'''
기사 분류기
나무위키로 학습한 워드 벡터를 이용.
주제별 단어와 기사의 유사성을 측정하여 max 분류로 한다.
기사와의 유사성 측정은?
기사내의 단어(조사 등 제외) 분포를 찾아 의미 있는 것을 찾아,스코어화.
'''
import codecs
from bs4 import BeautifulSoup
from konlpy.tag import Okt
from gensim.models import word2vec
import time
import numpy as np

category = ['정치', '경제', '사회', 'IT', '과학', '자동차', '부동산',
            '생활', '세계', '의학', '인테리어', '예술', '연예']

filename='namu_test_article.txt'
filename4 = './hub/namu.model'


def make_DF(filename):
    file = codecs.open(filename, 'r', encoding='utf-8')
    text = file.read()

    twitter = Okt()
    word_dic = {}
    lines = text.split('\r\n')
    for line in lines:
        malist = twitter.pos(line)
        for taeso, pumsa in malist:
            if pumsa == 'Noun':
                if not (taeso in word_dic):
                    word_dic[taeso] = 0
                word_dic[taeso] += 1
    print(word_dic)
    keys = sorted(word_dic.items(), key=lambda x: x[1], reverse=True)

    top20_dic = {}
    if len(keys)>20:
        for word, count in keys[:20]:
            top20_dic[word]=count
    else:
        for word, count in keys:
            top20_dic[word]=count
    return top20_dic


print('Model test')
t1 = time.time()
model = word2vec.Word2Vec.load(filename4)
t2 = time.time()
print('model load elapsed=', t2-t1)
top20_dic=make_DF(filename)
for ks in top20_dic.keys():
    print(ks, top20_dic[ks], end=" ,")

# 카테고리별 단어의 유사도
cascores=[]
for ca in category:
    sims = []
    dfs = []
    for ks in top20_dic.keys():
        try:
            v1 = model.similarity(ca, ks)
            sims.append( v1 )
        except KeyError:
            sims.append( 0.0 )
        v2 = top20_dic[ks]
        dfs.append( v2 )
        print(ca, ks, 'similarity=',v1, 'df=',v2)

    sims = np.asarray(sims)
    dfs = np.asarray(dfs)

    # 단어출연 빈도를 가중치로 한 스코어
    val = np.dot(sims, dfs)
    print('wsum=', val)
    sco=val/ np.sum(dfs)
    print('scor=', sco)
    cascores.append(sco)

cascores=np.asarray(cascores)
maxidx = np.argmax(cascores)

# print(category)
# print(cascores, maxidx)

categorydic = {
    cate:scor for cate, scor in zip(category, cascores)
}
pc=sorted(categorydic, key=lambda k:categorydic[k], reverse=True)
print(pc)
print( sorted(cascores, reverse=True) )
print( 'predict=',pc[0],"/",pc[1] )

Model test

model load elapsed= 21.66822624206543

{'디넷': 1, '코리아': 1, '봉삼': 1, '기자': 1, '방탄소년단': 3, '컴백': 2, '카카오': 2, '운영': 1, '음원': 1, '플랫폼': 1, '멜론': 13, '이': 1, '먹통': 1, '이용자': 3, '불편': 3, '오후': 1, '접속': 4, '문제': 2, '발생': 2, '오류': 3, '등': 2, '트래픽': 2, '번': 2, '불': 1, '안정': 1, '현상': 1, '것': 1, '추정': 1, '실행': 1, '인터넷': 2, '수': 2, '상태': 2, '확인': 1, '메시지': 1, '지속': 1, '검색': 1, '포털': 1, '실시간': 1, '급상승': 1, '검색어': 1, '단어': 1, '그': 1, '만큼': 1, '사용자': 1, '오류로': 1, '뜻': 1, '로고': 2, '팬': 1, '게시판': 1, '통해': 1, '호소': 1, '컴퓨터': 1, '겨우': 1, '몇': 1, '재생': 1, '목록': 1, '장난': 1, '말': 1, '다른': 1, '서버': 2, '위': 1, '관리': 1, '좀': 1, '측은': 1, '일시': 1, '면서': 1, '조속': 1, '정상화': 1, '고': 1}

멜론 13 ,접속 4 ,방탄소년단 3 ,이용자 3 ,불편 3 ,오류 3 ,컴백 2 ,카카오 2 ,문제 2 ,발생 2 ,등 2 ,트래픽 2 ,번 2 ,인터넷 2 ,수 2 ,상태 2 ,로고 2 ,서버 2 ,디넷 1 ,코리아 1 

  v1 = model.similarity(ca, ks)

정치 멜론 similarity= 0.10295023 df= 13

정치 접속 similarity= 0.14711404 df= 4

정치 방탄소년단 similarity= 0.14685516 df= 3

정치 이용자 similarity= 0.34247914 df= 3

정치 불편 similarity= 0.20862463 df= 3

정치 오류 similarity= 0.22307779 df= 3

... 중략....

연예 인터넷 similarity= 0.42752212 df= 2

연예 수 similarity= 0.11581904 df= 2

연예 상태 similarity= 0.17616633 df= 2

연예 로고 similarity= 0.3842223 df= 2

연예 서버 similarity= 0.114044465 df= 2

연예 디넷 similarity= 0.114044465 df= 1

연예 코리아 similarity= 0.3931188 df= 1

wsum= 14.374487452208996

scor= 0.2613543173128908

['연예', 'IT', '경제', '부동산', '사회', '정치', '예술', '과학', '세계', '자동차', '생활', '인테리어', '의학']

[0.2613543173128908, 0.24636683274399152, 0.22802816277200524, 0.22646296511996877, 0.20992791205644606, 0.2065649455243891, 0.19927997887134552, 0.18424408056519248, 0.17951921116222036, 0.17361825514923443, 0.15364200567657296, 0.14810618719255383, 0.13615213415839456]

predict= 연예 / IT

[{"namespace":"0","title":"!","text":"#redirect \ub290\ub08c\ud45c\n","contributors":["namubot","R:hoon12560"]},{"nam

espace":"0","title":"!!\uc544\uc557!!","text":"[[\ud30c\uc77c:3444050440.jpg]]\n([[\uc2e0 \uc138\uacc4\uc218\uc758 \u

bbf8\uad81 2]]\uc5d0\uc11c \ub72c !!\uc544\uc557!!)\n{{{+1 \uff01\uff01\u3042\u3042\u3063\u3068\uff01\uff01 }}}\n\n[[

\uc138\uacc4\uc218\uc758 \ubbf8\uad81 \uc2dc\ub9ac\uc988]]\uc5d0 \uc804\ud1b5\uc73c\ub85c \ub4f1\uc7a5\ud558\ub294 \u

b300\uc0ac. [[\uc138\uacc4\uc218\uc758 \ubbf8\uad81 2 \uc81c\uc655\uc758 \uc131\ubc30|2\ud3b8 \uc81c\uc655\uc758 \uc1

31\ubc30]]\ubd80\ud130 \ub4f1\uc7a5\ud588\uc73c\uba70, \ud6cc\ub96d\ud55c [[\uc0ac\ub9dd \ud50c\ub798\uadf8]]\uc758 \

uc608\uc2dc\uc774\ub2e4.\n\n\uc138\uacc4\uc218\uc758 \ubaa8\ud5d8\uac00\ub4e4\uc774 \ud0d0\ud5d8\ud558\ub294 \ub358\u

c804\uc778 \uc218\ud574\uc758 \uad6c\uc11d\uad6c\uc11d\uc5d0\ub294 \ucc44\ucde8\/\ubc8c\ucc44\/\ucc44\uad74 \ud3ec\uc

778\ud2b8\uac00 \uc788\uc73c\uba70, \uc774\ub97c \uc704\ud55c \ucc44\uc9d1 \uc2a4\ud0ac\uc5d0 \ud22c\uc790\ud558\uba7

4 \uc81c\ud55c\ub41c \ucc44\uc9d1 \uae30\ud68c\uc5d0 \ubcf4\ub2e4 \ud070 \uc774\ub4dd\uc744 \ucc59\uae38 \uc218 \uc78

8\ub2e4. \uadf8\ub7ec\ub098 \ubd84\ubc30\ud560 \uc218 \uc788\ub294 \uc2a4\ud0ac \ud3ec\uc778\ud2b8\ub294 \ud55c\uc815

\ub41c \ub9cc\ud07c \ucc44\uc9d1 \uc2a4\ud0ac\uc5d0 \ud22c\uc790\ud558\ub294 \ub9cc\ud07c \uc804\ud22c \uc2a4\ud0ac \

ub808\ubca8\uc740 \ub0ae\uc544\uc9c0\uac8c \ub41c\ub2e4.\n\n 1. \ucc44\uc9d1\uc6a9 \uce90\ub9ad\ud130\ub4e4\ub85c \uc

774\ub8e8\uc5b4\uc9c4 \uc57d\ud55c \ud30c\ud2f0(ex: [[\ub808\uc778\uc800(\uc138\uacc4\uc218\uc758 \ubbf8\uad81 2)|\ub

808\uc778\uc800]] 5\uba85)\uac00 \uc218\ud574\uc5d0 \uc785\uc7a5\ud55c\ub2e4.\n 1. \ud544\ub4dc \uc804\ud22c\ub97c \u

d68c\ud53c\ud558\uba74\uc11c \ucc44\uc9d1 \ud3ec\uc778\ud2b8\uc5d0 \ub3c4\ucc29\ud574 \uc5f4\uc2ec\ud788 \uc544\uc774

\ud15c\uc744 \uce90\ub294 \uc911\uc5d0...\n ..............................

import ijson

def load_json(filename):
    count=0
    with open(filename, 'r') as fd:
        parser = ijson.parse(fd)
        for prefix, event, value in parser:
            if prefix.endswith('.title'):
                print('index=', count+1)
                print("\nTITLE: %s" % value)
            elif prefix.endswith('.text'):
                print("\nCONTENT: %s" % value)
                count += 1
            # debug
            if count==10 :
                break

if __name__ == "__main__":
    print('load_json start')
    load_json('/home/psychic/download/namuwiki_20180326.json')
    print('load_json end')

tail -c 800 mini_namu.json  | grep -E "title|text"

:"#redirect \uc81c18\ub300 \uad6d\ud68c\uc758\uc6d0 \uc120\uac70\n","contributors":["namubot","R:ttgrgt24","211.187.44.56"]},{"namespace":"0","title":"18\ub85c \uc81c\ud6c4","text":"[include(\ud2c0:18\ub85c \uc81c\ud6c4)]\n\n\u5341\u516b\u8def\u8af8\u4faf\n\n[\ubaa9\ucc28]\n== \uc124\uba85 ==\n[[\ud6c4\ud55c]]\ub9d0 [[190\ub144]], [[\uc5ed\uc801]] [[\ub3d9\ud0c1#s-1|\ub3d9\ud0c1]]\uc744 \ud1a0\ubc8c\ud558\uace0 [[\ud55c\ub098\ub77c#s-1.2|\ud55c\ub098\ub77c]] \ud669\uc2e4\uc744 \uad73\uac74\ud788 \ubc1b\ub4e0\ub2e4\ub294 \uac83\uc744 \ub300\uc758 \uba85\ubd84\uc73c\ub85c \ud558\uc5ec \uc77c\uc5b4\uc120 [[\uc81c\ud6c4]]\ub4e4\uc758 [[\uc5f0\ud569]]. \uc774\ub978\ubc14 '[[\ubc18\ub3d9\ud0c1 \uc5f0\ud569]]'\uc774\ub77c\uace0 \ubd80\ub974\uba70, \ud568\uace1\uad00\uc758 \ub3d9\ucabd \uc9c0\ubc2

#echo -n "0\"}]" >> mini_namu.json

#tail -c 100 mini_namu.json 

ud569]]'\uc774\ub77c\uace0 \ubd80\ub974\uba70, \ud568\uace1\uad00\uc758 \ub3d9\ucabd \uc9c0\ubc20"}] 

import ijson

import codecs
import sys, os
import time

import nltk

from gensim.models import word2vec
from konlpy.tag import Okt

import pickle


'''
JSON 나무위키 파싱

'''


def load_json(filename):
    count=0
    with open(filename, 'r') as fd:
        # parser = ijson.parse(fd)
        # for prefix, event, value in parser:
        #      print ( 'prefix=', prefix, 'event=', event )

        #     if prefix.endswith('.title'):
        #         # print('index=', count+1)
        #         # print("\nTITLE: %s" % value)
        #     elif prefix.endswith('.text'):
        #         # print("\nCONTENT: %s" % value)
        #         count += 1
        # print ('total cnt=', count)
        for item in ijson.items(fd, 'item'):
            print(item['title'])
            # print(item['text'])

def load_and_write_content(filename, filename2):
    count=0
    file = codecs.open(filename2, 'w', encoding='utf-8')
    with open(filename, 'r') as fd:
        for item in ijson.items(fd, 'item'):
            count+=1
            file.write('[[제목]]: ')
            file.write(item['title'])
            file.write('\n')
            file.write('[[내용]]: \n')
            file.write(item['text'])
            file.write("\n")
    file.close()
    print('contents count=', count)


def make_wakati(f1, f2, f3):
    file = codecs.open(f1, 'r', encoding='utf-8')
    text = file.read()

    twitter = Okt()
    lines = text.split('\r\n')
    results = []
    for line in lines:
        r = []
        malist = twitter.pos(line)      # part of speech (POS)
        # pumsa : Noun, Josa, Verb, Eomi, Punctuation, Number, KoreanParticle, 
        for word, pumsa in malist:
            if not pumsa in ['Josa', 'Eomi', 'Punctuation']:
                r.append(word)
        results.append((" ".join(r)).strip())
    output = (" ".join(results)).strip()

    with open(f2, "w", encoding="utf-8") as fp:
        fp.write(output)

    print('make word2vec')
    data = word2vec.LineSentence(f2)
    model = word2vec.Word2Vec(data, size=200, window=10, hs=1, min_count=5, sg=1, workers=3)
    model.save(f3)
    print('end word2vec')

if __name__ == "__main__":
    print('load_json start')
    print('curdir=', os.getcwd())
    filedir = os.path.dirname( os.path.realpath(__file__))
    os.chdir(filedir)
    print('filedir=', filedir)

    filename ='mini_namu.json'
    filename2='mini_namu.txt'
    filename3='mini_namu.wakati'
    filename4='mini_namu.model'
    filename5='mini_namu.pkl'

    # load_json(filename)
    # 나무위키 JSON DB에서 제목과 컨텐트를 스트링으로 기록한 파일 생성. (txt)
    if True:
        print('Create WordTxT')
        t1=time.time()
        load_and_write_content(filename, filename2)
        t2=time.time()
        print('End WordTxt ', 'time=',t2-t1)

    if True:
        print('Create Wakati')
        t1=time.time()
        make_wakati(filename2, filename3, filename4)
        t2=time.time()
        print('End Wakati', 'time=',t2-t1)

    if True:
        print('Model test')
        t1=time.time()
        model = word2vec.Word2Vec.load(filename4)
        t2=time.time()
        print(model.most_similar(positive=["조선", "일본"]))
        print(model.most_similar(positive=["고려"]))
        print(model.most_similar(positive=["고려"], negative=["왕"]))
        print(model.most_similar(positive=["고려", "경제"]))
        print(model.most_similar(positive=["RPG", "게임"]))
        print(model.most_similar(positive=["임진왜란"]))
        print(model.most_similar(positive=["왕", "여자"], negative=["남자"]))

        print(model.similarity('고려','공민왕'))
        print(model.doesnt_match("아침 점심 저녁 조선 밥".split() ) )
        print(model.doesnt_match("총 무기 칼 게임 하늘".split() ) )

        print('time=',t2-t1)

[('1536년', 0.7473468780517578), ('음력', 0.7256854772567749), ('이이', 0.7070702910423279), ('도요토미', 0.7009462118148804), ('1552년', 0.6966617703437805), ('정조', 0.6919983625411987), ('문인', 0.6826009750366211), ('메이지', 0.6784789562225342), ('덴노', 0.6729492545127869), ('1337년', 0.6693580150604248)]

[('폐위', 0.7308624982833862), ('공민왕', 0.7183995246887207), ('정종', 0.7026641368865967), ('충숙왕', 0.7024295926094055), ('인종', 0.6962566375732422), ('즉위', 0.6933885216712952), ('충혜왕', 0.6890701651573181), ('명종', 0.6795019507408142), ('무신정권', 0.6763734817504883), ('원나라', 0.6668024659156799)]

[('마진', 0.2932665944099426), ('국호', 0.28764602541923523), ('발전', 0.27793973684310913), ('인종', 0.26443612575531006), ('말기', 0.2635525166988373), ('부족하다는', 0.2609822154045105), ('경향', 0.2581954002380371), ('1.6', 0.2573755383491516), ('줄어든', 0.2572077512741089), ('시기', 0.25421619415283203)]

[('천도', 0.6886128783226013), ('무신정권', 0.6836416721343994), ('인종', 0.6704342365264893), ('요나라', 0.6599131226539612), ('공민왕', 0.6575653553009033), ('권력자', 0.6571816205978394), ('도모', 0.6510674357414246), ('신변', 0.6495651006698608), ('폐위', 0.646056056022644), ('몽골', 0.6421409845352173)]

[('동봉', 0.6660709381103516), ('FPS', 0.6493582129478455), ('플래시', 0.6356097459793091), ('슈팅', 0.6346113681793213), ('로더', 0.629481852054596), ('플레이스테이션', 0.6227577924728394), ('TECHNIKA', 0.6164207458496094), ('DJMAX', 0.6132534742355347), ('온라인게임', 0.6100178360939026), ('오어', 0.606633186340332)]

[('1598년', 0.9288400411605835), ('이순신', 0.9017009735107422), ('1618년', 0.7677931189537048), ('1821년', 0.7676668167114258), ('도요토미', 0.7628185749053955), ('1865년', 0.7540280222892761), ('1839년', 0.7511029243469238), ('1868년', 0.7489482760429382), ('문인', 0.7481677532196045), ('1863년', 0.7473239898681641)]

[('속특', 0.4879986345767975), ('왕비', 0.48508623242378235), ('성종', 0.4771915674209595), ('선조', 0.4723415970802307), ('샤', 0.46159034967422485), ('세조', 0.4432491660118103), ('충혜왕', 0.43805891275405884), ('후궁', 0.4356191158294678), ('철종', 0.4354616701602936), ('문종', 0.43034279346466064)]

0.7184

조선

게임

make_wakati 수정

# make wakati file
twitter = Okt(max_heap_size=1024*8)     # heap size.. 8G
lines = text.split('\r\n')
t1=time.time()

print('making wakati start')
print('lines count=', len(lines))
fp = open(f2, "w", encoding="utf-8")
for line in lines:
    linelen = len(line)
    print("line length=", linelen)

    # split.
    # lineparts = map(''.join, zip(*[iter(line)] * 1000*1000*20))    # 20MB / 1line
    blocksize = 1000*1000*20
    if linelen % blocksize == 0:
        blockcnt = int(linelen/blocksize)
    else:
        blockcnt = int(linelen / blocksize) + 1

    for li in range(blockcnt):
        if li==blockcnt-1:
            linepart = line[li*blocksize:]
        else:
            linepart = line[li*blocksize:(li+1)*blocksize]
        print('progress=', li, '/', blockcnt, len(linepart) )
        malist = twitter.pos(linepart)      # part of speech (POS)
        # pumsa : Noun, Josa, Verb, Eomi, Punctuation, Number, KoreanParticle,
        for word, pumsa in malist:
            if not pumsa in ['Josa', 'Eomi', 'Punctuation']:
                fp.write(word.strip())
                fp.write(" ")
fp.close()
t2=time.time()
print('making wakati end time=', t1-t1)

다음 날 보니 wakati 파일 생성완료. 4G 크기.

질의 내용

print(model.most_similar(positive=["조선", "일본"]))
print(model.most_similar(positive=["고려"]))
print(model.most_similar(positive=["고려"], negative=["왕"]))
print(model.most_similar(positive=["고려", "경제"]))
print(model.most_similar(positive=["RPG", "게임"]))
print(model.most_similar(positive=["임진왜란"]))
print(model.most_similar(positive=["왕", "여자"], negative=["남자"]))

print(model.similarity('고려','공민왕'))
print(model.doesnt_match("아침 점심 저녁 조선 밥".split() ) )
print(model.doesnt_match("총 무기 칼 게임 하늘".split() ) )

Model test

[('대한제국', 0.6866917610168457), ('중국', 0.6801478862762451), ('우리나라', 0.6566961407661438), ('청나라', 0.6511589288711548), ('일제강점기', 0.6480269432067871), ('한국', 0.6324470043182373), ('조선국', 0.6295247077941895), ('경술국치', 0.6277073621749878), ('임진왜란', 0.623056173324585), ('역사', 0.6140305995941162)]

[('조선', 0.6692752838134766), ('고려왕조', 0.6515953540802002), ('공민왕', 0.6487005949020386), ('문종', 0.6360877752304077), ('현종', 0.6323516368865967), ('무신정권', 0.6292731761932373), ('태조', 0.6256691217422485), ('감안', 0.6252619624137878), ('冊封', 0.6246188879013062), ('충렬왕', 0.6229778528213501)]

[('감안', 0.4361225962638855), ('염두', 0.3996908664703369), ('제승방략', 0.39954277873039246), ('유념', 0.3989650011062622), ('바람직하지', 0.39659619331359863), ('해보아야', 0.39381834864616394), ('갖추자는', 0.39022499322891235), ('추세였음', 0.38095635175704956), ('타당', 0.3788178563117981), ('받아들여지기에', 0.37214767932891846)]

[('정책', 0.7264151573181152), ('동북아', 0.702474057674408), ('사회', 0.7002162933349609), ('정치', 0.6759592294692993), ('불평등', 0.6736841797828674), ('노동시장', 0.6685125827789307), ('경제정책', 0.668291449546814), ('한반도', 0.6653428673744202), ('민생', 0.6610323190689087), ('발전', 0.6543104648590088)]

[('TPS', 0.757465124130249), ('롤플레잉', 0.7494930028915405), ('로그라이크', 0.7455687522888184), ('FPS', 0.7437580823898315), ('플랫포머', 0.7379695177078247), ('RTS', 0.7236455082893372), ('모바일', 0.7205138206481934), ('MOBA', 0.718673586845398), ('SRPG', 0.7134039402008057), ('MMORPG', 0.7068561911582947)]

[('정유재란', 0.8903812170028687), ('병자호란', 0.8541854619979858), ('임란', 0.8498165607452393), ('왜란', 0.8400564193725586), ('정묘호란', 0.7854044437408447), ('1592년', 0.7841730117797852), ('의병장', 0.781627893447876), ('류성룡', 0.7787343263626099), ('왜군', 0.7751662731170654), ('광해군', 0.7695599794387817)]

[('왕인', 0.6398370265960693), ('준왕', 0.6377674341201782), ('왕중', 0.5966447591781616), ('극', 0.5917797088623047), ('최후', 0.5905873775482178), ('선왕', 0.5905437469482422), ('왕전', 0.588803231716156), ('中山王', 0.5885385274887085), ('楚穆王', 0.5835995674133301), ('南平王', 0.5762388706207275)]

0.648701

조선

게임

def is_divide_pt(x11,y11, x12,y12, x21,y21, x22,y22):

    '''input: 4 points
    output: True/False
    '''
    #  // line1 extension이 line2의 두 점을 양분하는지 검사..
    # 직선의 양분 판단
    f1= (x12-x11)*(y21-y11) - (y12-y11)*(x21-x11)
    f2= (x12-x11)*(y22-y11) - (y12-y11)*(x22-x11)
    if f1*f2 < 0 :
      return True
    else:
      return False

def is_cross_pt(x11,y11, x12,y12, x21,y21, x22,y22):
    b1 = is_divide_pt(x11,y11, x12,y12, x21,y21, x22,y22)
    b2 = is_divide_pt(x21,y21, x22,y22, x11,y11, x12,y12)
    if b1 and b2:
        return True

    return False

import numpy as np
import matplotlib.pyplot as plt


def get_crosspt(x11,y11, x12,y12, x21,y21, x22,y22):
    if x12==x11 or x22==x21:
        print('delta x=0')
        return None
    m1 = (y12 - y11) / (x12 - x11)
    m2 = (y22 - y21) / (x22 - x21)
    if m1==m2:
        print('parallel')
        return None
    print(x11,y11, x12, y12, x21, y21, x22, y22, m1, m2)
    cx = (x11 * m1 - y11 - x21 * m2 + y21) / (m1 - m2)
    cy = m1 * (cx - x11) + y11

    return cx, cy


x11 =np.random.randint(1,50)
y11 =np.random.randint(1,50)
x12 =np.random.randint(50,100)
y12 =np.random.randint(50,100)
x21 =np.random.randint(1,50)
y21 =np.random.randint(50,100)
x22 =np.random.randint(50,100)
y22 =np.random.randint(1,50)

plt.figure()
plt.plot([x11,x12], [y11,y12], c='r')
plt.plot([x21,x22], [y21,y22], c='b')

cx, cy = get_crosspt(x11, y11, x12, y12, x21, y21, x22, y22)
plt.plot(cx, cy, 'ro')
plt.show()

# line1:(x11,y11)-(x12,y12) , line2:(x21,y21)-(x22,y22)

def get_crosspt(x11,y11, x12,y12, x21,y21, x22,y22):
    if x12==x11 or x22==x21:
        print('delta x=0')
        if x12==x11:
            cx = x12
            m2 = (y22 - y21) / (x22 - x21)
            cy = m2 * (cx - x21) + y21
            return cx, cy
        if x22==x21:
            cx = x22
            m1 = (y12 - y11) / (x12 - x11)
            cy = m1 * (cx - x11) + y11
            return cx, cy

    m1 = (y12 - y11) / (x12 - x11)
    m2 = (y22 - y21) / (x22 - x21)
    if m1==m2:
        print('parallel')
        return None
    print(x11,y11, x12, y12, x21, y21, x22, y22, m1, m2)
    cx = (x11 * m1 - y11 - x21 * m2 + y21) / (m1 - m2)
    cy = m1 * (cx - x11) + y11

    return cx, cy