Python全栈+人工智能:回归算法应用笔记

美国工资水平预测：

adults = pd.read_csv('../data/adults.txt')
adults.head()

data = adults[['age', 'workclass', 'education',  'marital_status', 'occupation',  'race', 'sex', 'hours_per_week', 'native_country']].copy()
target = adults.salary

cols = [ 'workclass', 'education',  'marital_status', 'occupation',  'race', 'sex', 'native_country']
for col in cols:
    unique = data[col].unique()
    def convert(item):
        return np.argwhere(unique==item)[0,0]
    data[col] = data[col].map(convert)

X_train, X_test, y_train, y_test = train_test_split(data, target, test_size=0.2, random_state=1) # 通过seed控制是否是伪随机

np.random.seed(2)
np.random.rand(1)

knn = KNeighborsClassifier()
knn.fit(X_train, y_train)

knn.score(X_train, y_train)

y_ = knn.predict(X_test)
knn.score(X_test, y_test)

pd.crosstab(index=y_test, columns=y_, rownames=['真实值'], colnames=['预测值'], margins=True)

上述结果精度不够，因为数据没有标准化

改进：

定义标准化函数：

def normalized(x):
    return (x - x.min()) / (x.max() - x.min())

for col in data.columns:
    data[col] = data[col].transform(normalized)

再进行预测：

X_train, X_test, y_train, y_test = train_test_split(data, target, test_size=0.2, random_state=1)

knn = KNeighborsClassifier()
knn.fit(X_train, y_train)

knn.score(X_train, y_train)

y_ = knn.predict(X_test)
knn.score(X_test, y_test)