python离群点检测_GitHub - sladesha/Frcwp: 滴滴黑产识别的离群点检测python自用包

What is Frcwp

It means fast risk control with python.

It's a lightweight tool that automatic recognize the outliers from a large data pool.

This project aims to help people get easily method with abnormal recognition, especially forces password attacks.

We wish it could be a nice Open Source which could simplify the complexity of the Data Feature Project.

Theory

Compared with common methods

We got the correctness around 29 data sets below,however the speed of Frcwp comes last.

Usage

U can get it easily download from Pypi with pip install Frcwp.

import pandas as pd

from Frcwp import Frcwp

path = '../data/data_all.csv'

traindata = pd.read_csv(path)

frc = Frcwp()

traindata = frc.changeformat(traindata, index=0)

params = {

'na_rate': 0.4,

'single_dealed': 1,

'is_scale': 0,

'distince_method': 'Maha',

'outlier_rate': 0.05,

'strange_rate': 0.15,

'nestimators': 150,

'contamination': 0.2

}

frc.fit(traindata, **params)

predict_params = {

'output': 20,

'is_whole': 1

}

frc.predict(frc.potentialdata_set, **predict_params)

Dependence

Frcwp is implemented in Python 3.6, use Pandas.DataFrame to store data. These package can be easily installed using pip.

Reference

TODO

feature scanning

increase new outliers distinguishing methods