交差検証

Python によるモデル検証

Kasey Jones

Data Scientist

交差検証

学習、検証、テスト用データは全体を3分割して作成します。例: 学習60%、検証20%、テスト20%。

Python によるモデル検証

交差検証

交差検証では、データを1回の80:20ではなく、複数回に分割します。例: 5通りの80:20分割。

Python によるモデル検証

n_splits: 分割数(交差検証の折数)

shuffle: 分割前にシャッフルするか

random_state: 乱数シード

from sklearn.model_selection import KFold

X = np.array(range(40))
y = np.array([0] * 20 + [1] * 20)

kf = KFold(n_splits=5)

splits = kf.split(X)
Python によるモデル検証
kf = KFold(n_splits=5)
splits = kf.split(X)

for train_index, test_index in splits: print(len(train_index), len(test_index))
32 8 32 8 32 8 32 8 32 8
# インデックス集合の一例を表示
print(train_index, test_index)
[ 0  1  2  3  4  5  6  7  8  9 10 11 12 13 14 15 16 17 18 19 20 ...]
[32 33 34 35 36 37 38 39]
Python によるモデル検証
rfr = RandomForestRegressor(n_estimators=25, random_state=1111)

errors = [] for train_index, val_index in splits: X_train, y_train = X[train_index], y[train_index] X_val, y_val = X[val_index], y[val_index] rfr.fit(X_train, y_train) predictions = rfr.predict(X_val) errors.append(<some_accuracy_metric>)
print(np.mean(errors))
4.25
Python によるモデル検証

演習の時間

Python によるモデル検証

Preparing Video For Download...