1 读取数据
import pandas as pd
import numpy as np
import warnings
warnings.filterwarnings('ignore')
note: 将整型变量的类型尽量压缩,逐步判断并转化为int8,int16,int32,int64
def reduce_mem_usage(df):
""" iterate through all the columns of a dataframe and modify the data type
to reduce memory usage.
"""
start_mem = df.memory_usage().sum()
print('Memory usage of dataframe is {:.2f} MB'.format(start_mem))
for col in df.columns:
col_type = df[col].dtype
if col_type != object:
c_min = df[col].min()
c_max = df[col].max()
if str(col_type)[:3] == 'int':
if c_min > np.iinfo(np.int8).min and c_max < np.iinfo(np.int8).max:
df[col] = df[col].astype(np.int8)
elif c_min > np.iinfo(np.int16).min and c_max < np.iinfo(np.int16).max:
df[col] = df[col].astype(np.int16)
elif c_min > np.iinfo(np.int32).min and c_max < np.iinfo(np.int32).max:
df[col] = df[col].astype(np.int32)
elif c_min > np.iinfo(np.int64).min and c_max < np.iinfo(np.int64).max:
df[col] = df[col].astype(np.int64)
else:
if c_min > np.finfo(np.float16).min and c_max < np.finfo(np.float16).max:
df[col] = df[col].astype(np.float16)
elif c_min > np.finfo(np.float32).min and c_max < np.finfo(np.float32).max:
df[col] = df[col].astype(np.float32)
else:
df[col] = df[col].astype(np.float64)
else:
df[col] = df[col].astype('category')
end_mem = df.memory_usage().sum()
print('Memory usage after optimization is: {:.2f} MB'.format(end_mem))
print('Decreased by {:.1f}%'.format(100 * (start_mem - end_mem) / start_mem))
return df
sample_feature = reduce_mem_usage(pd.read_csv('data_for_tree.csv'))
Memory usage of dataframe is 62099672.00 MB
Memory usage after optimization is: 16520303.00 MB
Decreased by 73.4%
sample_feature.head()
SaleID | name | model | brand | bodyType | fuelType | gearbox | power | kilometer | notRepairedDamage | ... | used_time | city | brand_amount | brand_price_max | brand_price_median | brand_price_min | brand_price_sum | brand_price_std | brand_price_average | power_bin | |
---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
0 | 0 | 736 | 30.0 | 6 | 1.0 | 0.0 | 0.0 | 60 | 12.5 | 0.0 | ... | 4384.0 | 1.0 | 10192.0 | 35990.0 | 1800.0 | 13.0 | 36457520.0 | 4564.0 | 3576.0 | 5.0 |
1 | 1 | 2262 | 40.0 | 1 | 2.0 | 0.0 | 0.0 | 0 | 15.0 | - | ... | 4756.0 | 4.0 | 13656.0 | 84000.0 | 6400.0 | 15.0 | 124044600.0 | 8992.0 | 9080.0 | NaN |
2 | 2 | 14874 | 115.0 | 15 | 1.0 | 0.0 | 0.0 | 163 | 12.5 | 0.0 | ... | 4384.0 | 2.0 | 1458.0 | 45000.0 | 8496.0 | 100.0 | 14373814.0 | 5424.0 | 9848.0 | 16.0 |
3 | 3 | 71865 | 109.0 | 10 | 0.0 | 0.0 | 1.0 | 193 | 15.0 | 0.0 | ... | 7124.0 | NaN | 13992.0 | 92900.0 | 5200.0 | 15.0 | 113034208.0 | 8248.0 | 8076.0 | 19.0 |
4 | 4 | 111080 | 110.0 | 5 | 1.0 | 0.0 | 0.0 | 68 | 5.0 | 0.0 | ... | 1531.0 | 6.0 | 4664.0 | 31500.0 | 2300.0 | 20.0 | 15414322.0 | 3344.0 | 3306.0 | 6.0 |
5 rows × 39 columns
continuous_feature_names = [x for x in sample_feature.columns if x not in ['price','brand','model']]
continuous_feature_names
['SaleID',
'name',
'bodyType',
'fuelType',
'gearbox',
'power',
'kilometer',
'notRepairedDamage',
'seller',
'offerType',
'v_0',
'v_1',
'v_2',
'v_3',
'v_4',
'v_5',
'v_6',
'v_7',
'v_8',
'v_9',
'v_10',
'v_11',
'v_12',
'v_13',
'v_14',
'train',
'used_time',
'city',
'brand_amount',
'brand_price_max',
'brand_price_median',
'brand_price_min',
'brand_price_sum',
'brand_price_std',
'brand_price_average',
'power_bin']
2 线性回归 & 五折交叉验证 & 模拟真实业务情况
下面一段代码是对sample_feature(汽车有尚未修复的损坏)进行转化,这货原来是什么情况?来看一下
sample_feature.notRepairedDamage.value_counts()
0.0 147809
- 32251
1.0 18977
Name: notRepairedDamage, dtype: int64
notRepairedDamage原来是category类型,将列转换成float32类型。首先用0代替-,应该是0代表没有损坏,而1是有损坏
sample_feature = sample_feature.dropna().replace('-', 0).reset_index(drop=True)
sample_feature['notRepairedDamage'] = sample_feature['notRepairedDamage'].astype(np.float32)
train = sample_feature[continuous_feature_names + ['price']]
train_X = train[continuous_feature_names]
train_y = train['price']
然后将特征和标签提取出来了
简单建模
所谓简单就是线性回归了
from sklearn.linear_model import LinearRegression
虽然简单,但是这里输入参数normalize=True啥意思呢?就是归一化的意思,每个值先减平均数,然后再除以二范数
model = LinearRegression(normalize=True)
model = model.fit(train_X, train_y)
查看训练的线性回归模型的截距(intercept)与权重(coef)
下面这段代码就是求出模型的截距和系数,有一点需要自己学习吧,就是他这个函数用法,我自己是想破头也写不出来的。其实他做的就是让系数从大到小排序,肯定越靠前说明重要程度越大啦,但负值越大应该也影响越大
print('intercept:'+ str(model.intercept_))
sorted(dict(zip(continuous_feature_names, model.coef_)).items(), key=lambda x:x[1], reverse=True)
intercept:-110670.68277237505
[('v_6', 3367064.341641913),
('v_8', 700675.5609398851),
('v_9', 170630.27723221114),
('v_7', 32322.66193201868),
('v_12', 20473.670796932514),
('v_3', 17868.079541480864),
('v_11', 11474.938996683431),
('v_13', 11261.76456000961),
('v_10', 2683.9200905975536),
('gearbox', 881.8225039248213),
('fuelType', 363.90425072160974),
('bodyType', 189.60271012071914),
('city', 44.94975120522923),
('power', 28.55390161674886),
('brand_price_median', 0.5103728134079288),
('brand_price_std', 0.45036347092635),
('brand_amount', 0.14881120395065447),
('brand_price_max', 0.0031910186703120124),
('SaleID', 5.355989919860818e-05),
('train', 9.12696123123169e-08),
('seller', -1.2324308045208454e-06),
('offerType', -1.2362143024802208e-06),
('brand_price_sum', -2.1750068681875495e-05),
('name', -0.0002980012713068734),
('used_time', -0.0025158943328551053),
('brand_price_average', -0.40490484510119196),
('brand_price_min', -2.2467753486892215),
('power_bin', -34.42064411723048),
('v_14', -274.7841180773099),
('kilometer', -372.89752666072053),
('notRepairedDamage', -495.1903844629022),
('v_0', -2045.05495735435),
('v_5', -11022.98624056226),
('v_4', -15121.731109853255),
('v_2', -26098.299920495414),
('v_1', -45556.189297267025)]
from matplotlib import pyplot as plt
取随机样本画图
subsample_index = np.random.randint(low=0, high=len(train_y), size=50)
绘制特征v_9的值与标签的散点图,图片发现模型的预测结果(蓝色点)与真实标签(黑色点)的分布差异较大,且部分预测值出现了小于0的情况,说明我们的模型存在一些问题
一堆点,这能看出啥呀,哎,搞不懂,也可能是点取的太多了吧。但一个明显的错误是能看出来的,就是竟然能预测出小于0的值,这肯定不好吧
plt.scatter(train_X['v_9'][subsample_index], train_y[subsample_index], color='black')
plt.scatter(train_X['v_9'][subsample_index], model.predict(train_X.loc[subsample_index]), color='blue')
plt.xlabel('v_9')
plt.ylabel('price')
plt.legend(['True Price','Predicted Price'],loc='upper right')
print('The predicted price is obvious different from true price')
plt.show()
The predicted price is obvious different from true price
通过作图我们发现数据的标签(price)呈现长尾分布,不利于我们的建模预测。原因是很多模型都假设数据误差项符合正态分布,而长尾分布的数据违背了这一假设。
这个做了分布的图,说是长尾分布不是正态分布,所以会出现预测较大的偏差
import seaborn as sns
print('It is clear to see the price shows a typical exponential distribution')
plt.figure(figsize=(15,5))
plt.subplot(1,2,1)
train_y.hist()
plt.subplot(1,2,2)
train_y[train_y < np.quantile(train_y, 0.9)].hist()
It is clear to see the price shows a typical exponential distribution
在这里我们对标签进行了 l o g ( x + 1 ) log(x+1) log(x+1) 变换,使标签贴近于正态分布
note:用np.log进行对数运算,这一点好神奇,长尾分布取个log就变成正态分布了!不过自己的理解就是那些比较格路的幺蛾子值取个log就像均值靠拢了,所以更接近于正态分布了吧
train_y_ln = np.log(train_y + 1)
import seaborn as sns
print('The transformed price seems like normal distribution')
plt.figure(figsize=(15,5))
plt.subplot(1,2,1)
train_y_ln.hist()
plt.subplot(1,2,2)
train_y_ln[train_y_ln < np.quantile(train_y_ln, 0.9)].hist()
The transformed price seems like normal distribution
dict(zip(continuous_feature_names, model.coef_)).items()可以将键值对都取出来,然后组成一个tuple数组
model = model.fit(train_X, train_y_ln)
print('intercept:'+ str(model.intercept_))
sorted(dict(zip(continuous_feature_names, model.coef_)).items(), key=lambda x:x[1], reverse=True)
intercept:18.750749465562816
[('v_9', 8.052409900567445),
('v_5', 5.764236596653074),
('v_12', 1.6182081236792127),
('v_1', 1.4798310582986653),
('v_11', 1.1669016563599888),
('v_13', 0.9404711296034274),
('v_7', 0.7137273083566703),
('v_3', 0.6837875771084441),
('v_0', 0.008500518010074017),
('power_bin', 0.00849796930289183),
('gearbox', 0.00792237727832901),
('fuelType', 0.006684769706822828),
('bodyType', 0.004523520092702889),
('power', 0.0007161894205358969),
('brand_price_min', 3.334351114748527e-05),
('brand_amount', 2.8978797042779114e-06),
('brand_price_median', 1.2571172873010354e-06),
('brand_price_std', 6.659176363425468e-07),
('brand_price_max', 6.194956307517108e-07),
('brand_price_average', 5.999345965068619e-07),
('SaleID', 2.11941700396494e-08),
('seller', 4.986766555248323e-11),
('train', 1.0800249583553523e-11),
('offerType', -3.7552183584921295e-11),
('brand_price_sum', -1.5126504215930698e-10),
('name', -7.015512588892066e-08),
('used_time', -4.122479372352577e-06),
('city', -0.0022187824810425832),
('v_14', -0.004234223418120774),
('kilometer', -0.013835866226882912),
('notRepairedDamage', -0.27027942349846146),
('v_4', -0.8315701200994835),
('v_2', -0.9470842241619211),
('v_10', -1.6261466689762891),
('v_8', -40.34300748761719),
('v_6', -238.7903638550714)]
再次进行可视化,发现预测结果与真实值较为接近,且未出现异常状况
plt.figure(figsize=(30,10))
plt.scatter(train_X['v_9'][subsample_index], train_y[subsample_index], color='black')
plt.scatter(train_X['v_9'][subsample_index], np.exp(model.predict(train_X.loc[subsample_index])), color='blue')
plt.xlabel('v_9')
plt.ylabel('price')
plt.legend(['True Price','Predicted Price'],loc='upper right')
print('The predicted price seems normal after np.log transforming')
plt.show()
The predicted price seems normal after np.log transforming
五折交叉验证
这里面引用了很多函数需要注意
from sklearn.model_selection import cross_val_score
from sklearn.metrics import mean_absolute_error, make_scorer
note: np.nan_to_num()将nan变成0,将inf变成finit数
def log_transfer(func):
def wrapper(y, yhat):
result = func(np.log(y), np.nan_to_num(np.log(yhat)))
return result
return wrapper
scores = cross_val_score(model, X=train_X, y=train_y, verbose=1, cv = 5, scoring=make_scorer(log_transfer(mean_absolute_error)))
[Parallel(n_jobs=1)]: Using backend SequentialBackend with 1 concurrent workers.
[Parallel(n_jobs=1)]: Done 5 out of 5 | elapsed: 0.7s finished
使用线性回归模型,对未处理标签的特征数据进行五折交叉验证
print('AVG:', np.mean(scores))
AVG: 1.3658023920313753
使用线性回归模型,对处理过标签的特征数据进行五折交叉验证
scores = cross_val_score(model, X=train_X, y=train_y_ln, verbose=1, cv = 5, scoring=make_scorer(mean_absolute_error))
[Parallel(n_jobs=1)]: Using backend SequentialBackend with 1 concurrent workers.
[Parallel(n_jobs=1)]: Done 5 out of 5 | elapsed: 2.1s finished
print('AVG:', np.mean(scores))
AVG: 0.1932530183704746
pd.DataFrame(scores.reshape(1,-1))
0 | 1 | 2 | 3 | 4 | |
---|---|---|---|---|---|
0 | 1.348304 | 1.36349 | 1.380712 | 1.378401 | 1.358105 |
scores = pd.DataFrame(scores.reshape(1,-1))
scores.columns = ['cv' + str(x) for x in range(1, 6)]
scores.index = ['MAE']
scores
cv1 | cv2 | cv3 | cv4 | cv5 | |
---|---|---|---|---|---|
MAE | 1.348304 | 1.36349 | 1.380712 | 1.378401 | 1.358105 |
模拟真实业务情况
但在事实上,由于我们并不具有预知未来的能力,五折交叉验证在某些与时间相关的数据集上反而反映了不真实的情况。通过2018年的二手车价格预测2017年的二手车价格,这显然是不合理的,因此我们还可以采用时间顺序对数据集进行分隔。在本例中,我们选用靠前时间的4/5样本当作训练集,靠后时间的1/5当作验证集,最终结果与五折交叉验证差距不大
import datetime
sample_feature
SaleID | name | model | brand | bodyType | fuelType | gearbox | power | kilometer | notRepairedDamage | ... | used_time | city | brand_amount | brand_price_max | brand_price_median | brand_price_min | brand_price_sum | brand_price_std | brand_price_average | power_bin | |
---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
0 | 0 | 736 | 30.0 | 6 | 1.0 | 0.0 | 0.0 | 60 | 12.5 | 0.0 | ... | 4384.0 | 1.0 | 10192.0 | 35990.0 | 1800.0 | 13.0 | 36457520.0 | 4564.0 | 3576.0 | 5.0 |
1 | 2 | 14874 | 115.0 | 15 | 1.0 | 0.0 | 0.0 | 163 | 12.5 | 0.0 | ... | 4384.0 | 2.0 | 1458.0 | 45000.0 | 8496.0 | 100.0 | 14373814.0 | 5424.0 | 9848.0 | 16.0 |
2 | 4 | 111080 | 110.0 | 5 | 1.0 | 0.0 | 0.0 | 68 | 5.0 | 0.0 | ... | 1531.0 | 6.0 | 4664.0 | 31500.0 | 2300.0 | 20.0 | 15414322.0 | 3344.0 | 3306.0 | 6.0 |
3 | 5 | 137642 | 24.0 | 10 | 0.0 | 1.0 | 0.0 | 109 | 10.0 | 0.0 | ... | 2482.0 | 3.0 | 13992.0 | 92900.0 | 5200.0 | 15.0 | 113034208.0 | 8248.0 | 8076.0 | 10.0 |
4 | 6 | 2402 | 13.0 | 4 | 0.0 | 0.0 | 1.0 | 150 | 15.0 | 0.0 | ... | 6184.0 | 3.0 | 16576.0 | 99999.0 | 6000.0 | 12.0 | 138279072.0 | 8088.0 | 8344.0 | 14.0 |
... | ... | ... | ... | ... | ... | ... | ... | ... | ... | ... | ... | ... | ... | ... | ... | ... | ... | ... | ... | ... | ... |
91724 | 149994 | 43073 | 42.0 | 1 | 1.0 | 0.0 | 0.0 | 122 | 3.0 | 0.0 | ... | 1538.0 | 5.0 | 13656.0 | 84000.0 | 6400.0 | 15.0 | 124044600.0 | 8992.0 | 9080.0 | 12.0 |
91725 | 149995 | 163978 | 121.0 | 10 | 4.0 | 0.0 | 1.0 | 163 | 15.0 | 0.0 | ... | 5772.0 | 4.0 | 13992.0 | 92900.0 | 5200.0 | 15.0 | 113034208.0 | 8248.0 | 8076.0 | 16.0 |
91726 | 149996 | 184535 | 116.0 | 11 | 0.0 | 0.0 | 0.0 | 125 | 10.0 | 0.0 | ... | 2322.0 | 2.0 | 2944.0 | 34500.0 | 2900.0 | 30.0 | 13398006.0 | 4724.0 | 4548.0 | 12.0 |
91727 | 149997 | 147587 | 60.0 | 11 | 1.0 | 1.0 | 0.0 | 90 | 6.0 | 0.0 | ... | 2003.0 | 3.0 | 2944.0 | 34500.0 | 2900.0 | 30.0 | 13398006.0 | 4724.0 | 4548.0 | 8.0 |
91728 | 149998 | 45907 | 34.0 | 10 | 3.0 | 1.0 | 0.0 | 156 | 15.0 | 0.0 | ... | 3672.0 | 1.0 | 13992.0 | 92900.0 | 5200.0 | 15.0 | 113034208.0 | 8248.0 | 8076.0 | 15.0 |
91729 rows × 39 columns
重设索引,并将原索引丢掉
sample_feature = sample_feature.reset_index(drop=True)
split_point = len(sample_feature) // 5 * 4
train = sample_feature.loc[:split_point].dropna()
val = sample_feature.loc[split_point:].dropna()
train_X = train[continuous_feature_names]
train_y_ln = np.log(train['price'] + 1)
val_X = val[continuous_feature_names]
val_y_ln = np.log(val['price'] + 1)
model = model.fit(train_X, train_y_ln)
mean_absolute_error(val_y_ln, model.predict(val_X))
0.19577667270300989
绘制学习率曲线与验证曲线
from sklearn.model_selection import learning_curve, validation_curve
def plot_learning_curve(estimator, title, X, y, ylim=None, cv=None,n_jobs=1, train_size=np.linspace(.1, 1.0, 5 )):
plt.figure()
plt.title(title)
if ylim is not None:
plt.ylim(*ylim)
plt.xlabel('Training example')
plt.ylabel('score')
train_sizes, train_scores, test_scores = learning_curve(estimator, X, y, cv=cv, n_jobs=n_jobs, train_sizes=train_size, scoring = make_scorer(mean_absolute_error))
train_scores_mean = np.mean(train_scores, axis=1)
train_scores_std = np.std(train_scores, axis=1)
test_scores_mean = np.mean(test_scores, axis=1)
test_scores_std = np.std(test_scores, axis=1)
plt.grid()#区域
plt.fill_between(train_sizes, train_scores_mean - train_scores_std,
train_scores_mean + train_scores_std, alpha=0.1,
color="r")
plt.fill_between(train_sizes, test_scores_mean - test_scores_std,
test_scores_mean + test_scores_std, alpha=0.1,
color="g")
plt.plot(train_sizes, train_scores_mean, 'o-', color='r',
label="Training score")
plt.plot(train_sizes, test_scores_mean,'o-',color="g",
label="Cross-validation score")
plt.legend(loc="best")
return plt
plot_learning_curve(LinearRegression(), 'Liner_model', train_X[:1000], train_y_ln[:1000], ylim=(0.0, 0.5), cv=5, n_jobs=1)
多种模型对比
train = sample_feature[continuous_feature_names + ['price']].dropna()
train_X = train[continuous_feature_names]
train_y = train['price']
train_y_ln = np.log(train_y + 1)
线性模型 & 嵌入式特征选择
from sklearn.linear_model import LinearRegression
from sklearn.linear_model import Ridge
from sklearn.linear_model import Lasso
models = [LinearRegression(),
Ridge(),
Lasso()]
result = dict()
for model in models:
model_name = str(model).split('(')[0]
scores = cross_val_score(model, X=train_X, y=train_y_ln, verbose=0, cv = 5, scoring=make_scorer(mean_absolute_error))
result[model_name] = scores
print(model_name + ' is finished')
LinearRegression is finished
Ridge is finished
Lasso is finished
result = pd.DataFrame(result)
result.index = ['cv' + str(x) for x in range(1, 6)]
result
LinearRegression | Ridge | Lasso | |
---|---|---|---|
cv1 | 0.190792 | 0.194832 | 0.383899 |
cv2 | 0.193758 | 0.197632 | 0.381893 |
cv3 | 0.194132 | 0.198123 | 0.384090 |
cv4 | 0.191825 | 0.195670 | 0.380526 |
cv5 | 0.195758 | 0.199676 | 0.383611 |
model = LinearRegression().fit(train_X, train_y_ln)
print('intercept:'+ str(model.intercept_))
sns.barplot(abs(model.coef_), continuous_feature_names)
intercept:18.750749465575524
model = Ridge().fit(train_X, train_y_ln)
print('intercept:'+ str(model.intercept_))
sns.barplot(abs(model.coef_), continuous_feature_names)
intercept:4.671709786976683
L1正则化有助于生成一个稀疏权值矩阵,进而可以用于特征选择。如下图,我们发现power与userd_time特征非常重要。
model = Lasso().fit(train_X, train_y_ln)
print('intercept:'+ str(model.intercept_))
sns.barplot(abs(model.coef_), continuous_feature_names)
intercept:8.6721824626662
除此之外,决策树通过信息熵或GINI指数选择分裂节点时,优先选择的分裂特征也更加重要,这同样是一种特征选择的方法。XGBoost与LightGBM模型中的model_importance指标正是基于此计算的
非线性模型
除了线性模型以外,还有许多我们常用的非线性模型如下,在此篇幅有限不再一一讲解原理。我们选择了部分常用模型与线性模型进行效果比对。
下面用到的方法有:线性回归、支持向量机、决策树回归、随机森林、梯度提升回归?、多层感知器、xgboost、lightgbm
from sklearn.linear_model import LinearRegression
from sklearn.svm import SVC
from sklearn.tree import DecisionTreeRegressor
from sklearn.ensemble import RandomForestRegressor
from sklearn.ensemble import GradientBoostingRegressor
from sklearn.neural_network import MLPRegressor
from xgboost.sklearn import XGBRegressor
from lightgbm.sklearn import LGBMRegressor
models = [LinearRegression(),
DecisionTreeRegressor(),
RandomForestRegressor(),
GradientBoostingRegressor(),
MLPRegressor(solver='lbfgs', max_iter=100),
XGBRegressor(n_estimators = 100, objective='reg:squarederror'),
LGBMRegressor(n_estimators = 100)]
models[0]
[LinearRegression(copy_X=True, fit_intercept=True, n_jobs=None, normalize=False),
DecisionTreeRegressor(ccp_alpha=0.0, criterion='mse', max_depth=None,
max_features=None, 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, presort='deprecated',
random_state=None, splitter='best'),
RandomForestRegressor(bootstrap=True, ccp_alpha=0.0, criterion='mse',
max_depth=None, max_features='auto', max_leaf_nodes=None,
max_samples=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=100, n_jobs=None, oob_score=False,
random_state=None, verbose=0, warm_start=False),
GradientBoostingRegressor(alpha=0.9, ccp_alpha=0.0, criterion='friedman_mse',
init=None, learning_rate=0.1, loss='ls', max_depth=3,
max_features=None, 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=100,
n_iter_no_change=None, presort='deprecated',
random_state=None, subsample=1.0, tol=0.0001,
validation_fraction=0.1, verbose=0, warm_start=False),
MLPRegressor(activation='relu', alpha=0.0001, batch_size='auto', beta_1=0.9,
beta_2=0.999, early_stopping=False, epsilon=1e-08,
hidden_layer_sizes=(100,), learning_rate='constant',
learning_rate_init=0.001, max_fun=15000, max_iter=100,
momentum=0.9, n_iter_no_change=10, nesterovs_momentum=True,
power_t=0.5, random_state=None, shuffle=True, solver='lbfgs',
tol=0.0001, validation_fraction=0.1, verbose=False,
warm_start=False),
XGBRegressor(base_score=None, booster=None, colsample_bylevel=None,
colsample_bynode=None, colsample_bytree=None, gamma=None,
gpu_id=None, importance_type='gain', interaction_constraints=None,
learning_rate=None, max_delta_step=None, max_depth=None,
min_child_weight=None, missing=nan, monotone_constraints=None,
n_estimators=100, n_jobs=None, num_parallel_tree=None,
objective='reg:squarederror', random_state=None, reg_alpha=None,
reg_lambda=None, scale_pos_weight=None, subsample=None,
tree_method=None, validate_parameters=False, verbosity=None),
LGBMRegressor(boosting_type='gbdt', class_weight=None, colsample_bytree=1.0,
importance_type='split', learning_rate=0.1, max_depth=-1,
min_child_samples=20, min_child_weight=0.001, min_split_gain=0.0,
n_estimators=100, n_jobs=-1, num_leaves=31, objective=None,
random_state=None, reg_alpha=0.0, reg_lambda=0.0, silent=True,
subsample=1.0, subsample_for_bin=200000, subsample_freq=0)]
result = dict()
for model in models:
model_name = str(model).split('(')[0]
scores = cross_val_score(model, X=train_X, y=train_y_ln, verbose=0, cv = 5, scoring=make_scorer(mean_absolute_error))
result[model_name] = scores
print(model_name + ' is finished')
LinearRegression is finished
DecisionTreeRegressor is finished
RandomForestRegressor is finished
GradientBoostingRegressor is finished
MLPRegressor is finished
XGBRegressor is finished
LGBMRegressor is finished
result = pd.DataFrame(result)
result.index = ['cv' + str(x) for x in range(1, 6)]
result
LinearRegression | DecisionTreeRegressor | RandomForestRegressor | GradientBoostingRegressor | MLPRegressor | XGBRegressor | LGBMRegressor | |
---|---|---|---|---|---|---|---|
cv1 | 0.190792 | 0.196785 | 0.132981 | 0.168903 | 336.480763 | 0.142378 | 0.141544 |
cv2 | 0.193758 | 0.193241 | 0.134480 | 0.171857 | 399.941663 | 0.140922 | 0.145501 |
cv3 | 0.194132 | 0.189123 | 0.133667 | 0.170915 | 266.022859 | 0.139393 | 0.143887 |
cv4 | 0.191825 | 0.190084 | 0.132413 | 0.169083 | 353.765162 | 0.137492 | 0.142497 |
cv5 | 0.195758 | 0.204320 | 0.137153 | 0.174078 | 720.029717 | 0.143733 | 0.144852 |
可以看到随机森林模型在每一个fold中均取得了更好的效果
模型调参
在此我们介绍了三种常用的调参方法如下:
- 贪心算法 https://www.jianshu.com/p/ab89df9759c8
- 网格调参 https://blog.csdn.net/weixin_43172660/article/details/83032029
- 贝叶斯调参 https://blog.csdn.net/linxid/article/details/81189154
## LGB的参数集合:
objective = ['regression', 'regression_l1', 'mape', 'huber', 'fair']
num_leaves = [3,5,10,15,20,40, 55]# 叶子节点个数
max_depth = [3,5,10,15,20,40, 55]# 树最大深度
bagging_fraction = []
feature_fraction = []
drop_rate = []
1 贪心调参
best_obj = dict()
for obj in objective:
model = LGBMRegressor(objective=obj)
score = np.mean(cross_val_score(model, X=train_X, y=train_y_ln, verbose=0, cv = 5, scoring=make_scorer(mean_absolute_error)))
best_obj[obj] = score
best_leaves = dict()
for leaves in num_leaves:
model = LGBMRegressor(objective=min(best_obj.items(), key=lambda x:x[1])[0], num_leaves=leaves)
score = np.mean(cross_val_score(model, X=train_X, y=train_y_ln, verbose=0, cv = 5, scoring=make_scorer(mean_absolute_error)))
best_leaves[leaves] = score
best_depth = dict()
for depth in max_depth:
model = LGBMRegressor(objective=min(best_obj.items(), key=lambda x:x[1])[0],
num_leaves=min(best_leaves.items(), key=lambda x:x[1])[0],
max_depth=depth)
score = np.mean(cross_val_score(model, X=train_X, y=train_y_ln, verbose=0, cv = 5, scoring=make_scorer(mean_absolute_error)))
best_depth[depth] = score
2 Grid Search 调参
from sklearn.model_selection import GridSearchCV
parameters = {'objective': objective , 'num_leaves': num_leaves, 'max_depth': max_depth}
model = LGBMRegressor()
clf = GridSearchCV(model, parameters, cv=5)
clf = clf.fit(train_X, train_y)
clf.best_params_
{'max_depth': 15, 'num_leaves': 55, 'objective': 'regression'}
model = LGBMRegressor(objective='regression',
num_leaves=55,
max_depth=15)
np.mean(cross_val_score(model, X=train_X, y=train_y_ln, verbose=0, cv = 5, scoring=make_scorer(mean_absolute_error)))
0.13754833106731224
3 贝叶斯调参
from bayes_opt import BayesianOptimization
首先定义一个函数,返回的应该是k折验证的准确率
def rf_cv(num_leaves, max_depth, subsample, min_child_samples):
val = cross_val_score(
LGBMRegressor(objective = 'regression_l1',
num_leaves=int(num_leaves),
max_depth=int(max_depth),
subsample = subsample,
min_child_samples = int(min_child_samples)
),
X=train_X, y=train_y_ln, verbose=0, cv = 5, scoring=make_scorer(mean_absolute_error)
).mean()
return 1 - val
rf_bo = BayesianOptimization(
rf_cv,
{
'num_leaves': (2, 100),
'max_depth': (2, 100),
'subsample': (0.1, 1),
'min_child_samples' : (2, 100)
}
)
rf_bo.maximize()
[31mInitialization[0m
[94m----------------------------------------------------------------------------------------------[0m
Step | Time | Value | max_depth | min_child_samples | num_leaves | subsample |
1 | 00m18s | [35m 0.86150[0m | [32m 41.7364[0m | [32m 9.9106[0m | [32m 43.8343[0m | [32m 0.4737[0m |
2 | 00m25s | [35m 0.86523[0m | [32m 20.0225[0m | [32m 11.1218[0m | [32m 62.7188[0m | [32m 0.3173[0m |
3 | 00m17s | 0.86143 | 45.6899 | 64.7975 | 43.3648 | 0.8202 |
4 | 00m10s | 0.83443 | 35.0734 | 94.4035 | 8.8507 | 0.8622 |
5 | 00m18s | 0.86244 | 85.3179 | 23.6400 | 46.9726 | 0.7228 |
[31mBayesian Optimization[0m
[94m----------------------------------------------------------------------------------------------[0m
Step | Time | Value | max_depth | min_child_samples | num_leaves | subsample |
6 | 00m36s | [35m 0.86920[0m | [32m 99.4589[0m | [32m 99.1108[0m | [32m 99.7631[0m | [32m 0.8204[0m |
7 | 00m24s | 0.80646 | 2.5444 | 96.8373 | 99.3405 | 0.8690 |
8 | 00m35s | 0.86895 | 92.8784 | 2.6496 | 99.7885 | 0.2488 |
9 | 00m36s | 0.86895 | 42.1611 | 2.5265 | 99.9897 | 0.1828 |
10 | 00m37s | [35m 0.86921[0m | [32m 99.1173[0m | [32m 57.5858[0m | [32m 99.6229[0m | [32m 0.1268[0m |
11 | 00m31s | 0.85987 | 99.4359 | 99.6975 | 37.0181 | 0.3138 |
12 | 00m41s | 0.86780 | 66.8997 | 21.5978 | 81.4402 | 0.9551 |
13 | 00m41s | 0.86692 | 99.9751 | 75.6942 | 72.3071 | 0.8105 |
14 | 00m29s | 0.82570 | 3.3240 | 2.7883 | 98.3646 | 0.8390 |
15 | 00m18s | 0.77190 | 99.7659 | 5.9422 | 2.2912 | 0.4362 |
16 | 00m29s | 0.86532 | 57.5752 | 99.7148 | 61.8887 | 0.1556 |
17 | 00m20s | 0.82546 | 3.3740 | 99.9999 | 42.6172 | 0.1799 |
18 | 00m33s | 0.86700 | 52.7029 | 2.9349 | 73.3034 | 0.1083 |
19 | 00m20s | 0.80204 | 3.1007 | 3.6067 | 3.0243 | 0.5249 |
20 | 00m39s | [35m 0.86921[0m | [32m 64.1085[0m | [32m 98.4171[0m | [32m 99.6344[0m | [32m 0.1344[0m |
21 | 00m41s | [35m 0.86927[0m | [32m 43.9552[0m | [32m 47.2108[0m | [32m 99.2895[0m | [32m 0.1297[0m |
22 | 00m38s | 0.86555 | 74.0348 | 52.8610 | 61.7144 | 0.1191 |
23 | 00m42s | 0.86703 | 99.5047 | 28.7613 | 74.8431 | 0.1029 |
24 | 00m41s | 0.86601 | 36.6047 | 34.0949 | 64.7690 | 0.1102 |
25 | 00m55s | 0.86916 | 66.7270 | 65.1533 | 99.8953 | 0.9703 |
26 | 00m44s | [35m 0.86940[0m | [32m 72.6885[0m | [32m 25.6077[0m | [32m 99.7277[0m | [32m 0.1478[0m |
27 | 00m47s | 0.86909 | 99.9445 | 20.9597 | 98.0581 | 0.9664 |
28 | 00m45s | 0.86758 | 83.0911 | 99.8907 | 79.5884 | 0.6953 |
29 | 00m48s | 0.86923 | 49.6717 | 17.7407 | 99.3922 | 0.9610 |
30 | 00m48s | 0.86895 | 65.9624 | 2.8017 | 99.2133 | 0.6879 |
rf_bo.res['max']
{'max_val': 0.8693980187983674,
'max_params': {'num_leaves': 99.72769486848688,
'max_depth': 72.68851665306862,
'subsample': 0.14775375691262158,
'min_child_samples': 25.60766609629669}}
1-rf_bo.res['max']['max_val']
0.13060198120163258
总结
在本章中,我们完成了建模与调参的工作,并对我们的模型进行了验证。此外,我们还采用了一些基本方法来提高预测的精度,提升如下图所示。
plt.figure(figsize=(13,5))
sns.lineplot(x=['0_origin','1_log_transfer','2_L1_&_L2','3_change_model','4_parameter_turning'], y=[1.36 ,0.19, 0.19, 0.14, 0.13])