PrefixSpan/HwanWooCode/prefix_1025.py

#!/usr/bin/env python
# coding: utf-8

# In[1]:


import pandas as pd
import numpy as np
from mlxtend.preprocessing import TransactionEncoder
from mlxtend.frequent_patterns import association_rules, fpgrowth
from prefixspan import PrefixSpan



df = pd.read_csv("ts_data_accident-2020_sample.csv", low_memory=False, encoding='ISO-8859-1')
pd.set_option('display.max_columns',None)
df=df[['RISK_V2','INST_NM','DRULE_ATT_TYPE_CODE1','TW_ATT_IP','TW_ATT_PORT','TW_DMG_IP','TW_DMG_PORT','ACCD_DMG_PROTO_NM','TW_ATT_CT_NM','ACCD_FIND_MTD_CODE','DRULE_NM']].dropna()
len(df)
##################### NTM section #####################
NTM_df=df[df['ACCD_FIND_MTD_CODE']==1]
NTM_df

# Pick out it in order to get the asset, risk, intent, black IP out
RISK_V2=NTM_df['RISK_V2']
RISK_V2_FILTERED=RISK_V2.dropna()
## 결측값 제거.

import json
from pandas import json_normalize

# modified
def get_asset_desc(asset_field):
    if asset_field == 'ASSETS_VAL_1':
        return '공인-전체IP대역(유선)'
    elif asset_field == 'ASSETS_VAL_2':
        return '공인-전체IP대역(무선)'
    elif asset_field == 'ASSETS_VAL_3':
        return '공인-WEB서버'
    elif asset_field == 'ASSETS_VAL_4':
        return '공인-내부응용서버'
    elif asset_field == 'ASSETS_VAL_5':
        return '공인-DB서버'
    elif asset_field == 'ASSETS_VAL_6':
        return '공인-패치서버'
    elif asset_field == 'ASSETS_VAL_7':
        return '공인-네트워크'
    elif asset_field == 'ASSETS_VAL_8':
        return '공인-보안'
    elif asset_field == 'ASSETS_VAL_9':
        return '공인-업무용PC'
    elif asset_field == 'ASSETS_VAL_10':
        return '공인-비업무용PC'
    elif asset_field == 'ASSETS_VAL_11':
        return '공인-기타'
    elif asset_field == 'ASSETS_VAL_12':
        return '사설-전체IP대역(유선)'
    elif asset_field == 'ASSETS_VAL_13':
        return '사설-전체IP대역(무선)'
    elif asset_field == 'ASSETS_VAL_14':
        return '사설-WEB서버'
    elif asset_field == 'ASSETS_VAL_15':
        return '사설-내부응용서버'
    elif asset_field == 'ASSETS_VAL_16':
        return '사설-DB서버'
    elif asset_field == 'ASSETS_VAL_17':
        return '사설-패치서버'
    elif asset_field == 'ASSETS_VAL_18':
        return '사설-네트워크'
    elif asset_field == 'ASSETS_VAL_19':
        return '사설-보안'
    elif asset_field == 'ASSETS_VAL_20':
        return '사설-업무용PC'
    elif asset_field == 'ASSETS_VAL_21':
        return '사설-비업무용PC'
    elif asset_field == 'ASSETS_VAL_22':
        return '사설-기타'
    else:
        return ''

def get_intent_desc(intent_field):
    if intent_field == 'INTENT_VAL_1':
        return '파괴'
    elif intent_field == 'INTENT_VAL_2':
        return '유출'
    elif intent_field == 'INTENT_VAL_3':
        return '지연'
    elif intent_field == 'INTENT_VAL_4':
        return '잠복'
    elif intent_field == 'INTENT_VAL_5':
        return '단순침입'
    elif intent_field == 'INTENT_VAL_6':
        return 'MD5'
    elif intent_field == 'INTENT_VAL_0':
        return 'Default'
    else:
        return ''

def get_source_desc(source_field):
    if source_field=='SOURCE_VAL_1':
        return '북한IP'
    if source_field=='SOURCE_VAL_3':
        return 'ECSC Black IP'
    else:
        return ''    
# New assets column

## ASSETS_VAL을 아예 JSON항목으로 만들어서 새로운 데이터프레임으로 생성.
risk_df = pd.DataFrame()
for risk in RISK_V2_FILTERED:
    risk = risk.replace("'", "\"") #json으로 만들려고.
    json_string = json.loads(risk)
    json_df = json_normalize(json_string)
    risk_df = pd.concat([risk_df,json_df],ignore_index=True)    #DataFrame 합쳐주기. ignore_index = True를 해야 index가 재구성 된다.
risk_df_column_names = risk_df.columns

assets_df = []
intents_df = []
sources_df = []
def filter_all(risk):
    for i in range(0,len(risk)):
        risks=[]
        intents=[]
        sources=[]
        for column in risk_df_column_names:
            # filter_asset
            if 'ASSETS_VAL_' in column and risk.iloc[i][column]:
                risk_key_desc = 'RISK_V2.' + column + '=' + get_asset_desc(column)
                risks.append(risk_key_desc)
            
            # filter_intent
            if 'INTENT_VAL_' in column and risk.iloc[i][column]:
                intent_key_desc = 'RISK_V2.' + column + '=' + get_intent_desc(column)
                intents.append(intent_key_desc)
                
            if 'SOURCE_VAL_' in column and risk.iloc[i][column]:
                source_key_desc='RISK_V2.' + column + '=' + get_source_desc(column)
                sources.append(source_key_desc)
        
        assets_df.append(risks)
        intents_df.append(intents)
        sources_df.append(sources)
        
filter_all(risk_df)
## 여기까지 내가 만든 것.

NTM_df['ASSETS_VAL'] = assets_df
NTM_df['ASSETS_VAL'] = NTM_df['ASSETS_VAL'].astype(str)
NTM_df['ASSETS_VAL'] = NTM_df['ASSETS_VAL'].str.replace('[','',regex=True)
NTM_df['ASSETS_VAL'] = NTM_df['ASSETS_VAL'].str.replace(']','',regex=True)
NTM_df['ASSETS_VAL']

NTM_df['INTENT_VAL'] = intents_df
NTM_df['INTENT_VAL'] = NTM_df['INTENT_VAL'].astype(str)
NTM_df['INTENT_VAL'] = NTM_df['INTENT_VAL'].str.replace('[','',regex=True)
NTM_df['INTENT_VAL'] = NTM_df['INTENT_VAL'].str.replace(']','',regex=True)
NTM_df['INTENT_VAL']

NTM_df['SOURCE_VAL'] = sources_df
NTM_df['SOURCE_VAL'] = NTM_df['SOURCE_VAL'].astype(str)
NTM_df['SOURCE_VAL'] = NTM_df['SOURCE_VAL'].str.replace('[','',regex=True)
NTM_df['SOURCE_VAL'] = NTM_df['SOURCE_VAL'].str.replace(']','',regex=True)
NTM_df['SOURCE_VAL']
NTM_df.drop(columns=['RISK_V2'], inplace=True)


##################### 여기서부터 진행하시면 됩니다. #####################
##################### 아래 12개 아이템(12. 장비 ACCD_FIND_MTD_CODE 제외)에 대해서 모든 아이템 조합에 알고리즘 적용하기#####################

# It should be 13 columns in total

# 1. 기관 INST_NM
# 2. 공격 DRULE_ATT_TYPE_CODE1
# 3. 자산 ASSETS_VAL
# 4. 위협공격ip TW_ATT_IP
# 5. 위협공격port TW_ATT_PORT
# 6. 위협피해ip TW_DMG_IP
# 7. 위협피해port TW_DMG_PORT
# 8. 위협피해프로토콜 ACCD_DMG_PROTO_NM
# 9. 공격국가 TW_ATT_CT_NM
# 10. 의도(7개) INTENT_VAL
# 11. IP/URL 가중치 SOURCE_VAL
# 12. 장비 ACCD_FIND_MTD_CODE
# 13. 탐지규칙명 DRULE_NM

NTM_df.isna().sum()

# Change the Nan to zero
NTM_df['ACCD_DMG_PROTO_NM']=NTM_df['ACCD_DMG_PROTO_NM'].replace(np.nan,'')
NTM_df['INST_NM']=NTM_df['INST_NM'].replace(np.nan,'')
NTM_df['DRULE_ATT_TYPE_CODE1']=NTM_df['DRULE_ATT_TYPE_CODE1'].replace(np.nan,'')
NTM_df['TW_ATT_IP']=NTM_df['TW_ATT_IP'].replace(np.nan,0)
NTM_df['TW_ATT_PORT']=NTM_df['TW_ATT_PORT'].replace(np.nan,0)
NTM_df['TW_DMG_IP']=NTM_df['TW_DMG_IP'].replace(np.nan,0)
NTM_df['TW_DMG_PORT']=NTM_df['TW_DMG_PORT'].replace(np.nan,0)
NTM_df['TW_ATT_CT_NM']=NTM_df['TW_ATT_CT_NM'].replace(np.nan,'')
NTM_df['ASSETS_VAL']=NTM_df['ASSETS_VAL'].replace(np.nan,0)
NTM_df['INTENT_VAL']=NTM_df['INTENT_VAL'].replace(np.nan,0)
NTM_df['SOURCE_VAL']=NTM_df['SOURCE_VAL'].replace(np.nan,0)
NTM_df['DRULE_NM']=NTM_df['DRULE_NM'].replace(np.nan,'')
NTM_df['TW_ATT_IP']=NTM_df['TW_ATT_IP']
NTM_df['TW_ATT_PORT']=NTM_df['TW_ATT_PORT']
NTM_df['TW_DMG_IP']=NTM_df['TW_DMG_IP']
NTM_df['TW_DMG_PORT']=NTM_df['TW_DMG_PORT']
# Check NaN out again
NTM_df.isna().sum()

copy_df = NTM_df.copy() #원본도 안건드리고, 실행시킬 때마다 오류 떠서 copy로 하는게 좋을 것 같다.
copy_df.drop(columns=['ACCD_FIND_MTD_CODE'],inplace=True)
data_len = len(NTM_df)
    
# Combination
import itertools

# Combination 조합들 생성하는 함수. row마다 mCn 생성.
def get_comb_df(df, n):
    nCr = list(itertools.combinations(df.columns.tolist(),n))
    nCr = [column for column in nCr if 'DRULE_ATT_TYPE_CODE1' in column]
    ret_list = []
    for l in range(len(nCr)):
        for i in range(len(df)):
            temp = []
            temp_df = df.loc[i]
            for col in nCr[l]:
                new_string = col
                new_string = new_string + ":" + str(temp_df[col])
                temp.append(new_string)
            ret_list.append(temp)
    return ret_list                  

# item들은 이 순서다.
#item1 = 'INST_NM'
#item2 = 'DRULE_ATT_TYPE_CODE1'
#item3 = 'TW_ATT_IP'
#item4 = 'TW_ATT_PORT'
#item5 = 'TW_DMG_IP'
#item6 = 'TW_DMG_PORT'
#item7 = 'ACCD_DMG_PROTO_NM'
#item8 = 'TW_ATT_CT_NM'
#item9 = 'DRULE_NM'
#item10 = 'ASSETS_VAL'
#item11 = 'INTENT_VAL'
#item12 = 'SOURCE_VAL'

nonnull_list = []
for i in range(0,data_len):
    item1 = 'INST_NM:' + NTM_df.loc[i]['INST_NM']
    item2 = 'DRULE_ATT_TYPE_CODE1:' + NTM_df.loc[i]['DRULE_ATT_TYPE_CODE1']
    item3 = 'TW_ATT_IP:' + NTM_df.loc[i]['TW_ATT_IP'].astype(str)
    item4 = 'TW_ATT_PORT:' + NTM_df.loc[i]['TW_ATT_PORT'].astype(str)
    item5 = 'TW_DMG_IP:' + NTM_df.loc[i]['TW_DMG_IP'].astype(str)
    item6 = 'TW_DMG_PORT:' + NTM_df.loc[i]['TW_DMG_PORT'].astype(str)
    item7 = 'ACCD_DMG_PROTO_NM:' + NTM_df.loc[i]['ACCD_DMG_PROTO_NM']
    item8 = 'TW_ATT_CT_NM:' + NTM_df.loc[i]['TW_ATT_CT_NM']
    item9 = 'DRULE_NM:' + NTM_df.loc[i]['DRULE_NM']
    item10 = NTM_df.loc[i]['ASSETS_VAL']
    item11 = NTM_df.loc[i]['INTENT_VAL']
    item12 = NTM_df.loc[i]['SOURCE_VAL']
    not_null_arr = []
    ## 리스트안에 빈 값을 빼버리자.
    null_check_list = [item1,item2,item3,item4,item5,item6,item7,item8,item9,item10,item11,item12]
    for item in null_check_list:
        if item and item != '[]':
            not_null_arr.append(item)
    nonnull_list.append(not_null_arr)
    
get_comb_df(copy_df,9)

def get_prefix_span(df_list, n): #n이상 길이를 갖는 규칙들만. 거기다가 Frequency기준 정렬 까지.
    prefix_span = PrefixSpan(df_list)
    n_ps = prefix_span.frequent(1,filter = lambda patt, matches:len(patt)>n)
    ps_df = pd.DataFrame(n_ps)
    ps_df.rename(columns={0:'Frequency', 1:'Cause'}, inplace=True)
    ps_df['Effect']= np.nan
    ps_df = ps_df[['Cause','Effect','Frequency']]
    ps_sort_df = ps_df.sort_values(by=['Frequency'],ascending=False,ignore_index=True)
    return ps_sort_df
  
test = get_prefix_span(nonnull_list,8)
test

# Define the function that find the rule name 
# 데이터 크기를 줄여서 실행해본 결과 정상 작동함.
def get_Effect(df):
    for i in range(0,10000):
        drules=['Attack','DDOS','HACK','MAIL','Malwr','WEB']
        for item in df.loc[i,'Cause']:
            for drule in drules:
                drule_str = 'DRULE_ATT_TYPE_CODE1:' + drule
                if item == drule_str:
                    df.loc[i,'Effect'] = drule
                    break         
    return df                

tdf = get_Effect(test)
tdf.head(10000)	# 10000개로 했을 때, DRULE_ATT_TYPE_CODE 가 있는 항목들은 Effect정상 추출.
tdf = get_Effect(test)
testdf = tdf.tail(1000)
testdf

### Effect = NaN인 값 지우기.
testdf = tdf.head(1000)
test_result = testdf
testdf
test_result['Effect'] = test_result['Effect'].replace(np.nan,0)
test_result = test_result[test_result.Effect != 0]
test_result.reset_index(drop=True)

# 정규표현식 사용해서 매칭하기.
# 정규표현식 사용하는 틀. words에 배열만 넣으면 된다.
tdf['Cause'] = [','.join(map(str, word))for word in tdf['Cause']]

def regbase(words):
    base = r'^{}'
    expr = '(?=.*{})'
    ret = base.format(''.join(expr.format(w) for w in words))
    return ret

def result(n):
    comlist = get_comb_df(copy_df,n)
    for i in range(0,len(comlist)):
        print(comlist[i])
        print(tdf[tdf['Cause'].str.contains(regbase(comlist[i]),na=False,regex=True)].reset_index(drop=True,inplace=False))