整体流程图
3.2.4 案例:预测facebook签到位置
流程分析:
1)获取数据
2)数据处理
目的:
特征值 x
目标值 y
a.缩小数据范围
2 < x < 2.5
1.0 < y < 1.5
b.time -> 年月日时分秒
c.过滤签到次数少的地点
数据集划分
3)特征工程:标准化
4)KNN算法预估流程
5)模型选择与调优
6)模型评估
1.获取数据
import pandas as pd
import numpy as np
#1.获取数据
data = pd.read_csv("../input/train.csv")
print(data.head())
print(data.shape)
row_id x y accuracy time place_id
0 0 0.7941 9.0809 54 470702 8523065625
1 1 5.9567 4.7968 13 186555 1757726713
2 2 8.3078 7.0407 74 322648 1137537235
3 3 7.3665 2.5165 65 704587 6567393236
4 4 4.0961 1.1307 31 472130 7440663949
(29118021, 6)
2.基本的数据处理
# 1) 缩小数据范围
data = data.query("x<2.5 & x>2 & y<1.5 & y> 1.0")
# 2)时间戳转换为时间
time_value = pd.to_datetime(data['time'],unit='s')
time_value.head()
112 1970-01-08 05:06:14
180 1970-01-08 01:29:55
367 1970-01-07 17:01:07
874 1970-01-02 15:52:46
1022 1970-01-03 09:46:33
Name: time, dtype: datetime64[ns]
# 3) 将时间从Series的格式转换到DatetimeIndex的格式
time = pd.DatetimeIndex(time_value)
time.weekday # 星期几
time.month # 几月
time.year # 年份
time.hour # 小时
Int64Index([ 5, 1, 17, 15, 9, 19, 13, 22, 14, 16,
...
9, 10, 11, 3, 22, 8, 12, 20, 18, 22],
dtype='int64', name='time', length=83197)
data["day"] = time.day
data["weekday"] = time.weekday
data["hour"] = time.weekday
data.head()
row_id | x | y | accuracy | time | place_id | day | weekday | hour | |
---|---|---|---|---|---|---|---|---|---|
112 | 112 | 2.2360 | 1.3655 | 66 | 623174 | 7663031065 | 8 | 3 | 3 |
180 | 180 | 2.2003 | 1.2541 | 65 | 610195 | 2358558474 | 8 | 3 | 3 |
367 | 367 | 2.4108 | 1.3213 | 74 | 579667 | 6644108708 | 7 | 2 | 2 |
874 | 874 | 2.0822 | 1.1973 | 320 | 143566 | 3229876087 | 2 | 4 | 4 |
1022 | 1022 | 2.0160 | 1.1659 | 65 | 207993 | 3244363975 | 3 | 5 | 5 |
# 4) 过滤掉签到次数少的地点 分组+聚合函数
# 因为都是重复的没必要全都要,我们拿一个“row_id”
place_count = data.groupby("place_id").count()["row_id"]
# 挑选正确的地点,签到次数小于3次以下的都去除掉
place_count[place_count > 3].head()
place_id
1014605271 28
1015645743 4
1017236154 31
1024951487 5
1028119817 4
Name: row_id, dtype: int64
# 在data中通过地点过滤数据
boo_ = data["place_id"].isin(place_count[place_count > 3].index.values) # !!! 后面加index和values
data_final = data[boo_]
# 筛选特征值和目标值
x = data_final[['x','y','accuracy','day','weekday','hour']]
y = data_final['place_id']
x.head()
x | y | accuracy | day | weekday | hour | |
---|---|---|---|---|---|---|
112 | 2.2360 | 1.3655 | 66 | 8 | 3 | 3 |
367 | 2.4108 | 1.3213 | 74 | 7 | 2 | 2 |
874 | 2.0822 | 1.1973 | 320 | 2 | 4 | 4 |
1022 | 2.0160 | 1.1659 | 65 | 3 | 5 | 5 |
1045 | 2.3859 | 1.1660 | 498 | 6 | 1 | 1 |
y.head()
112 7663031065
367 6644108708
874 3229876087
1022 3244363975
1045 6438240873
Name: place_id, dtype: int64
# 数据集的划分
from sklearn.model_selection import train_test_split
x_train,x_test,y_train,y_test = train_test_split(x,y)
3.特征工程 标准化
from sklearn.preprocessing import StandardScaler
from sklearn.neighbors import KNeighborsClassifier
from sklearn.model_selection import GridSearchCV
# 1) 特征工程标准化
transfer = StandardScaler()
x_train = transfer.fit_transform(x_train)
x_test = transfer.transform(x_test)
4.KNN算法预估流程
# 1) KNN算法预估器
estimator = KNeighborsClassifier()
# 2)加入网格搜索与交叉验证
# 参数准备
param_dict = {"n_neighbors":[1,3,5,7,9,11,13]}
estimator = GridSearchCV(estimator,param_grid=param_dict,cv=3)
estimator.fit(x_train,y_train)
5.模型评估
# 计算准确率
score = estimator.score(x_test,y_test)
print("测试集的准确率为:",score)
# 最佳参数:best_params
print("最佳参数为:",estimator.best_params_)
print("最佳验证集结果为:",estimator.best_score_)
print("最佳估计器为:",estimator.best_estimator_)
准确率为: 0.43726517698240064
最佳参数为: {'n_neighbors': 9}
最佳结果为: 0.42577700141722424
最佳估计器为: KNeighborsClassifier(algorithm='auto', leaf_size=30, metric='minkowski',
metric_params=None, n_jobs=None, n_neighbors=9, p=2,
weights='uniform')
7.使用全部数据来,训练模型,然后来预测test的数据
未完待续。。。