BIDI_ML_INFO_EXTRACTION/BiddingKG/maxcompute/documentMerge.py

#coding:UTF8
from odps.udf import annotate
from odps.udf import BaseUDAF
from odps.udf import BaseUDTF


def getSet(list_dict,key):
    _set = set()
    for item in list_dict:
        if key in item:
            if item[key]!='' and item[key] is not None:
                if re.search("^[\d\.]+$",item[key]) is not None:
                    _set.add(str(float(item[key])))
                else:
                    _set.add(str(item[key]))
    return _set

def split_with_time(list_dict,sort_key,timedelta=86400*120):
    if len(list_dict)>0:
        if sort_key in list_dict[0]:
            list_dict.sort(key=lambda x:x[sort_key])
            list_group = []
            _begin = 0
            for i in range(len(list_dict)-1):
                if abs(list_dict[i][sort_key]-list_dict[i+1][sort_key])<timedelta:
                    continue
                else:
                    _group = []
                    for j in range(_begin,i+1):
                        _group.append(list_dict[j])
                    if len(_group)>1:
                        list_group.append(_group)
                    _begin = i + 1
            if len(list_dict)>1:
                _group = []
                for j in range(_begin,len(list_dict)):
                    _group.append(list_dict[j])
                if len(_group)>1:
                    list_group.append(_group)
            return list_group
    return [list_dict]

@annotate('bigint,bigint,string,string,string,string,string,string,bigint->string')
class f_merge_rule_limit_num_contain_greater(BaseUDAF):
    '''
    项目编号、中标单位、len(项目编号)>7、中标单位<> ""、合并后非空招标单位数<2、合并后同公告类型非空金额相同
    '''
    def __init__(self):
        import logging
        import json,re
        global json,logging,re
        logging.basicConfig(level = logging.INFO,format = '%(asctime)s - %(name)s - %(levelname)s - %(message)s')

    def new_buffer(self):
        return [list()]

    def iterate(self, buffer,docid,page_time_stamp,set_limit_column1,set_limit_column2,set_limit_column3,set_limit_column4,contain_column,greater_column,MAX_NUM):
        buffer[0].append({"docid":docid,"page_time_stamp":page_time_stamp,"set_limit_column1":set_limit_column1,
                          "set_limit_column2":set_limit_column2,"set_limit_column3":set_limit_column3,"set_limit_column4":set_limit_column4,
                          "contain_column":contain_column,"greater_column":greater_column,"MAX_NUM":MAX_NUM})

    def merge(self, buffer, pbuffer):
        buffer[0].extend(pbuffer[0])

    def terminate(self, buffer):
        MAX_NUM = 5
        if len(buffer[0])>0:
            MAX_NUM = buffer[0][0]["MAX_NUM"]
        list_split = split_with_time(buffer[0],"page_time_stamp")
        list_group = []
        for _split in list_split:
            flag = True
            keys = ["set_limit_column1","set_limit_column2","set_limit_column3","set_limit_column4"]
            dict_set = {}
            for _key in keys:
                dict_set[_key] = set()
            if len(_split)>MAX_NUM:
                flag = False
            else:
                for _key in keys:
                    logging.info(_key+str(getSet(_split,_key)))
                    if len(getSet(_split,_key))>1:
                        flag = False
                        break

            MAX_CONTAIN_COLUMN = None
            #判断组内每条公告是否包含
            if flag:
                for _d in _split:
                    contain_column = _d["contain_column"]
                    if contain_column is not None and contain_column !="":
                        if MAX_CONTAIN_COLUMN is None:
                            MAX_CONTAIN_COLUMN = contain_column
                        else:
                            if len(MAX_CONTAIN_COLUMN)<len(contain_column):
                                if contain_column.find(MAX_CONTAIN_COLUMN)==-1:
                                    flag = False
                                    break
                                MAX_CONTAIN_COLUMN = contain_column
                            else:
                                if MAX_CONTAIN_COLUMN.find(contain_column)==-1:
                                    flag = False
                                    break
            if len(getSet(_split,"greater_column"))==1:
                flag = False
                break
            if flag:
                _set_docid = set()
                for item in _split:
                    _set_docid.add(item["docid"])
                if len(_set_docid)>1:
                    list_group.append(list(_set_docid))
        return json.dumps(list_group)

def getDiffIndex(list_dict,key):
    _set = set()
    for _i in range(len(list_dict)):
        item = list_dict[_i]
        if key in item:
            if item[key]!='' and item[key] is not None:
                if re.search("^\d[\d\.]*$",item[key]) is not None:
                    _set.add(str(float(item[key])))
                else:
                    _set.add(str(item[key]))
        if len(_set)>1:
            return _i
    return len(list_dict)

@annotate('bigint,bigint,string,string,string,string,string,string,string,bigint->string')
class f_remege_limit_num_contain(BaseUDAF):
    '''
    项目编号、中标单位、len(项目编号)>7、中标单位<> ""、合并后非空招标单位数<2、合并后同公告类型非空金额相同
    '''
    def __init__(self):
        import logging
        import json,re
        global json,logging,re
        logging.basicConfig(level = logging.INFO,format = '%(asctime)s - %(name)s - %(levelname)s - %(message)s')

    def new_buffer(self):
        return [list()]

    def iterate(self, buffer,docid,page_time_stamp,set_limit_column1,set_limit_column2,set_limit_column3,set_limit_column4,contain_column1,contain_column2,notLike_column,confidence):
        buffer[0].append({"docid":docid,"page_time_stamp":page_time_stamp,"set_limit_column1":set_limit_column1,
                          "set_limit_column2":set_limit_column2,"set_limit_column3":set_limit_column3,"set_limit_column4":set_limit_column4,
                          "contain_column1":contain_column1,"contain_column2":contain_column2,"notLike_column":notLike_column,"confidence":confidence})

    def merge(self, buffer, pbuffer):
        buffer[0].extend(pbuffer[0])

    def getNotLikeSet(self,_dict,column_name):
        column_value = _dict.get(column_name,None)
        _set = set()
        if column_value is not None:
            for _i in range(1,len(column_value)):
                _set.add(column_value[_i-1:_i+1])
        _dict["notLike_set"] = _set

    def getSimilarity(self,_set1,_set2):
        _sum = max([1,min([len(_set1),len(_set2)])])
        return len(_set1&_set2)/_sum

    def terminate(self, buffer):
        list_group = []
        the_group = buffer[0]

        SIM_PROB = 0.6
        for _d in the_group:
            self.getNotLikeSet(_d,"notLike_column")

        #判断多个值与否
        keys = ["set_limit_column1","set_limit_column2","set_limit_column3","set_limit_column4"]
        re_merge = False
        for _key in keys:
            if len(getSet(the_group,_key))>1:
                re_merge = True
                break
        #判断是否相似而不相同
        re_merge_sim = False
        for _i1 in range(0,len(the_group)):
            for _j1 in range(_i1+1,len(the_group)):
                _set1 = the_group[_i1]["notLike_set"]
                _set2 = the_group[_j1]["notLike_set"]
                _sim = self.getSimilarity(_set1,_set2)
                if _sim>SIM_PROB and _sim<1:
                    re_merge_sim = True
                    break
        contain_keys = ["contain_column1","contain_column2"]

        logging.info(the_group)
        if re_merge or re_merge_sim:
            the_group.sort(key=lambda x:x["confidence"],reverse=True)
            the_group.sort(key=lambda x:x["page_time_stamp"])
            #重新成组
            dict_docid_doc = {}
            for _doc in the_group:
                dict_docid_doc[_doc["docid"]] = _doc
            for _doc in the_group:
                merge_flag = False
                for _index in range(len(list_group)):
                    _g = list_group[_index]
                    hit_count = 0
                    dict_temp = dict()
                    #多个值的异常
                    if re_merge:
                        for _c_key in contain_keys:
                            dict_temp[_c_key] = _g[_c_key]
                            if _g[_c_key] is not None and _doc[_c_key] is not None:
                                if len(_g[_c_key])>len(_doc[_c_key]):
                                    if str(_g[_c_key]).find(str(_doc[_c_key]))>=0:
                                        dict_temp[_c_key] = _g[_c_key]
                                        hit_count += 1
                                else:
                                    if str(_doc[_c_key]).find(str(_g[_c_key]))>=0:
                                        dict_temp[_c_key] = _doc[_c_key]
                                        _g[_c_key] = _doc[_c_key]
                                        hit_count += 1
                    else:
                        hit_count = 1
                    # if hit_count==len(contain_keys):
                    if hit_count>0:
                        _flag_sim = False
                        #相似而不相同的异常
                        if re_merge_sim:
                            for _docid in _g["docid"]:
                                tmp_d = dict_docid_doc[_docid]
                                _sim = self.getSimilarity(tmp_d["notLike_set"],_doc["notLike_set"])
                                if _sim>SIM_PROB and _sim<1:
                                    _flag_sim = True
                        if not _flag_sim:
                            for _c_key in dict_temp.keys():
                                _g[_c_key] = dict_temp[_c_key]
                            _g["docid"].append(_doc["docid"])
                            merge_flag = True
                            break
                if not merge_flag:
                    _dict = dict()
                    _dict["docid"] = [_doc["docid"]]
                    for _c_key in contain_keys:
                        _dict[_c_key] = _doc[_c_key]
                    list_group.append(_dict)

            final_group = []
            #判断是否符合一个值
            for _group in list_group:
                _split = []
                for _docid in _group["docid"]:
                    _split.append(dict_docid_doc[_docid])

                #通过置信度排序，尽可能保留组
                _split.sort(key=lambda x:x["confidence"],reverse=True)
                #置信度
                list_key_index = []
                for _k in keys:
                    list_key_index.append(getDiffIndex(_split,_k))

                _index = min(list_key_index)


                final_group.append([_c["docid"] for _c in _split[:_index]])
                for _c in _split[_index:]:
                    final_group.append([_c["docid"]])


                #若是找到两个以上，则全部单独成组，否则成一组
                # _flag = True
                # for _key in keys:
                #     if len(getSet(_split,_key))>1:
                #         _flag = False
                #         break
                # if not _flag:
                #     for _docid in _group["docid"]:
                #         final_group.append([_docid])
                # else:
                #     final_group.append(list(set(_group["docid"])))
        else:
            final_group = [list(set([item["docid"] for item in the_group]))]

        return json.dumps(final_group)

@annotate('string -> string')
class f_get_remerge_group(BaseUDTF):
    '''
    将多个组拆解成多条记录
    '''

    def __init__(self):
        import logging
        import json
        global json,logging
        logging.basicConfig(level = logging.INFO,format = '%(asctime)s - %(name)s - %(levelname)s - %(message)s')

    def process(self,json_remerge):
        if json_remerge is not None:
            list_group = json.loads(json_remerge)
            for _group in list_group:
                l_g = list(set(_group))
                l_g.sort(key=lambda x:x)
                list_docid = [str(_docid) for _docid in l_g]
                self.forward(",".join(list_docid))

@annotate('bigint,bigint,string->string')
class f_merge_probability(BaseUDAF):
    '''
    合并组为一条记录
    '''
    def __init__(self):
        import json
        global json

    def new_buffer(self):
        return [[]]

    def iterate(self, buffer,docid,page_time_stamp,_type):
        buffer[0].append({"docid":docid,"page_time_stamp":page_time_stamp,"type":_type})

    def merge(self, buffer, pbuffer):
        buffer[0].extend(pbuffer[0])

    def terminate(self, buffer):
        list_dict = buffer[0]
        list_dict = list_dict[:10000]
        list_group = split_with_time(list_dict,sort_key="page_time_stamp",timedelta=86400*120)

        return json.dumps(list_group)

@annotate('string -> bigint,bigint,bigint,bigint,string')
class f_split_merge_probability(BaseUDTF):

    def __init__(self):
        import logging
        import json
        global logging,json
        logging.basicConfig(level=logging.INFO,format='%(asctime)s - %(name)s - %(levelname)s - %(message)s')

    def process(self,list_group_str):
        logging.info("0")
        logging.info(list_group_str)
        if list_group_str is not None:
            logging.info("1")
            try:
                list_group = json.loads(list_group_str)
                logging.info("2")
                for _group in list_group:
                    if len(_group)>0:
                        _type = _group[0].get("type","")
                    logging.info("3%d"%len(list_group))
                    # _group.sort(key=lambda x:x["page_time_stamp"])
                    _len = min(100,len(_group))
                    for _index_i in range(_len):
                        _count = 0
                        for _index_j in range(_index_i+1,_len):
                            if abs(_group[_index_j]["page_time_stamp"]-_group[_index_i]["page_time_stamp"])>86400*120:
                                break
                                _count += 1
                            _docid1 = _group[_index_i]["docid"]
                            _docid2 = _group[_index_j]["docid"]
                            if _docid1<_docid2:
                                self.forward(_docid1,_docid2,1,_len,_type)
                            else:
                                self.forward(_docid2,_docid1,1,_len,_type)
            except Exception as e:
                logging(str(e))


@annotate('bigint,bigint,string->string')
class f_merge_groupPairs(BaseUDAF):
    '''
    合并组为一条记录
    '''
    def __init__(self):
        import json
        global json

    def new_buffer(self):
        return [[]]

    def iterate(self, buffer,is_exists,counts,_type):
        buffer[0].append({"is_exists":is_exists,"counts":counts,"_type":_type})

    def merge(self, buffer, pbuffer):
        buffer[0].extend(pbuffer[0])

    def terminate(self, buffer):
        list_dict = buffer[0]
        list_dict = list_dict[:10000]

        return json.dumps(list_dict)

@annotate("string -> bigint,bigint,bigint")
class f_merge_getLabel(BaseUDTF):

    def __init__(self):
        import logging
        import json
        global logging,json

    def process(self,str_docids):
        if str_docids is not None:
            list_docids = [int(i) for i in str_docids.split(",")]
            list_docids.sort(key=lambda x:x)
            _len = min(100,len(list_docids))
            for index_i in range(_len):
                docid_less = list_docids[index_i]

                for index_j in range(index_i+1,_len):
                    docid_greater = list_docids[index_j]
                    self.forward(docid_less,docid_greater,1)

def getSimilarityOfString(str1,str2):
    _set1 = set()
    _set2 = set()
    if str1 is not None:
        for i in range(1,len(str1)):
            _set1.add(str1[i-1:i+1])
    if str2 is not None:
        for i in range(1,len(str2)):
            _set2.add(str2[i-1:i+1])
    _len = max(1,min(len(_set1),len(_set2)))
    return len(_set1&_set2)/_len

def check_columns(tenderee_less,tenderee_greater,
                  agency_less,agency_greater,project_code_less,project_code_greater,project_name_less,project_name_greater,
                  win_tenderer_less,win_tenderer_greater,win_bid_price_less,win_bid_price_greater,
                  bidding_budget_less,bidding_budget_greater,doctitle_refine_less,doctitle_refine_greater):
    flag = True
    _set_tenderee = set()
    if tenderee_less is not None and tenderee_less!="":
        _set_tenderee.add(tenderee_less)
    if tenderee_greater is not None and tenderee_greater!="":
        _set_tenderee.add(tenderee_greater)
    if len(_set_tenderee)>1:
        return False
    code_sim = getSimilarityOfString(project_code_less,project_code_greater)
    if code_sim>0.6 and code_sim<1:
        return False
    _set_win_tenderer = set()
    if win_tenderer_less is not None and tenderee_less!="":
        _set_win_tenderer.add(win_tenderer_less)
    if win_tenderer_greater is not None and win_tenderer_greater!="":
        _set_win_tenderer.add(win_tenderer_greater)
    if len(_set_win_tenderer)>1:
        return False
    _set_win_bid_price = set()
    if win_bid_price_less is not None and win_bid_price_less!="":
        _set_win_bid_price.add(win_bid_price_less)
    if win_bid_price_greater is not None and win_bid_price_greater!="":
        _set_win_bid_price.add(win_bid_price_greater)
    if len(_set_win_bid_price)>1:
        return False
    _set_bidding_budget = set()
    if bidding_budget_less is not None and bidding_budget_less!="":
        _set_bidding_budget.add(bidding_budget_less)
    if bidding_budget_greater is not None and bidding_budget_greater!="":
        _set_bidding_budget.add(bidding_budget_greater)
    if len(_set_bidding_budget)>1:
        return False
    return True

def getSimLevel(str1,str2):
    str1_null = False
    str2_null = False
    _v = 0
    if str1 is None or str1=="":
        str1_null = True
    if str2 is None or str2=="":
        str2_null = True
    if str1_null and str2_null:
        _v = 2
    elif str1_null and not str2_null:
        _v = 4
    elif not str1_null and str2_null:
        _v = 6
    elif not str1_null and not str2_null:
        if str1==str2:
            _v = 10
        else:
            _v = 0
    return _v

import math
def featurnCount(_count,max_count=100):
    return max(0,min(1,_count))*(1/math.sqrt(max(1,_count-1)))

@annotate("string,string,string,string,string,string,string,string,string,string,string,string,string,string,string,string,string->string")
class f_merge_featureMatrix(BaseUDTF):

    def __init__(self):
        import logging
        import json
        global logging,json

    def process(self,json_context,tenderee_less,tenderee_greater,
                                    agency_less,agency_greater,project_code_less,project_code_greater,project_name_less,project_name_greater,
                                    win_tenderer_less,win_tenderer_greater,win_bid_price_less,win_bid_price_greater,
                                    bidding_budget_less,bidding_budget_greater,doctitle_refine_less,doctitle_refine_greater):
        # if not check_columns(tenderee_less,tenderee_greater,
        #                      agency_less,agency_greater,project_code_less,project_code_greater,project_name_less,project_name_greater,
        #                      win_tenderer_less,win_tenderer_greater,win_bid_price_less,win_bid_price_greater,
        #                      bidding_budget_less,bidding_budget_greater,doctitle_refine_less,doctitle_refine_greater)
        #     return

        _context = json.loads(json_context)
        dict_context = {}
        for item in _context:
            dict_context[item["_type"]] = [item["is_exists"],item["counts"]]
        context_key = ["tenderee","agency","project_code","project_name","win_tenderer","win_bid_price","bidding_budget","doctitle_refine"]
        list_matrix = []
        for index_i in range(len(context_key)):
            for index_j in range(index_i+1,len(context_key)):
                _key = "%s&%s"%(context_key[index_i],context_key[index_j])
                _v = featurnCount(dict_context.get(_key,[0,0])[1])
                list_matrix.append(_v)
        context3_key = ["tenderee","agency","win_tenderer","win_bid_price","bidding_budget"]
        for index_i in range(len(context3_key)):
            for index_j in range(index_i+1,len(context3_key)):
                for index_k in range(index_j+1,len(context3_key)):
                    _key = "%s&%s&%s"%(context3_key[index_i],context3_key[index_j],context3_key[index_k])
                    _v = featurnCount(dict_context.get(_key,[0,0])[1])
                    list_matrix.append(_v)
        list_matrix.append(getSimLevel(tenderee_less,tenderee_greater)/10)
        list_matrix.append(getSimLevel(agency_less,agency_greater)/10)
        list_matrix.append(getSimilarityOfString(project_code_less,project_code_greater))
        list_matrix.append(getSimilarityOfString(project_name_less,project_name_greater))
        list_matrix.append(getSimLevel(win_tenderer_less,win_tenderer_greater)/10)
        list_matrix.append(getSimLevel(win_bid_price_less,win_bid_price_greater)/10)
        list_matrix.append(getSimLevel(bidding_budget_less,bidding_budget_greater)/10)
        list_matrix.append(getSimilarityOfString(doctitle_refine_less,doctitle_refine_greater))
        self.forward(json.dumps(list_matrix))







@annotate('string -> bigint,bigint')
class f_check_remerge(BaseUDTF):
    '''
    将多个组拆解成多条记录
    '''

    def __init__(self):
        import logging
        import json
        global json,logging
        logging.basicConfig(level = logging.INFO,format = '%(asctime)s - %(name)s - %(levelname)s - %(message)s')

    def process(self,json_remerge):
        if json_remerge is not None:
            list_group = json.loads(json_remerge)
            for _group in list_group:
                for _docid in _group:
                    self.forward(_group[-1],_docid)

def getConfidence(rule_id):
    if rule_id >=1 and rule_id <=20:
        return 30
    elif rule_id>=31 and rule_id<=50:
        return 20
    else:
        return 10

@annotate('string,bigint -> bigint,bigint,bigint')
class f_arrange_group_single(BaseUDTF):
    '''
    将多个组拆解成多条记录
    '''

    def __init__(self):
        import logging
        import json
        global json,logging
        logging.basicConfig(level = logging.INFO,format = '%(asctime)s - %(name)s - %(levelname)s - %(message)s')

    def process(self,json_set_docid,rule_id):
        if json_set_docid is not None:
            list_group = json.loads(json_set_docid)
            for _group in list_group:
                for index_i in range(len(_group)):
                    for index_j in range(len(_group)):
                        # if index_i!=index_j and _group[index_i]!=_group[index_j]:
                        if index_i!=index_j:
                            self.forward(_group[index_i],_group[index_j],getConfidence(rule_id))

@annotate('bigint,bigint->string')
class f_get_merge_docids(BaseUDAF):
    '''
    合并组为一条记录
    '''
    def __init__(self):
        import json
        global json

    def new_buffer(self):
        return [set()]

    def iterate(self, buffer,docid1,docid2):
        buffer[0].add(docid1)
        buffer[0].add(docid2)

    def merge(self, buffer, pbuffer):
        buffer[0] |= pbuffer[0]

    def terminate(self, buffer):
        set_docid = buffer[0]
        list_docid = list(set_docid)
        list_docid.sort(key=lambda x:x)
        list_docid_str = []
        for _docid in list_docid:
            list_docid_str.append(str(_docid))
        return ",".join(list_docid_str)