scikit-learn

scikit-learn

常用

• 划分验证集

from sklearn.cross_validation import train_test_split
X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=0.2, random_state=0)

• Grid Search

param_grid = {'n_estimators': [300, 500], 'max_features': [10, 12, 14]}
model = grid_search.GridSearchCV(estimator=rfr, param_grid=param_grid, n_jobs=1, cv=10, verbose=20, scoring=RMSE)
model.fit(X_train, y_train)

• LabelEncoder

from sklearn.preprocessing import LabelEncoder
le = LabelEncoder()
le.fit([1,5,67,100])
le.transform([1,1,100,67,5])
#输出： array([0,0,3,2,1])
#可通过le.inverse_transform(x)转换回去

• sklearn.utils.shuffle(多个数组按同样顺序打乱)

def fill_feed_dict(data_X, data_Y, batch_size):
    """Generator datasets to yield batches"""
    # Shuffle data first.
    shuffled_X, shuffled_Y = shuffle(data_X, data_Y)
    for idx in range(data_X.shape[0] // batch_size):
        x_batch = shuffled_X[batch_size * idx: batch_size * (idx + 1)]
        y_batch = shuffled_Y[batch_size * idx: batch_size * (idx + 1)]
        yield x_batch, y_batch

算法

• GBDT（MART）迭代决策树入门教程 | 简介

实现

• 常用算法调用(LR/ RF/ GBDT/ knn/ SVM)
• logistic回归
• 皮尔逊相关度
• 利用kmeans对图片颜色聚类并可视化
• 高维数据可视化tSNE

Choosing the right estimator