sklearn.datasets
.1 生成数据
from sklearn import datasets
团状样本点集
X, y = datasets.make_blobs(n_samples=1000,
centers=10, # 团的个数
# 默认随机生成分布, 也可以自定义质心
# 例如 centers=[[1,2], [2,3]]
n_features=2, # 数据维度
cluster_std=0.5,
random_state=4,
s=10 # 样本点的大小
)
内外两个环状分布的样本点集
X, y = datasets.make_circles(n_samples=1000,
factor=0.5, # 内外环距离，取值 [0, 1)
# ->0, 内环在中央聚合成一个团; ->1, 内环外扩与外环合并
noise=0.05 # 噪音
)
其他数据类型
波士顿房价数据
X, y = datasets.load_boston(return_X_y=True) # X, y np.array()
一般情况下会把 X 转化为 pd.DataFrame()，因此需要给 Column Headers 赋值（来源于官方文档）
X = pd.DataFrame(X)
X.columns = [...]
sklearn.linear_model
from sklearn.linear_model import LinearRegression
lr = LinearRegression()
lr.fit(X, y)
lr.coef_ # 系数
lr.intercept_ # 截距
sklearn.matrics
.1 Value the clustering 检查聚类算法的性能
sklearn.matrics.silhouette_score(X_blobs, # 原数据
class_prediction # 分类标签, 例如以下表示有三类 array([0, 1, 2, 1, 1, 2])
)
sklearn.model_selection
from sklearn.model_selection import ...
划分训练集/测试集
# By deafult, sklearn use 75/25 split
X_train, X_test, y_train, y_test = train_test_split(X, y, random_state=0)
sklearn.processing
from sklearn.processing import ...
Scale Data
LINKs Back Jupyter Notebook: Simple_neural_network_with_sklearn
Standard Scale \(z=\frac{x-\mu}{\sigma}\)
scaler = StandardScaler()
# Scale X_train
# 基于 X_train 的 mu/sigma 创建一个 Scale transform，并应用
X_train_scaled = scaler.fit_transform(X_train)
# Scale X_test: 把由 X_train 创建的 transfrom 应用于 X_test
x_test_scaled = scaler.transform(X_test)
poly = PolynomialFeatures(degree=2)
sklearn.neural_network
from sklearn.neural_network import ...
Standard multi-layer perceptron regressor
reg = MLPRegressor(random_state=0, max_iter = 100000, learning_rate_init=0.0001)
learning_rate_init: 学习速率（步长），一般取 [0.0001, 0.01]
reg.fit(X_train_scaled, y_train)
reg.score(X_test_scaled, y_test) # return R2 score
Document Information
- Author: Zeka Lee
- Link: https://zhekaili.github.io/wiki/python-libs/python-lib-sklearn/
- Copyright: 自由转载-非商用-非衍生-保持署名（创意共享3.0许可证）