PrefixSpan/HwanWooCode/prefix_span.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.

## ASSETS_VAL 확인
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,'')

# Check NaN out again
NTM_df.isna().sum()

# # Merge all

# # Make one string from all of elements
NTM_df['Combined']=NTM_df['INST_NM'].astype(str)+' '+NTM_df['TW_ATT_IP'].astype(str)+' '+NTM_df['TW_ATT_PORT'].astype(str)+' '+NTM_df['TW_DMG_IP'].astype(str)+' '+NTM_df['TW_DMG_PORT'].astype(str) +' '+NTM_df['ACCD_DMG_PROTO_NM'].astype(str)+' '+NTM_df['TW_ATT_CT_NM']+' '+NTM_df['ASSETS_VAL']+' '+NTM_df['INTENT_VAL']+' '+NTM_df['SOURCE_VAL']+' '+NTM_df['DRULE_ATT_TYPE_CODE1']+' '+NTM_df['DRULE_NM']

NTM_com=NTM_df['Combined']
## 내가 만든 컴바인
## 모든 조합을 돌릴거면, [1,2,3,4,5] 처럼 배열의 원소로 만들어서 넣는게 가장 베스트 아닌가.
## 그런데 순서를 다 바꿔줘야 하는데, 그건 어떻게 할 것인지 내일 물어보자.

data_len = len(NTM_df)
hwan_list = []
for i in range(0,data_len):
    accd_dmg_proto_nm = NTM_df.loc[i]['ACCD_DMG_PROTO_NM']
    inst_nm = NTM_df.loc[i]['INST_NM']
    drule_att_type_code1 = NTM_df.loc[i]['DRULE_ATT_TYPE_CODE1']
    tw_att_ip = NTM_df.loc[i]['TW_ATT_IP']
    tw_att_port = NTM_df.loc[i]['TW_ATT_PORT']
    tw_dmg_ip = NTM_df.loc[i]['TW_DMG_IP']
    tw_dmg_port = NTM_df.loc[i]['TW_DMG_PORT']
    tw_att_ct_nm = NTM_df.loc[i]['TW_ATT_CT_NM']
    assets_val = NTM_df.loc[i]['ASSETS_VAL']
    intent_val = NTM_df.loc[i]['INTENT_VAL']
    source_val = NTM_df.loc[i]['SOURCE_VAL']
    drule_nm = NTM_df.loc[i]['DRULE_NM']
    null_check_list = [accd_dmg_proto_nm, inst_nm, drule_att_type_code1, tw_att_ip, tw_att_port,
            tw_dmg_ip, tw_dmg_port, tw_att_ct_nm, assets_val, intent_val, source_val, drule_nm]
    not_null_arr = []
    ## 리스트안에 빈 값을 빼버리자.
    for item in null_check_list:
        if item and item != '[]':
            not_null_arr.append(item)
    hwan_list.append(not_null_arr)

new_ps = PrefixSpan(hwan_list)
# new_ps : hwan_list안에 순서대로 null값을 제외한 모든값들이 [1,2,3,4,5,6] 이런식으로 들어가 있는데,
# 이 값을 PrefixSpan 수행한 코드.

## 여기도 내 코드
test_ntm = new_ps.frequent(1)
test_ntm_df = pd.DataFrame(test_ntm)
test_ntm_df.rename(columns={0:'Frequency', 1:'Cause'}, inplace=True)
print(test_ntm_df)
test_sort_values = test_ntm_df.sort_values(by=['Frequency'],ascending=False,ignore_index=True)
# test_sort_values : PrefixSpan을 수행하여  Frequency가 나온 값. Frequency 를 기준으로 정렬했는데, 2900만가지나 된다.


# Effect Rule Name Case 1.
prefix_NTM_df = test_sort_values
for i in range(0,len(prefix_NTM_df)):
    drules=['Attack','DDOS','HACK','MAIL','Malwr','WEB']
    for drule in drules:
        if drule in prefix_NTM_df.Cause[i]:
            prefix_NTM_df.Effect[i] = drule
        elif prefix_NTM_df.Cause[i] == NaN:
            prefix_NTM_df.Effect[i] = ''

# Effect Rule Name Case 2.
prefix_NTM_df = test_sort_values

for i in range(0,len(prefix_NTM_df)):
    drules=['Attack','DDOS','HACK','MAIL','Malwr','WEB']
    for drule in drules:
        iloc_value = prefix_NTM_df.iloc[i]['Cause']
        if drule in iloc_value:
            prefix_NTM_df.Cause[i] = drule
            
# Case 1을 20분동안 돌렸는데도 결과가 나오질 않아서 iloc으로 바꾸어 Case2로 돌려보았다. 이것도 오래걸리는 중..

# Rule Name 적용한 코드는 prefix_NTM_df 에 적용.