【问题4】：kaggle练习题《自行车租赁业务预测》--带数据分析，用了随机森林，支持向量机，岭回归等

不念不忘少年蓝@ 2022-03-19 02:08 156阅读 0赞

数据来源：https://www.kaggle.com/c/bike-sharing-demand

### **第一步：读取数据    并对数据进行分析** ###

import numpy as np
    import pandas as pd
    
    df_train = pd.read_csv('data/kaggle_bike_competition_train.csv')
    print(df_train.head())
    print(df_train.shape)
    print(df_train.dtypes)  # 看一下各字段类型
    # 看时候有缺失值
    print(df_train.isnull().sum().sum())  # 输出为0，表示没有缺失值

读取出来的数据   下面这个表columns名与数据错一位。 datetime对应的是日期那一列，第一列是序号。

<table> 
 <thead> 
  <tr> 
   <th>datetime</th> 
   <th>season</th> 
   <th>holiday</th> 
   <th>workingday</th> 
   <th>weather</th> 
   <th>temp</th> 
   <th>atemp</th> 
   <th>humidity</th> 
   <th>windspeed</th> 
   <th>casual</th> 
   <th>registered</th> 
   <th>count</th> 
  </tr> 
 </thead> 
 <tbody> 
  <tr> 
   <th>0</th> 
   <td>2011-01-01 00:00:00</td> 
   <td>1</td> 
   <td>0</td> 
   <td>0</td> 
   <td>1</td> 
   <td>9.84</td> 
   <td>14.395</td> 
   <td>81</td> 
   <td>0.0000</td> 
   <td>3</td> 
   <td>13</td> 
   <td>16</td> 
  </tr> 
  <tr> 
   <th>1</th> 
   <td>2011-01-01 01:00:00</td> 
   <td>1</td> 
   <td>0</td> 
   <td>0</td> 
   <td>1</td> 
   <td>9.02</td> 
   <td>13.635</td> 
   <td>80</td> 
   <td>0.0000</td> 
   <td>8</td> 
   <td>32</td> 
   <td>40</td> 
  </tr> 
  <tr> 
   <th>2</th> 
   <td>2011-01-01 02:00:00</td> 
   <td>1</td> 
   <td>0</td> 
   <td>0</td> 
   <td>1</td> 
   <td>9.02</td> 
   <td>13.635</td> 
   <td>80</td> 
   <td>0.0000</td> 
   <td>5</td> 
   <td>27</td> 
   <td>32</td> 
  </tr> 
  <tr> 
   <th>3</th> 
   <td>2011-01-01 03:00:00</td> 
   <td>1</td> 
   <td>0</td> 
   <td>0</td> 
   <td>1</td> 
   <td>9.84</td> 
   <td>14.395</td> 
   <td>75</td> 
   <td>0.0000</td> 
   <td>3</td> 
   <td>10</td> 
   <td>13</td> 
  </tr> 
  <tr> 
   <th>4</th> 
   <td>2011-01-01 04:00:00</td> 
   <td>1</td> 
   <td>0</td> 
   <td>0</td> 
   <td>1</td> 
   <td>9.84</td> 
   <td>14.395</td> 
   <td>75</td> 
   <td>0.0000</td> 
   <td>0</td> 
   <td>1</td> 
   <td>1</td> 
  </tr> 
 </tbody> 
</table>

从上面得出了各字段的类型，并且得出没有缺失值。。。因为最后一行代码输出为0.

![watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L3NoYXdyb2FkODg_size_16_color_FFFFFF_t_70][]

### **第二步：处理一些字段    如：datetime那一列。**  ###

# 把月，日和小时单独拎出来，搞成三列
    df_train['month'] = pd.DatetimeIndex(df_train.datetime).month
    df_train['day'] = pd.DatetimeIndex(df_train.datetime).dayofweek
    df_train['hour'] = pd.DatetimeIndex(df_train.datetime).hour
    # print(df_train.head())
    
    # 那个，既然时间大串已经被我们处理过了，那这个字段放着太占地方，干脆就不要了吧
    # 先上一个粗暴的版本，咱们把注册租户和未注册租户也先丢掉，回头咱们再看另外一种处理方式
    
    # 保险起见，保存一下刚才处理的数据
    df_train_origin = df_train
    # 删除掉不需要的字段
    df_train = df_train.drop(['datetime', 'casual', 'registered'], axis=1)
    # print(df_train.head())  # 查看处理后的数据
    # print(df_train.shape)   # 显然列变为了12
    
    # 开始分数据  count是标签
    df_train_target = df_train['count'].values
    df_train_data = df_train.drop(['count'], axis=1).values

这一步我们将日期拆开，分别增加三列：month, day, hour。 并且把注册租户和未注册租户以及原先datetime列删除 。。防止他们干扰。  最后得出 特征值和标签值。

### **第三步：模型走起** ###

**1. 岭回归**

print("岭回归\n")
    kfold = KFold(n_splits=3, shuffle=True, random_state=42)
    for train, test in kfold.split(df_train_data):
        clf = Ridge()
        clf.fit(df_train_data[train], df_train_target[train])
        print("训练集准确率:{},测试集准确率:{}".format(
            clf.score(df_train_data[train], df_train_target[train]),
            clf.score(df_train_data[test], df_train_target[test])
        ))

先进行3折交叉验证，然后用岭回归进行训练。  最后得出的结果很尴尬

![2019021416000947.png][]

**    2.支持向量机模型**

print("支持向量机\n")
    for train, test in kfold.split(df_train_data):
        clf = SVR(kernel='rbf', C=10, gamma=0.001)
        clf.fit(df_train_data[train], df_train_target[train])
        print("训练集准确率:{},测试集准确率:{}".format(
            clf.score(df_train_data[train], df_train_target[train]),
            clf.score(df_train_data[test], df_train_target[test])
        ))

结果：

![2019021416073719.png][]

这个结果也挺尴尬 。 支持向量机曾几何时也是神器。

**    3.随机森林**

print("随机森林\n")
    for train, test in kfold.split(df_train_data):
        clf = RandomForestRegressor(n_estimators=80)
        clf.fit(df_train_data[train], df_train_target[train])
        print("训练集准确率:{},测试集准确率:{}".format(
            clf.score(df_train_data[train], df_train_target[train]),
            clf.score(df_train_data[test], df_train_target[test])
        ))

结果还是不错的。

![20190214161103691.png][]

**4.用网格搜索给随机森林找一组参数，然后在用随机森林预测**

x_train, x_test, y_train, y_test = train_test_split(df_train_data, df_train_target, test_size=0.2, random_state=42)
    params = {"n_estimators": [30, 60, 90]}
    scores = ['r2']
    for score in scores:
        print(score)
        clf = GridSearchCV(RandomForestRegressor(), params, cv=5, scoring=score)
        clf.fit(x_train, y_train)
        print(clf.best_estimator_)

找到了一组合适的参数。如下

![20190214161518647.png][]

我们用这组参数再次进行随机森林预测

print("随机森林\n")
    for train, test in kfold.split(df_train_data):
        clf = RandomForestRegressor(bootstrap=True, criterion='mse', max_depth=None,
               max_features='auto', max_leaf_nodes=None,
               min_impurity_decrease=0.0, min_impurity_split=None,
               min_samples_leaf=1, min_samples_split=2,
               min_weight_fraction_leaf=0.0, n_estimators=90, n_jobs=None,
               oob_score=False, random_state=None, verbose=0, warm_start=False)
        clf.fit(df_train_data[train], df_train_target[train])
        print("训练集准确率:{},测试集准确率:{}".format(
            clf.score(df_train_data[train], df_train_target[train]),
            clf.score(df_train_data[test], df_train_target[test])
        ))

![20190214163519236.png][]

### **附加： 走一波数据分析** ###

画出几个特征与流通量的关系

# 风速与流通量
    df_train_origin.groupby('windspeed').mean().plot(y='count', marker='o')
    plt.show()
    
    # 湿度与流通量
    df_train_origin.groupby('humidity').mean().plot(y='count', marker='o')
    plt.show()
    
    # 温度与流通量
    df_train_origin.groupby('temp').mean().plot(y='count', marker='o')
    plt.show()
    
    
    # 温度湿度变化
    df_train_origin.plot(x='temp', y='humidity', kind='scatter')
    plt.show()

![watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L3NoYXdyb2FkODg_size_16_color_FFFFFF_t_70 1][]

![watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L3NoYXdyb2FkODg_size_16_color_FFFFFF_t_70 2][]

![watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L3NoYXdyb2FkODg_size_16_color_FFFFFF_t_70 3][]

![watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L3NoYXdyb2FkODg_size_16_color_FFFFFF_t_70 4][]

画出各个维度与count的关系

fig, axs = plt.subplots(nrows=2, ncols=3)
    df_train_origin.plot(kind='scatter', x='temp', y='count', ax=axs[0, 0], figsize=(16, 8), color='magenta')
    df_train_origin.plot(kind='scatter', x='atemp', y='count', ax=axs[0, 1], color='cyan')
    df_train_origin.plot(kind='scatter', x='humidity', y='count', ax=axs[0, 2], color='red')
    df_train_origin.plot(kind='scatter', x='windspeed', y='count', ax=axs[1, 0], color='yellow')
    df_train_origin.plot(kind='scatter', x='month', y='count', ax=axs[1, 1], color='blue')
    df_train_origin.plot(kind='scatter', x='hour', y='count', ax=axs[1, 2], color='green')
    plt.show()

![watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L3NoYXdyb2FkODg_size_16_color_FFFFFF_t_70 5][]

画出各个特征之间的相关性

corr = df_train_origin[['temp', 'weather', 'windspeed', 'day', 'month', 'hour', 'count']].corr()
    print(corr)
    
    
    # 用颜色深浅来表示相关度
    
    plt.matshow(corr)
    plt.colorbar()
    plt.show()

相关性矩阵     带负号表示负相关

![20190214164451611.png][]

![watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L3NoYXdyb2FkODg_size_16_color_FFFFFF_t_70 6][] 由上向下，相关性一次减小。。

# 本次就分析到这里 。。。未完待续   连载中。。。。 #

[watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L3NoYXdyb2FkODg_size_16_color_FFFFFF_t_70]: /images/20220319/50e0aa11e3b1449da5276d5114a7777d.png
[2019021416000947.png]: /images/20220319/3404a6d6190e4602b9eba55b12779f4d.png
[2019021416073719.png]: /images/20220319/96f275a07acb4194a09494514d7208c7.png
[20190214161103691.png]: /images/20220319/b83fb14840cb4b81b22b195a5fe922c7.png
[20190214161518647.png]: /images/20220319/c787a60ad0ef4aed9911e939651913c2.png
[20190214163519236.png]: /images/20220319/ed44cc9bb0be4414be69d20c750b306c.png
[watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L3NoYXdyb2FkODg_size_16_color_FFFFFF_t_70 1]: /images/20220319/65aecff1ca554788aeb484b132c37368.png
[watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L3NoYXdyb2FkODg_size_16_color_FFFFFF_t_70 2]: /images/20220319/03c47cc0fc9b4dbe9e3fc4a4eb8756c6.png
[watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L3NoYXdyb2FkODg_size_16_color_FFFFFF_t_70 3]: /images/20220319/c1cc093e799e428cb60ba0d3e3964b43.png
[watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L3NoYXdyb2FkODg_size_16_color_FFFFFF_t_70 4]: /images/20220319/c3fa4ec5a2fb4eccb0c8492cdc399aad.png
[watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L3NoYXdyb2FkODg_size_16_color_FFFFFF_t_70 5]: /images/20220319/5eccbe9fa1674bb895e22cf0a240745b.png
[20190214164451611.png]: /images/20220319/fb3e7004187640629339456990572c3e.png
[watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L3NoYXdyb2FkODg_size_16_color_FFFFFF_t_70 6]: /images/20220319/c06ab267e87a43d79d0443a71ca3c650.png