otto案例介绍 -- Otto Group Product Classification Challenge【xgboost实现】

1. 背景介绍

奥托集团是世界上最⼤的电⼦商务公司之⼀，在20多个国家设有⼦公司。该公司每天都在世界各地销售数百万种产品, 所以对其产品根据性能合理的分类⾮常重要。

不过,在实际⼯作中,⼯作⼈员发现,许多相同的产品得到了不同的分类。本案例要求,你对奥拓集团的产品进⾏正确的分分 类。尽可能的提供分类的准确性。

链接：https://www.kaggle.com/c/otto-group-product-classification-challenge/overview

2. 思路分析

• 1.数据获取
• 2.数据基本处理
  • 2.1 截取部分数据
  • 2.2 把标签纸转换为数字
  • 2.3 分割数据(使⽤StratifiedShuffleSplit)
  • 2.4 数据标准化
  • 2.5 数据pca降维
• 3.模型训练
  • 3.1 基本模型训练
  • 3.2 模型调优
    • 3.2.1 调优参数:
      • n_estimator,
      • max_depth,
      • min_child_weights,
      • subsamples,
      • consample_bytrees,
      • etas
    • 3.2.2 确定最后最优参数

3. 代码实现

import numpy as np
import pandas as pd
import matplotlib.pyplot as plt

3.1 数据获取

data = pd.read_csv("./data/otto/train.csv")

data.shape
(61878, 95)

data.describe()

# 图形可视化,查看数据分布
import seaborn as sns

sns.countplot(data.target)

plt.show()

由上图可以看出,该数据类别不均衡,所以需要后期处理

3.2 数据基本处理

数据已经经过脱敏,不再需要特殊处理

3.2.1 截取部分数据

new1_data = data[:10000]
new1_data.shape

# 图形可视化,查看数据分布
import seaborn as sns

sns.countplot(new1_data.target)

plt.show()

使用上面方式获取数据不可行,然后使用随机欠采样获取响应的数据

# 随机欠采样获取数据
# 首先需要确定特征值\标签值

y = data["target"]
x = data.drop(["id", "target"], axis=1)

# 欠采样获取数据
from imblearn.under_sampling import RandomUnderSampler

rus = RandomUnderSampler(random_state=0)

X_resampled, y_resampled = rus.fit_resample(x, y)

x.shape, y.shape
X_resampled.shape, y_resampled.shape

# 图形可视化,查看数据分布
import seaborn as sns

sns.countplot(y_resampled)

plt.show()

3.2.2 把标签值转换为数字

y_resampled.head()

from sklearn.preprocessing import LabelEncoder

le = LabelEncoder()
y_resampled = le.fit_transform(y_resampled)

3.2.3 分割数据

from sklearn.model_selection import train_test_split

x_train, x_test, y_train, y_test = train_test_split(X_resampled, y_resampled, test_size=0.2)

x_train.shape, y_train.shape

# 图形可视化
import seaborn as sns

sns.countplot(y_test)
plt.show()

# 通过StratifiedShuffleSplit实现数据分割

from sklearn.model_selection import StratifiedShuffleSplit

sss = StratifiedShuffleSplit(n_splits=1, test_size=0.2, random_state=0)

for train_index, test_index in sss.split(X_resampled.values, y_resampled):
    print(len(train_index))
    print(len(test_index))
    
    x_train = X_resampled.values[train_index]
    x_val = X_resampled.values[test_index]
    
    y_train = y_resampled[train_index]
    y_val = y_resampled[test_index]

13888
3473

print(x_train.shape, x_val.shape)

(13888, 93) (3473, 93)

# 图形可视化
import seaborn as sns

sns.countplot(y_val)
plt.show()

3.2.4 数据标准化

from sklearn.preprocessing import StandardScaler

scaler = StandardScaler()
scaler.fit(x_train)

x_train_scaled = scaler.transform(x_train)
x_val_scaled = scaler.transform(x_val)

3.2.5 数据PCA降维

x_train_scaled.shape
(13888, 93)

from sklearn.decomposition import PCA

pca = PCA(n_components=0.9)

x_train_pca = pca.fit_transform(x_train_scaled)
x_val_pca = pca.transform(x_val_scaled)

print(x_train_pca.shape, x_val_pca.shape)

# 可视化数据降维信息变化程度
plt.plot(np.cumsum(pca.explained_variance_ratio_))

plt.xlabel("元素数量")
plt.ylabel("表达信息百分占比")

plt.show()

3.3 模型训练

3.3.1 基本模型训练

from xgboost import XGBClassifier

xgb = XGBClassifier()
xgb.fit(x_train_pca, y_train)

# 输出预测值,一定输出带有百分占比的预测值
y_pre_proba = xgb.predict_proba(x_val_pca)

y_pre_proba

# logloss评估
from sklearn.metrics import log_loss

log_loss(y_val, y_pre_proba, eps=1e-15, normalize=True)

0.735851001353164

xgb.get_params

3.3.2 模型调优

3.3.2.1 确定最优的estimators

scores_ne = []
n_estimators = [100, 200, 300, 400, 500, 550, 600, 700]

for nes in n_estimators:
    print("n_estimators:", nes)
    xgb = XGBClassifier(max_depth=3,
                        learning_rate=0.1, 
                        n_estimators=nes, 
                        objective="multi:softprob", 
                        n_jobs=-1, 
                        nthread=4, 
                        min_child_weight=1,
                        subsample=1,
                        colsample_bytree=1,
                        seed=42)
    
    xgb.fit(x_train_pca, y_train)
    y_pre = xgb.predict_proba(x_val_pca)
    score = log_loss(y_val, y_pre)
    scores_ne.append(score)
    
    print("每次测试的logloss值是:{}".format(score))

# 图形化展示相应的logloss值
plt.plot(n_estimators, scores_ne, "o-")

plt.xlabel("n_estimators")
plt.ylabel("log_loss")
plt.show()

print("最优的n_estimators值是:{}".format(n_estimators[np.argmin(scores_ne)]))

3.3.2.2 确定最优的max_depth

scores_md = []
max_depths = [1,3,5,6,7]

for md in max_depths:
    print("max_depth:", md)
    xgb = XGBClassifier(max_depth=md,
                        learning_rate=0.1, 
                        n_estimators=n_estimators[np.argmin(scores_ne)], 
                        objective="multi:softprob", 
                        n_jobs=-1, 
                        nthread=4, 
                        min_child_weight=1,
                        subsample=1,
                        colsample_bytree=1,
                        seed=42)
    
    xgb.fit(x_train_pca, y_train)
    y_pre = xgb.predict_proba(x_val_pca)
    score = log_loss(y_val, y_pre)
    scores_md.append(score)
    
    print("每次测试的logloss值是:{}".format(score))

# 图形化展示相应的logloss值
plt.plot(max_depths, scores_md, "o-")

plt.xlabel("max_depths")
plt.ylabel("log_loss")
plt.show()

print("最优的max_depths值是:{}".format(max_depths[np.argmin(scores_md)]))

3.3.2.3 依据上面模式,运行调试下面参数

min_child_weights,

subsamples,

consample_bytrees,

etas

3.3.3 最优模型

xgb = XGBClassifier(learning_rate =0.1, 
                    n_estimators=550, 
                    max_depth=3, 
                    min_child_weight=3, 
                    subsample=0.7, 
                    colsample_bytree=0.7, 
                    nthread=4, 
                    seed=42, 
                    objective='multi:softprob')

xgb.fit(x_train_scaled, y_train)

y_pre = xgb.predict_proba(x_val_scaled)

print("测试数据的log_loss值为 : {}".format(log_loss(y_val, y_pre, eps=1e-15, normalize=True)))

加油!

感谢!

努力!