max_features: Maximum number of features at each split. With one-hot
encoding, this should be set to None. Defaults to "sqrt", which is
Breiman's version of Random Forest.
num_trees: Number of trees.
"""
def __init__(
self, max_features: Optional[str] = "sqrt", num_trees: int = 500
) -> None:
self.max_features = max_features
self.num_trees = num_trees
self.models: List[RandomForestRegressor] = []
[docs] @copy_doc(TorchModel.fit)
def fit(
self,
datasets: List[SupervisedDataset],
search_space_digest: SearchSpaceDigest,
candidate_metadata: Optional[List[List[TCandidateMetadata]]] = None,
) -> None:
Xs, Ys, Yvars = _datasets_to_legacy_inputs(datasets=datasets)
for X, Y, Yvar in zip(Xs, Ys, Yvars):
self.models.append(
_get_rf(
X=X.numpy(),
Y=Y.numpy(),
Yvar=Yvar.numpy(),
num_trees=self.num_trees,
max_features=self.max_features,
)
)
[docs] @copy_doc(TorchModel.predict)
def predict(self, X: Tensor) -> Tuple[Tensor, Tensor]:
return _rf_predict(self.models, X)
[docs] @copy_doc(TorchModel.cross_validate)
def cross_validate( # pyre-ignore [14]: not using metric_names or ssd
self,
datasets: List[SupervisedDataset],
X_test: Tensor,
**kwargs: Any,
) -> Tuple[Tensor, Tensor]:
Xs, Ys, Yvars = _datasets_to_legacy_inputs(datasets=datasets)
cv_models: List[RandomForestRegressor] = []
for X, Y, Yvar in zip(Xs, Ys, Yvars):
cv_models.append(
_get_rf(
X=X.numpy(),
Y=Y.numpy(),
Yvar=Yvar.numpy(),
num_trees=self.num_trees,
max_features=self.max_features,
)
)
return _rf_predict(cv_models, X_test)
def _get_rf(
X: np.ndarray,
Y: np.ndarray,
Yvar: np.ndarray,
num_trees: int,
max_features: Optional[str],
) -> RandomForestRegressor:
"""Fit a Random Forest model.
Args:
X: X
Y: Y
Yvar: Variance for Y
num_trees: Number of trees
max_features: Max features specifier
Returns: Fitted Random Forest.
"""
r = RandomForestRegressor(
n_estimators=num_trees, max_features=max_features, bootstrap=True
)
# pyre-fixme[16]: `RandomForestRegressor` has no attribute `estimators_`.
r.estimators_ = [DecisionTreeRegressor() for i in range(r.n_estimators)]
for estimator in r.estimators_:
# Parametric bootstrap
y = np.random.normal(loc=Y[:, 0], scale=np.sqrt(Yvar[:, 0]))
estimator.fit(X, y)
return r
def _rf_predict(
models: List[RandomForestRegressor], X: Tensor
) -> Tuple[Tensor, Tensor]:
"""Make predictions with Random Forest models.
Args:
models: List of models for each outcome
X: X to predict
Returns:
mean and covariance estimates
"""
f = np.zeros((X.shape[0], len(models)))
cov = np.zeros((X.shape[0], len(models), len(models)))
for i, m in enumerate(models):
# pyre-fixme[16]: `RandomForestRegressor` has no attribute `estimators_`.
preds = np.vstack([tree.predict(X.numpy()) for tree in m.estimators_])
f[:, i] = preds.mean(0)
cov[:, i, i] = preds.var(0)
return torch.from_numpy(f), torch.from_numpy(cov)
