【机器学习】2.1.一元回归模型 房产成交价格预测

包含全部示例的代码仓库见GIthub

1 导入库

import pandas as pd
import numpy as np
import matplotlib.pyplot as plt
%matplotlib inline
from sklearn.linear_model import LinearRegression
from sklearn.preprocessing import PolynomialFeatures

2 数据准备

data_list = []
for i in range(1,8):
    try:
        data = pd.read_csv('./dataset/lianjia{}.csv'.format(i), encoding='gbk')
    except:
        data = pd.read_csv('./dataset/lianjia{}.csv'.format(i))
    finally:
        data_list.append(data)
data = pd.concat(data_list)
data = data.dropna()
data.info()
# output
<class 'pandas.core.frame.DataFrame'>
Int64Index: 144531 entries, 0 to 6659
Data columns (total 14 columns):
 #   Column          Non-Null Count   Dtype  
---  ------          --------------   -----  
 0   cjtaoshu        144531 non-null  int64  
 1   mendian         144531 non-null  object 
 2   cjzongjia       144531 non-null  float64
 3   zhiwei          144531 non-null  object 
 4   haoping         144531 non-null  object 
 5   cjdanjia        144531 non-null  object 
 6   cjxiaoqu        144531 non-null  object 
 7   xingming        144531 non-null  object 
 8   cjzhouqi        144531 non-null  object 
 9   biaoqian        144531 non-null  object 
 10  cjlouceng       144531 non-null  object 
 11  cjshijian       144531 non-null  object 
 12  congyenianxian  144531 non-null  object 
 13  bankuai         144531 non-null  object 
dtypes: float64(1), int64(1), object(12)
memory usage: 16.5+ MB

data.head()
# output
cjtaoshu	mendian	cjzongjia	zhiwei	haoping	cjdanjia	cjxiaoqu	xingming	cjzhouqi	biaoqian	cjlouceng	cjshijian	congyenianxian	bankuai
0	37	红莲北里店	251.0	店经理	97% 141	43997元/平	红莲北里 3室1厅 57平	郭海龙	36	房东信赖;销售达人;带看活跃	南 北/高楼层/6层	签约时间：2015-05-24	4-5年	马连道
1	37	红莲北里店	159.0	店经理	97% 141	36969元/平	红莲南里 1室1厅 43平	郭海龙	36	房东信赖;销售达人;带看活跃	南/高楼层/7层	签约时间：2015-05-10	4-5年	马连道
2	37	红莲北里店	257.0	店经理	97% 141	39046元/平	常青藤嘉园 1室1厅 65平	郭海龙	36	房东信赖;销售达人;带看活跃	北/低楼层/16层	签约时间：2015-04-26	4-5年	马连道
3	37	红莲北里店	243.0	店经理	97% 141	41313元/平	红莲北里 2室1厅 58平	郭海龙	36	房东信赖;销售达人;带看活跃	南 北/高楼层/6层	签约时间：2015-04-04	4-5年	马连道
4	37	红莲北里店	372.5	店经理	97% 141	42053元/平	广安门外大街 3室1厅 88平	郭海龙	36	房东信赖;销售达人;带看活跃	东 南 西 北/中楼层/18层	签约时间：2015-04-01	4-5年	马连道

data.cjdanjia = np.round(data.cjdanjia.str.replace('元/平','').astype(np.float32).map(lambda x:x/10000),2)
data.cjdanjia
# output
0       4.40
1       3.70
2       3.90
3       4.13
4       4.21
        ... 
6653    2.28
6656    8.10
6657    1.91
6658    1.21
6659    1.18

for value in data.cjshijian:# 遍历并拿到数据
    print(type(value))
    print(value[5:])
# output
<class 'str'>
2015-05-24
<class 'str'>
2015-05-10
<class 'str'>
2015-04-26

data.cjshijian = pd.to_datetime(data.cjshijian.map(lambda x:x[5:]))
data.cjshijian
# output
0      2015-05-24
1      2015-05-10
2      2015-04-26
3      2015-04-04
4      2015-04-01
          ...    
6653   2016-07-13
6656   2016-09-11
6657   2016-04-10
6658   2016-01-29
6659   2015-12-20

data = data[data.cjxiaoqu.str.contains('远洋山水')]
data = data.sort_values(by = 'cjshijian')
data = data.set_index('cjshijian')['2012':]
plt.figure(figsize=(10,8))
plt.scatter(data.index, data.cjdanjia)

data = data[data.cjdanjia>1]
(data.index - pd.to_datetime('2012-01-01')).days
# output
Int64Index([   6,    6,   43,   45,   49,   50,   50,   50,   50,   55,
            ...
            1727, 1728, 1729, 1729, 1729, 1759, 1763, 1763, 1765, 1771],
           dtype='int64', name='cjshijian', length=481)

处理成时间序列

data = data.assign(time = (data.index - pd.to_datetime('2012-01-01')).days)  # assign 添加一列
data = data[['time', 'cjdanjia']]
data
# output
          time	cjdanjia
cjshijian		
2012-01-07	6	2.54
2012-01-07	6	2.54
2012-02-13	43	2.22
2012-02-15	45	2.06
2012-02-19	49	2.23
...	...	...
2016-10-25	1759	6.43
2016-10-29	1763	6.47
2016-10-29	1763	6.62
2016-10-31	1765	6.60
2016-11-06	1771	8.22

plt.figure(figsize=(10,8))
plt.scatter(data.time, data.cjdanjia)

X = data.time
Y = data.cjdanjia

3 模型构建

3.1 y = ax+b

model = LinearRegression()
model.fit(pd.DataFrame(X),Y)  # sklearn 与 dataframe比较契合，x转为dataframe就不会报错

x = [[0], [1800]]
type(x)
# output
list

y = model.predict(pd.DataFrame(x))
plt.figure(figsize=(10,8))
plt.plot(x, y, color='r')
plt.scatter(data.time, data.cjdanjia)

3.2 y = ax2+bx+c

model2 = LinearRegression()
model2.fit(pd.DataFrame({'x2':X*X, 'x1':X}), Y)
x = np.linspace(0, 1800)
y = model2.predict(pd.DataFrame({'x2':x*x, 'x1':x}))
plt.figure(figsize=(10,8))
plt.plot(x, y, color='r')  # 测试数据绘图
plt.scatter(data.time, data.cjdanjia)

训练数据绘图

model2.predict(pd.DataFrame({'x2':X*X, 'x1':X})).shape
model2.predict(pd.DataFrame({'x2':x*x, 'x1':x})).shape
y = model2.predict(pd.DataFrame({'x2':X*X, 'x1':X}))  # X（训练数据）绘图
plt.figure(figsize=(10,8))
plt.plot(X, y, color='r')
plt.scatter(data.time, data.cjdanjia)

3.3 y = ax3+bx2+x+c

q3 = PolynomialFeatures(degree=3)  # 转换成3阶多项式
X3 = q3.fit_transform(pd.DataFrame(X))
model3 = LinearRegression()
model3.fit(X3, Y)

x = np.linspace(0, 1800)
x_ = q3.fit_transform(pd.DataFrame(x))
y = model3.predict(x_)
plt.figure(figsize=(10,8))
plt.plot(x, y, color='r')
plt.scatter(data.time, data.cjdanjia)

3.4 4阶多项式

q4 = PolynomialFeatures(degree=4)  # 转换成4阶多项式
X4 = q4.fit_transform(pd.DataFrame(X))
model4 = LinearRegression()
model4.fit(X4, Y)
x = np.linspace(0, 1800)
x_ = q4.fit_transform(pd.DataFrame(x))
y = model4.predict(x_)
plt.figure(figsize=(10,8))
plt.plot(x, y, color='r')
plt.scatter(data.time, data.cjdanjia)

4 测试模型

X_train, X_test = X[:'2016-5'], X['2016-5':]
Y_train, Y_test = Y[:'2016-5'], Y['2016-5':]

q3 = PolynomialFeatures(degree=3)  # 转换成3阶多项式
X3 = q3.fit_transform(pd.DataFrame(X_train))
model3 = LinearRegression()
model3.fit(X3, Y_train)
X3_ = q3.fit_transform(pd.DataFrame(X_test))
np.sum((model3.predict(X3_) - Y_test)**2)
# output
78.51184688848316

q4 = PolynomialFeatures(degree=4)  # 转换成4阶多项式
X4 = q4.fit_transform(pd.DataFrame(X_train))
model4 = LinearRegression()
model4.fit(X4, Y_train)X4_ = q4.fit_transform(pd.DataFrame(X_test))
np.sum((model4.predict(X4_) - Y_test)**2)
# output
68.29625037920916

q5 = PolynomialFeatures(degree=5)  # 转换成5阶多项式
X5 = q5.fit_transform(pd.DataFrame(X_train))
model5 = LinearRegression()
model5.fit(X5, Y_train)
X5_ = q5.fit_transform(pd.DataFrame(X_test))
np.sum((model5.predict(X5_) - Y_test)**2)
# output
140.7305646522754

q2 = PolynomialFeatures(degree=2)  # 转换成2阶多项式
X2 = q2.fit_transform(pd.DataFrame(X_train))
model2 = LinearRegression()
model2.fit(X2, Y_train)
X2_ = q2.fit_transform(pd.DataFrame(X_test))
np.sum((model2.predict(X2_) - Y_test)**2)
# output
258.2695617180388

5 改变训练集大小

X_train, X_test = X['2016-1':'2016-5'], X['2016-5':]
Y_train, Y_test = Y['2016-1':'2016-5'], Y['2016-5':]

model = LinearRegression()
model.fit(X_train.values.reshape(-1,1), Y_train)
np.sum((model.predict(X_test.values.reshape(-1,1)) - Y_test)**2)
# output
67.65526416726503

q2 = PolynomialFeatures(degree=2)  # 转换成2阶多项式
X2 = q2.fit_transform(pd.DataFrame(X_train))
model2 = LinearRegression()
model2.fit(X2, Y_train)
X2_ = q2.fit_transform(pd.DataFrame(X_test))
np.sum((model2.predict(X2_) - Y_test)**2)
# output
343.769340055577

q3 = PolynomialFeatures(degree=3)  # 转换成3阶多项式
X3 = q3.fit_transform(pd.DataFrame(X_train))
model3 = LinearRegression()
model3.fit(X3, Y_train)
X3_ = q3.fit_transform(pd.DataFrame(X_test))
np.sum((model3.predict(X3_) - Y_test)**2)
# output
38874.83884770577

q4 = PolynomialFeatures(degree=4)  # 转换成4阶多项式
X4 = q4.fit_transform(pd.DataFrame(X_train))
model4 = LinearRegression()
model4.fit(X4, Y_train)
X4_ = q4.fit_transform(pd.DataFrame(X_test))
np.sum((model4.predict(X4_) - Y_test)**2)
# output
203047.92547841667