如何选择机器学习模型进行数据分析_简要笔记

Supervised 监督学习
Unsuperivised 非监督学习
Reinforcement 强化学习(alphago,我将Action给环境，环境给我Reward))

Supervised Learning
- Classification 分类
- Regression 回归

Unsupervised Learning
- Clustering 聚类
- Compression 降维(压缩)
如何选择一个模型

大体流程

```
//训练集x(N * d), y(N * 1); 测试集x, y(同分布的)
train_x, train_y, test_x, test_y = getData()  // MNIST

model = somemodel()            // SVM()，LASSO()
model.fit(train_x, train_y)       // 学习参数
predictions = model.predict(test_x)  // 预测模型

//验证模型
//分类任务: 分对的百分比
//回归任务: 计算MSE等
//11种评价指标
//一个预测的数，和真实数据之间的差距
score = score_function(test_y, predictions)  
```

# iris 花的数据集
from sklearn.datasets import load_iris
iris = load_iris()
from sklearn.datasets import load_iris
iris = load_iris()

# The resulting dataset is a Bunch object: you can see what's available using the method keys():

In [4]: iris.keys()

Out[4]:
dict_keys(['target', 'data', 'target_names', 'DESCR', 'feature_names'])

Generating Synthetic Data 自己造数据

In [5]: from sklearn.datasets import make_regression # 自己造出来数据

I. Supervised 监督学习

estimator.fit(X_train, y_train)
estimator.predict(X_test)
estimator.score(X_test, y_test) #评估

监督学习的工作流程

II.Unsupervised Transformers (无监督)

transformer.fit(X_train) # 进行一个拟合
X_train_transf = transformer.transform(X_train) # 降维到所期待的维度
X_test_transf = transformer.transform(X_test) # n component, trans = TSN(n_col=3, [..])

Feature Scaling (特征规则化)
min-max scalling
"normalization" (归到0~1之间)
Principal Component Analysis (PCA)

把一个高维空间映射到

PCA for Dimensionality Reduction (降维)
K-means Clustering (聚类手法)
- 需要告诉他分成 N 类
- 欧几里得距离 (

# ### Scikit-learn API
estimator.fit(X_train, [y_train])
estimator.predict(X_test)  estimator.transform(X_test)
Classification             Preprocessing (预处理：图像灰度值归一化等 )
Regression                 Dimensinality Reduction (降维)
Clustering                 Feature Extraction （特征提取）
                           Feature selection   (特征选择)