10. RF (Random Forest; ランダムフォレスト)#
import numpy as np
import pandas as pd
import scipy
from sklearn.tree import DecisionTreeClassifier, DecisionTreeRegressor
from sklearn.datasets import load_wine
from sklearn.model_selection import train_test_split
from sklearn.preprocessing import LabelEncoder
from copy import copy
from sklearn.ensemble import RandomForestClassifier
try:
from palmerpenguins import load_penguins
except:
!pip install palmerpenguins
from palmerpenguins import load_penguins
SEED = 2023_02_15
10.1. データの準備#
print("-------Original DataFrame-------------------")
df = load_penguins()
print(df.shape)
display(df.head())
#display(df.info())
print("-------Preprocessed DataFrame-------------------")
df = df.dropna()
labelencoder = LabelEncoder()
df.island = labelencoder.fit_transform(df.island)
df.sex = labelencoder.fit_transform(df.sex)
df.species = labelencoder.fit_transform(df.species)
print(df.shape)
display(df.head())
X = df.drop("species", axis=1).to_numpy()
y = df["species"].to_numpy()
X_train,X_test,y_train,y_test=train_test_split(X,y,test_size=0.5,)#stratify=y)
-------Original DataFrame-------------------
(344, 8)
| species | island | bill_length_mm | bill_depth_mm | flipper_length_mm | body_mass_g | sex | year | |
|---|---|---|---|---|---|---|---|---|
| 0 | Adelie | Torgersen | 39.1 | 18.7 | 181.0 | 3750.0 | male | 2007 |
| 1 | Adelie | Torgersen | 39.5 | 17.4 | 186.0 | 3800.0 | female | 2007 |
| 2 | Adelie | Torgersen | 40.3 | 18.0 | 195.0 | 3250.0 | female | 2007 |
| 3 | Adelie | Torgersen | NaN | NaN | NaN | NaN | NaN | 2007 |
| 4 | Adelie | Torgersen | 36.7 | 19.3 | 193.0 | 3450.0 | female | 2007 |
-------Preprocessed DataFrame-------------------
(333, 8)
| species | island | bill_length_mm | bill_depth_mm | flipper_length_mm | body_mass_g | sex | year | |
|---|---|---|---|---|---|---|---|---|
| 0 | 0 | 2 | 39.1 | 18.7 | 181.0 | 3750.0 | 1 | 2007 |
| 1 | 0 | 2 | 39.5 | 17.4 | 186.0 | 3800.0 | 0 | 2007 |
| 2 | 0 | 2 | 40.3 | 18.0 | 195.0 | 3250.0 | 0 | 2007 |
| 4 | 0 | 2 | 36.7 | 19.3 | 193.0 | 3450.0 | 0 | 2007 |
| 5 | 0 | 2 | 39.3 | 20.6 | 190.0 | 3650.0 | 1 | 2007 |
10.2. scikit-learnを使った決定木の実験#
dtc = DecisionTreeClassifier(max_depth=None,random_state=SEED)
dtc.fit(X_train,y_train)
train_acc=dtc.score(X_train,y_train)
test_acc=dtc.score(X_test,y_test)
print(f"{train_acc=}\n{test_acc=}")
train_acc=1.0
test_acc=0.9880239520958084
10.3. scikit-learnを使ったRandom Forestの実験#
rfc = RandomForestClassifier(
n_estimators=100, # 弱学習器を何個作るか
criterion="gini", # 損失関数をginiやentropyから指定
max_depth=None, # 弱学習器として使った決定木の深さ上限
max_features="sqrt", # ブートストラップサンプルの特徴数。sqrtはsqrt(n_features)
n_jobs=-1, # 何個並列で計算するか。-1は使えるコアを全て使う。
random_state=SEED, # 擬似乱数のSEED
)
rfc.fit(X_train,y_train)
RandomForestClassifier(n_jobs=-1, random_state=20230215)In a Jupyter environment, please rerun this cell to show the HTML representation or trust the notebook.
On GitHub, the HTML representation is unable to render, please try loading this page with nbviewer.org.
RandomForestClassifier(n_jobs=-1, random_state=20230215)
rfc.predict(X_test)
array([0, 2, 0, 0, 0, 2, 0, 2, 1, 2, 2, 0, 0, 2, 0, 2, 2, 0, 1, 2, 1, 2,
0, 1, 0, 2, 0, 1, 2, 0, 0, 2, 2, 1, 2, 2, 0, 2, 2, 0, 2, 0, 0, 2,
2, 2, 1, 2, 0, 0, 0, 0, 2, 1, 0, 0, 1, 0, 2, 0, 2, 2, 2, 0, 0, 0,
2, 2, 1, 0, 2, 0, 0, 2, 0, 2, 0, 2, 0, 0, 2, 0, 1, 0, 0, 2, 0, 1,
0, 2, 2, 2, 1, 2, 0, 0, 0, 2, 2, 0, 0, 0, 2, 2, 0, 2, 2, 2, 0, 1,
0, 1, 0, 1, 2, 1, 2, 2, 2, 0, 2, 0, 1, 1, 0, 1, 2, 2, 2, 1, 0, 0,
2, 1, 0, 0, 1, 1, 2, 2, 1, 0, 0, 2, 0, 1, 1, 0, 0, 2, 0, 2, 0, 1,
1, 0, 2, 2, 2, 1, 2, 2, 1, 2, 1, 1, 0])
rfc.score(X_test,y_test)
0.9820359281437125
10.4. RandomForestClassifierのシンプルな実装例#
NumPyを使ってRandom Forest Classifierを実装しましょう。ただし、sklearnのDecisionTreeClassifierを使います。
import numpy as np
import scipy
from sklearn.tree import DecisionTreeClassifier
from copy import copy
# 最頻値を求める
def mode(Mat: np.ndarray, axis: int = None) -> np.ndarray:
return scipy.stats.mode(Mat, axis=axis, keepdims=False).mode
def get_bootstrap_sample_indices(rng:np.random._generator.Generator, X:np.ndarray, bootstrap_sample_size:int)->np.ndarray:
"""ブートストラップサンプルを一つ作る
Args:
rng (np.random._generator.Generator): 擬似乱数生成器
X (np.ndarray): 二次元配列
bootstrap_sample_size (int): サンプルサイズ
Returns:
np.ndarray: サンプルのindexを持った一次元配列
"""
return rng.integers(low=0, high=X.shape[0],size=bootstrap_sample_size,)
class MyRandomForestClassifier:
def __init__(self,
bootstrap_sample_size: int,
max_features: int = None,
n_estimators: int = 100,
rng: np.random._generator.Generator = np.random.default_rng(
np.random.randint(2**20)),
**estimator_params,
):
self.n_estimators = n_estimators
self.bootstrap_sample_size = bootstrap_sample_size
self.max_features = max_features
self.rng = rng
self.estimator_params = estimator_params
self.estimators_ = []
self.selected_features_ = []
self.is_fitted = False
def fit(self, X, y):
# ブートストラップサンプルを作成
for _x, _y in self.get_bootstrap_sample(X, y):
# 弱識別器の訓練を行う
_estimator = DecisionTreeClassifier(**self.estimator_params,
random_state=self.rng.integers(0, 2**20),)
_estimator.fit(_x, _y)
# 学習済み弱識別器をリストに保存
self.estimators_.append(_estimator)
self.is_fitted = True
return self
def get_bootstrap_sample(self, X: np.ndarray, y: np.ndarray):
"""ブートストラップサンプルを作成し、データとラベルのペアを一つ一つ返すメソッド
"""
if self.is_fitted:
print("warning! 2回目以降のfitです。bootstrap sampleの作り方が初期化されます。")
for _ in range(self.n_estimators):
_sample_data_indices = get_bootstrap_sample_indices(self.rng,X,self.bootstrap_sample_size)
# ランダムに特徴を選択する
_feature_indices = np.arange(X.shape[1])
if self.max_features is not None:
self.rng.shuffle(_feature_indices)
_feature_indices = _feature_indices[:self.max_features]
self.selected_features_.append(_feature_indices)
# ブートストラップサンプルを切り出す
X_sample = X[_sample_data_indices][:, _feature_indices]
y_sample = y[_sample_data_indices]
yield X_sample, y_sample
def predict(self, X):
assert self.is_fitted, "このメソッドは訓練後に利用してください。"
_pred_labels = []
for _index in range(len(self.estimators_)):
# _index番目の弱識別器を使ってXのラベルを推論する
_estimator = self.estimators_[_index]
_feature_indices = self.selected_features_[_index]
_pred_labels.append(_estimator.predict(X[:, _feature_indices]))
_pred_labels = np.vstack(_pred_labels)
# 多数決で予測値を決定する（_pred_labelsの各列の最頻値を返す)
pred_labels = mode(_pred_labels, axis=0)
return pred_labels
def score(self, X, y):
"正答率を計算する"
assert self.is_fitted, "このメソッドは訓練後に利用してください。"
_pred_labels = self.predict(X,)
return (_pred_labels == y).sum()/y.size
rf = MyRandomForestClassifier(
bootstrap_sample_size=int(X_train.shape[0]*0.9),
max_features = int(X_train.shape[1]*0.8),
n_estimators = 100,
rng = np.random.default_rng(SEED),
max_depth=None,
)
rf.fit(X_train,y_train)
train_acc=rf.score(X_train,y_train)
test_acc=rf.score(X_test,y_test)
print(f"{train_acc=}\n{test_acc=}")
train_acc=1.0
test_acc=0.9940119760479041
10.5. RandomForestRegressorのシンプルな実装例#
NumPyを使ってRandom Forest Regressorを実装しましょう。ただし、sklearnのDecisionTreeRegressorを使います。
class MyRandomForestRegressor:
def __init__(self,
bootstrap_sample_size: int,
max_features: int = None,
n_estimators: int = 100,
rng: np.random._generator.Generator = np.random.default_rng(
np.random.randint(2**20)),
**estimator_params,
):
self.n_estimators = n_estimators
self.bootstrap_sample_size = bootstrap_sample_size
self.max_features = max_features
self.rng = rng
self.estimator_params = estimator_params
self.estimators_ = []
self.selected_features_ = []
self.is_fitted = False
def fit(self, X, y):
# ブートストラップサンプルを作成
for _x, _y in self.get_bootstrap_sample(X, y):
# 弱識別器の訓練を行う
_estimator = DecisionTreeRegressor(**self.estimator_params,
random_state=self.rng.integers(0, 2**20),)
_estimator.fit(_x, _y)
# 学習済み弱識別器をリストに保存
self.estimators_.append(_estimator)
self.is_fitted = True
return self
def get_bootstrap_sample(self, X: np.ndarray, y: np.ndarray):
"""ブートストラップサンプルを作成し、データとラベルのペアを一つ一つ返すメソッド
"""
if self.is_fitted:
print("warning! 2回目以降のfitです。bootstrap sampleの作り方が初期化されます。")
for _ in range(self.n_estimators):
_sample_data_indices = get_bootstrap_sample_indices(self.rng,X,self.bootstrap_sample_size)
# ランダムに特徴を選択する
_feature_indices = np.arange(X.shape[1])
if self.max_features is not None:
self.rng.shuffle(_feature_indices)
_feature_indices = _feature_indices[:self.max_features]
self.selected_features_.append(_feature_indices)
# ブートストラップサンプルを切り出す
X_sample = X[_sample_data_indices][:, _feature_indices]
y_sample = y[_sample_data_indices]
yield X_sample, y_sample
def predict(self, X):
assert self.is_fitted, "このメソッドは訓練後に利用してください。"
_pred_labels = []
for _index in range(len(self.estimators_)):
# _index番目の弱識別器を使ってXのラベルを推論する
_estimator = self.estimators_[_index]
_feature_indices = self.selected_features_[_index]
_pred_labels.append(_estimator.predict(X[:, _feature_indices]))
_pred_labels = np.vstack(_pred_labels)
# 平均で予測値を決定する（_pred_labelsの各列の平均を返す)
pred_labels = np.mean(_pred_labels, axis=0)
return pred_labels
def score(self, X, y):
"正答率を計算する"
assert self.is_fitted, "このメソッドは訓練後に利用してください。"
_pred_labels = self.predict(X,)
return (_pred_labels == y).sum()/y.size