from xml.etree import ElementTree import re from enum import Enum from collections import defaultdict import sys import logging import argparse import pickle import time import subprocess import concurrent.futures import tempfile from msd_translate import MSD_TRANSLATE MAX_NUM_COMPONENTS = 5 CODES = { "Noun": "N", "Verb": "V", "Adjective": "A", "Adverb": "R", "Pronoun": "P", "Numeral": "M", "Preposition": "S", "Conjunction": "C", "Particle": "Q", "Interjection": "I", "Abbreviation": "Y", "Residual": "X", 'common': 'c', 'proper': 'p', 'masculine': 'm', 'feminine': 'f', 'neuter': 'n', "singular": "s", "dual": "d", "plural": "p", "nominative": "n", "genitive": "g", "dative": "d", "accusative": "a", "locative": "l", "instrumental": "i", "no": "n", "yes": "y", "main": "m", "auxiliary": "a", "perfective": "e", "progressive": "p", "biaspectual": "b", "infinitive": "n", "supine": "u", "participle": "p", "present": "r", "future": "f", "conditional": "c", "imperative": "m", "first": "1", "second": "2", "third": "3", "general": "g", "possessive": "s", "positive": "p", "comparative": "c", "superlative": "s", "personal": "p", "demonstrative": "d", "relative": "r", "reflexive": "x", "interrogative": "q", "indefinite": "i", "negative": "z", "bound": "b", "digit": "d", "roman": "r", "letter": "l", "cardinal": "c", "ordinal": "o", "pronominal": "p", "special": "s", "coordinating": "c", "subordinating": "s", "foreign": "f", "typo": "t", "program": "p", } TAGSET = { "N": ['type', 'gender', 'number', 'case', 'animate'], "V": ['type', 'aspect', 'vform', 'person', 'number', 'gender', 'negative'], "A": ['type', 'degree', 'gender', 'number', 'case', 'definiteness'], "R": ['type', 'degree'], "P": ['type', 'person', 'gender', 'number', 'case', 'owner_number', 'owned_gender', 'clitic'], "M": ['form', 'type', 'gender', 'number', 'case', 'definiteness'], "S": ['case'], "C": ['type'], "Q": [], "I": [], "Y": [], "X": ['type'] } CATEGORY_BASES = { "N": ['.'] * 5, "V": ['.'] * 7, "A": ['.'] * 6, "R": ['.'] * 2, "P": ['.'] * 6, "M": ['.'] * 6, "S": ['.'] * 1, "C": ['.'] * 1, "Q": [], "I": [], "Y": [], "X": ['.'] * 1 } class RestrictionType(Enum): Morphology = 0 Lexis = 1 MatchAll = 2 class Rendition(Enum): Lemma = 0 WordForm = 1 Lexis = 2 Unknown = 3 class WordFormSelection(Enum): All = 0 Msd = 1 Agreement = 2 Any = 3 class Order(Enum): FromTo = 0 ToFrom = 1 Any = 2 @staticmethod def new(order): if order is not None: if order == "to-from": return Order.ToFrom elif order == "from-to": return Order.FromTo else: raise NotImplementedError("What kind of ordering is: {}".format(order)) else: return Order.Any def match(self, from_w, to_w): if self is Order.Any: return True fi = from_w.int_id ti = to_w.int_id if self is Order.FromTo: return fi < ti elif self is Order.ToFrom: return ti < fi else: raise NotImplementedError("Should not be here: Order match") class ComponentRendition: def __init__(self): self.more = None self.rendition = Rendition.Unknown def _set_rendition(self, r): assert(self.rendition is Rendition.Unknown) self.rendition = r def _set_more(self, m): self.more = m def add_feature(self, feature): if 'rendition' in feature: if feature['rendition'] == "lemma": self._set_rendition(Rendition.Lemma) elif feature['rendition'] == "word_form": self._set_rendition(Rendition.WordForm) self._set_more((WordFormSelection.Any, None)) elif feature['rendition'] == "lexis": self._set_rendition(Rendition.Lexis) self._set_more(feature['string']) else: raise NotImplementedError("Representation rendition: {}".format(feature)) elif 'selection' in feature: if feature['selection'] == "msd": selectors = {k: v for k, v in feature.items() if k != 'selection'} self._set_more((WordFormSelection.Msd, selectors)) elif feature['selection'] == "all": self._set_more((WordFormSelection.All, None)) elif feature['selection'] == 'agreement': assert(feature['head'][:4] == 'cid_') assert(feature['msd'] is not None) self._set_more((WordFormSelection.Agreement, (feature['head'][4:], feature['msd'].split('+')))) else: raise NotImplementedError("Representation selection: {}".format(feature)) else: return None def isit(self, rendition): return self.rendition is rendition @staticmethod def set_representations(matches, structure, word_renderer): representations = { c.idx: [[], None] if c.representation.isit(Rendition.WordForm) else [True, ""] for c in structure.components } representations_to_check = [] word_component_id = {} # doprint = structure.id == '1' and matches[0]['1'].text.startswith('evrop') and matches[0]['2'].text.startswith('prv') doprint = False def render_all(lst): return "/".join([w.text for w in set(lst)]) def render_form(lst): sorted_lst = sorted(set(lst), key=lst.count) for word in sorted_lst: othw = are_agreements_ok(word, representations_to_check) if othw is not None: matches.representations[word_component_id[othw.id]] = othw.most_frequent_text(word_renderer) matches.representations[word_component_id[word.id]] = word.most_frequent_text(word_renderer) return def are_agreements_ok(word, words_to_try): for w_id, other_word, agreements in words_to_try: if check_agreement(word, other_word, agreements): if doprint: print("GOOD :)") return other_word def check_msd(word, selectors): for key, value in selectors.items(): t = word.msd[0] v = TAGSET[t].index(key.lower()) f1 = word.msd[v + 1] f2 = CODES[value] if '-' not in [f1, f2] and f1 != f2: return False return True def check_agreement(w1, w2, agreements): if doprint: print("CHECK", w1.text, w1, w2.text, w2) for agr_case in agreements: t1 = w1.msd[0] v1 = TAGSET[t1].index(agr_case) assert(v1 >= 0) # if none specified: nedolocnik, always agrees if v1 + 1 >= len(w1.msd): continue # first is uppercase, not in TAGSET m1 = w1.msd[v1 + 1] # REPEAT (not DRY!) t2 = w2.msd[0] v2 = TAGSET[t2].index(agr_case) assert(v2 >= 0) if v2 + 1 >= len(w2.msd): continue m2 = w2.msd[v2 + 1] # match! if '-' not in [m1, m2] and m1 != m2: return False return True for words in matches.matches: # first pass, check everything but agreements for w_id, w in words.items(): component = structure.get_component(w_id) rep = component.representation word_component_id[w.id] = w_id if rep.isit(Rendition.Lemma): representations[w_id][0] = False representations[w_id][1] = w.lemma elif rep.isit(Rendition.Lexis): representations[w_id][0] = False representations[w_id][1] = rep.more elif rep.isit(Rendition.Unknown): representations[w_id][0] = False representations[w_id][1] = "" # it HAS to be word_form now else: assert(rep.isit(Rendition.WordForm)) wf_type, more = rep.more if wf_type is WordFormSelection.All: add = True func = render_all elif wf_type is WordFormSelection.Msd: add = check_msd(w, more) func = render_form elif wf_type is WordFormSelection.Any: add = True func = render_form else: assert(wf_type is WordFormSelection.Agreement) other_w, agreements = more representations_to_check.append((other_w, w, agreements)) add = True func = lambda x: None if add: representations[w_id][0].append(w) representations[w_id][1] = func if doprint: print(len(matches), len(representations_to_check)) # for w1i, w2i, agreements in representations_to_check: # w1, w2 = words[w1i], words[w2i] # if doprint: # print("? ", w1.msd, w2.msd, end="") # if w2i not in bad_words: # # if check_agreement(w1, w2, agreements): # representations[w1i][0].append(w1.text) # if doprint: # print(" :)") # elif doprint: # print(" :(") # elif doprint: # print(" :((") # just need to set representation to first group, # but in correct order, agreements last! representation_sorted_words = [] for w_id, w in matches.matches[0].items(): rep = component.representation if rep.isit(Rendition.WordForm) and rep.more[0] is WordFormSelection.Agreement: representation_sorted_words.append((w_id, w)) else: representation_sorted_words.insert(0, (w_id, w)) for w_id, w in representation_sorted_words: data = representations[w_id] if doprint: print([(r.text, r.lemma, r.msd) for r in data[0]]) if type(data[1]) is str: matches.representations[w_id] = None if data[0] else data[1] elif len(data[0]) == 0: matches.representations[w_id] = None else: data[1](data[0]) if doprint: print(matches.representations) print('--') def __str__(self): return str(self.rendition) class ComponentStatus(Enum): Optional = 0 Required = 1 Forbidden = 2 def __str__(self): if self == ComponentStatus.Optional: return "?" elif self == ComponentStatus.Required: return "!" else: #Forbidden return "X" def get_level(restriction): for feature in restriction: if "level" in feature.keys(): lvl = feature.get("level") else: continue raise RuntimeError("Unreachable!") def build_morphology_regex(restriction): restr_dict = {} for feature in restriction: feature_dict = dict(feature.items()) match_type = True if "filter" in feature_dict: assert(feature_dict['filter'] == "negative") match_type = False del feature_dict['filter'] assert(len(feature_dict) == 1) key, value = next(iter(feature_dict.items())) restr_dict[key] = (value, match_type) assert('POS' in restr_dict) category = restr_dict['POS'][0].capitalize() cat_code = CODES[category] rgx = [cat_code] + CATEGORY_BASES[cat_code] del restr_dict['POS'] min_msd_length = 1 for attribute, (value, typ) in restr_dict.items(): index = TAGSET[cat_code].index(attribute.lower()) assert(index >= 0) if '|' in value: match = "".join(CODES[val] for val in value.split('|')) else: match = CODES[value] match = "[{}{}]".format("" if typ else "^", match) rgx[index + 1] = match if typ: min_msd_length = max(index + 1, min_msd_length) def matcher(text): if len(text) <= min_msd_length: return False for c, r in zip(text, rgx): if not re.match(r, c): return False return True return " ".join(rgx), matcher def build_lexis_regex(restriction): restr_dict = {} for feature in restriction: restr_dict.update(feature.items()) assert("lemma" in restr_dict) match_list = restr_dict['lemma'].split('|') return match_list, lambda text: text in match_list class Restriction: def __init__(self, restriction_tag): if restriction_tag is None: self.type = RestrictionType.MatchAll self.matcher = None self.present = None return restriction_type = restriction_tag.get('type') if restriction_type == "morphology": self.type = RestrictionType.Morphology self.present, self.matcher = build_morphology_regex(list(restriction_tag)) elif restriction_type == "lexis": self.type = RestrictionType.Lexis self.present, self.matcher = build_lexis_regex(list(restriction_tag)) else: raise NotImplementedError() def match(self, word): if self.type == RestrictionType.Morphology: match_to = word.msd elif self.type == RestrictionType.Lexis: match_to = word.lemma elif self.type == RestrictionType.MatchAll: return True else: raise RuntimeError("Unreachable!") return self.matcher(match_to) def __str__(self): return "({:s} {})".format(str(self.type).split('.')[1], self.present) def __repr__(self): return str(self) class Component: def __init__(self, info): idx = info['cid'] name = info['name'] if 'name' in info else None if 'status' not in info: status = ComponentStatus.Required elif info['status'] == 'forbidden': status = ComponentStatus.Forbidden elif info['status'] == 'obligatory': status = ComponentStatus.Required elif info['status'] == 'optional': status = ComponentStatus.Optional else: raise NotImplementedError("strange status: {}".format(info['status'])) self.status = status self.name = name self.idx = idx self.restriction = None self.next_element = [] self.representation = ComponentRendition() self.selection = {} self.iter_ctr = 0 def add_next(self, next_component, link_label, order): self.next_element.append((next_component, link_label, Order.new(order))) def set_restriction(self, restrictions_tag): if restrictions_tag is None: self.restriction = Restriction(None) elif restrictions_tag.tag == "restriction": self.restriction = Restriction(restrictions_tag) elif restrictions_tag.tag == "restriction_or": self.restriction = [Restriction(el) for el in restrictions_tag] else: raise RuntimeError("Unreachable") def set_representation(self, representation): for feature in representation: self.representation.add_feature(feature.attrib) def find_next(self, deps, comps, restrs, reprs): to_ret = [] for d in deps: if d[0] == self.idx: _, idx, dep_label, order = d next_component = Component(comps[idx]) next_component.set_restriction(restrs[idx]) next_component.set_representation(reprs[idx]) to_ret.append(next_component) self.add_next(next_component, dep_label, order) others = next_component.find_next(deps, comps, restrs, reprs) to_ret.extend(others) return to_ret def name_str(self): return "_" if self.name is None else self.name def __str__(self): n = self.name_str() return "{:s}) {:7s}:{} [{}] :{}".format( self.idx, n, self.status, self.restriction, self.representation) def tree(self): el = [] for next, link, order in self.next_element: s = "{:3} -- {:5} --> {:3}".format(self.idx, link, next.idx) if order != Order.Any: s += " " + str(order)[6:] el.append(s) el.extend(next.tree()) return el def __repr__(self): return str(self) def match(self, word): m1 = self._match_self(word) if m1 is None: return None mn = self._match_next(word) if mn is None: return None to_ret = [m1] for cmatch in mn: # if good match but nothing to add, just continue if len(cmatch) == 0: continue # if more than one match found for particular component elif len(cmatch) > 1: logging.debug("MULTIPLE: {}, {}".format(self.idx, cmatch)) # if more than one match in multiple components, NOPE! if len(to_ret) > 1: logging.warning("Strange multiple match: {}".format( str([w.id for w in cmatch[0].values()]))) for tr in to_ret: tr.update(cmatch[0]) continue # yeah, so we have found more than one match, => # more than one element in to_ret to_ret = [{**dict(to_ret[0]), **m} for m in cmatch] else: for tr in to_ret: tr.update(cmatch[0]) logging.debug("MA: {}".format(str(to_ret))) return to_ret def _match_self(self, word): matched = None # matching if type(self.restriction) is list: for restr in self.restriction: matched = restr.match(word) if matched: # match either break else: matched = self.restriction.match(word) logging.debug("SELF MATCH({}: {} -> {}".format(self.idx, word.text, matched)) # recurse to next if not matched: return None else: return {self.idx: word} def _match_next(self, word): # matches for every component in links from this component to_ret = [] # need to get all links that match for next, link, order in self.next_element: next_links = word.get_links(link) logging.debug("FIND LINKS FOR: {} -> {}: #{}".format(self.idx, next.idx, len(next_links))) to_ret.append([]) # good flag good = next.status != ComponentStatus.Required for next_word in next_links: logging.debug("link: {}: {} -> {}".format(link, word.id, next_word.id)) if not order.match(word, next_word): continue match = next.match(next_word) if match is not None: # special treatement for forbidden if next.status == ComponentStatus.Forbidden: good = False break else: assert(type(match) is list) to_ret[-1].extend(match) good = True # if none matched, nothing found! if not good: logging.debug("BAD") return None return to_ret class SyntacticStructure: def __init__(self): self.id = None self.lbs = None self.components = [] @staticmethod def from_xml(xml): st = SyntacticStructure() st.id = xml.get('id') st.lbs = xml.get('LBS') assert(len(list(xml)) == 1) system = next(iter(xml)) assert(system.get('type') == 'JOS') components, dependencies, definitions = list(system) deps = [ (dep.get('from'), dep.get('to'), dep.get('label'), dep.get('order')) for dep in dependencies ] comps = { comp.get('cid'): dict(comp.items()) for comp in components } restrs, forms = {}, {} for comp in definitions: n = comp.get('cid') restrs[n] = None forms[n] = [] for el in comp: if el.tag.startswith("restriction"): assert(restrs[n] is None) restrs[n] = el elif el.tag.startswith("representation"): st.add_representation(n, el, forms) else: raise NotImplementedError("Unknown definition: {} in structure {}".format(el.tag, st.id)) fake_root_component = Component({'cid': '#', 'type': 'other'}) st.components = fake_root_component.find_next(deps, comps, restrs, forms) return st def add_representation(self, n, rep_el, forms): if rep_el.tag == "representation_and": rep_el = rep_el[0] logging.warning("Only using first reprentation in representation_and in structure {}".format(self.id)) assert(rep_el.tag == "representation") for el in rep_el: assert(el.tag == "feature") if 'rendition' in el.attrib: forms[n].append(el) elif 'selection' in el.attrib: forms[n].append(el) else: logging.warning("Strange representation feature in structure {}. Skipping" .format(self.id)) continue def __str__(self): comp_str = "\n".join(str(comp) for comp in self.components) links_str = "\n".join(self.components[0].tree()) return "{} LBS {}\nCOMPONENTS\n{}\n\nLINKS\n{}\n{}".format( self.id, self.lbs, comp_str, links_str, "-" * 40) def get_component(self, idx): for c in self.components: if c.idx == idx: return c raise RuntimeError("Unknown component id: {}".format(idx)) def check_agreements(self, match): for agr in self.agreements: w1 = match[agr['n1']] w2 = match[agr['n2']] for agr_case in agr['match']: t1 = w1.msd[0] v1 = TAGSET[t1].index(agr_case) assert(v1 >= 0) # if none specified: nedolocnik, always agrees if v1 + 1 >= len(w1.msd): continue # first is uppercase, not in TAGSET m1 = w1.msd[v1 + 1] # REPEAT (not DRY!) t2 = w2.msd[0] v2 = TAGSET[t2].index(agr_case) assert(v2 >= 0) if v2 + 1 >= len(w2.msd): continue m2 = w2.msd[v2 + 1] # match! if '-' not in [m1, m2] and m1 != m2: return False return True def check_form(self, match): for midx, w in match.items(): c = self.get_component(midx) for key, value in c.selection.items(): t = w.msd[0] v = TAGSET[t].index(key.lower()) f1 = w.msd[v + 1] f2 = CODES[value] if '-' not in [f1, f2] and f1 != f2: return False return True def match(self, word): matches = self.components[0].match(word) return [] if matches is None else matches # for m in matches: # to_ret.append((m, self.check_agreements(m))) def build_structures(filename): structures = [] with open(filename, 'r') as fp: et = ElementTree.XML(fp.read()) for structure in et.iter('syntactic_structure'): to_append = SyntacticStructure.from_xml(structure) if to_append is None: continue structures.append(to_append) return structures def get_msd(comp): d = dict(comp.items()) if 'msd' in d: return d['msd'] elif 'ana' in d: return d['ana'][4:] else: logging.error(d, file=sys.stderr) raise NotImplementedError("MSD?") class Word: def __init__(self, xml, do_msd_translate): self.lemma = xml.get('lemma') self.msd = MSD_TRANSLATE[get_msd(xml)] if do_msd_translate else get_msd(xml) self.id = xml.get('id') self.text = xml.text self.links = defaultdict(list) last_num = self.id.split('.')[-1] if last_num[0] not in '0123456789': last_num = last_num[1:] self.int_id = int(last_num) assert(None not in (self.id, self.lemma, self.msd)) @staticmethod def pcWord(pc, do_msd_translate): pc.set('lemma', pc.text) pc.set('msd', "N" if do_msd_translate else "U") return Word(pc, do_msd_translate) def add_link(self, link, to): self.links[link].append(to) def get_links(self, link): if link not in self.links and "|" in link: for l in link.split('|'): self.links[link].extend(self.links[l]) return self.links[link] def most_frequent_text(self, word_renderer): return word_renderer.render(self.lemma, self.msd) class WordMsdRenderer: def __init__(self): self.all_words = [] self.rendered_words = {} def add_words(self, words): self.all_words.extend(words) def generate_renders(self): data = defaultdict(lambda: defaultdict(list)) for w in self.all_words: data[w.lemma][w.msd].append(w.text) for lemma, ld in data.items(): self.rendered_words[lemma] = {} for msd, texts in ld.items(): rep = max(set(texts), key=texts.count) self.rendered_words[lemma][msd] = rep def render(self, lemma, msd): if lemma in self.rendered_words: if msd in self.rendered_words[lemma]: return self.rendered_words[lemma][msd] def is_root_id(id_): return len(id_.split('.')) == 3 def load_files(args): filenames = args.input skip_id_check = args.skip_id_check do_msd_translate = not args.no_msd_translate for n, fname in enumerate(filenames): if args.count_files: status = " :: {} / {}".format(n, len(filenames)) else: status = "" yield load_tei_file(fname, skip_id_check, do_msd_translate, args.pc_tag, status) def load_tei_file(filename, skip_id_check, do_msd_translate, pc_tag, status): logging.info("LOADING FILE: {}{}".format(filename, status)) with open(filename, 'r') as fp: xmlstring = re.sub(' xmlns="[^"]+"', '', fp.read(), count=1) xmlstring = xmlstring.replace(' xml:', ' ') et = ElementTree.XML(xmlstring) words = {} for w in et.iter("w"): words[w.get('id')] = Word(w, do_msd_translate) for pc in et.iter(pc_tag): words[pc.get('id')] = Word.pcWord(pc, do_msd_translate) for l in et.iter("link"): if 'dep' in l.keys(): ana = l.get('afun') lfrom = l.get('from') dest = l.get('dep') else: ana = l.get('ana') if ana[:4] != 'syn:': # dont bother... continue ana = ana[4:] lfrom, dest = l.get('target').replace('#', '').split() if lfrom in words: if not skip_id_check and is_root_id(lfrom): logging.error("NOO: ", lfrom) sys.exit(1) if dest in words: next_word = words[dest] words[lfrom].add_link(ana, next_word) else: logging.error("Unknown id: {}".format(dest)) sys.exit(1) else: # strange errors, just skip... pass return list(words.values()) class Writer: @staticmethod def make_output_writer(args): return Writer(False, args.output, args.multiple_output, int(args.sort_by), args.sort_reversed) @staticmethod def make_all_writer(args): return Writer(True, args.all, False, -1, False) def __init__(self, all, filename, multiple_output, sort_by, sort_reversed): self.all = all self.output_file = filename self.multiple_output = multiple_output self.sort_by = sort_by self.sort_order = sort_reversed def header(self): cols = ["Lemma"] if self.all: cols = ["Token_ID", "Word_form"] + cols + ["Msd"] else: cols.extend(["Representative_form", "RF_scenario"]) assert(len(cols) == self.length()) cols = ["C{}_{}".format(i + 1, thd) for i in range(MAX_NUM_COMPONENTS) for thd in cols] cols = ["Structure_ID"] + cols + ["Colocation_ID"] if not self.all: cols += ["Joint_representative_form", "Frequency"] return cols def length(self): return 4 if self.all else 3 def from_word(self, word, representation): if word is None: return [""] * self.length() elif self.all: return [word.id, word.text, word.lemma, word.msd] else: print("1", word) if representation is None: return [word.lemma, word.lemma, "lemma_fallback"] else: return [word.lemma, representation, "ok"] def sorted_rows(self, rows): if self.sort_by < 0 or len(rows) < 2: return rows if len(rows[0]) <= self.sort_by: logging.warning("Cannot sort by column #{}: Not enough columns!".format(len(rows[0]))) return rows try: int(rows[0][self.sort_by]) key=lambda row: int(row[self.sort_by]) except ValueError: key=lambda row: row[self.sort_by].lower() return sorted(rows, key=key, reverse=self.sort_order) def write_header(self, file_handler): file_handler.write(", ".join(self.header()) + "\n") def write_out_worker(self, file_handler, structure_id, components, colocation_ids): rows = [] for cid, m, freq, rprsnt in colocation_ids.get_matches_for(structure_id, not self.all): to_write = [] representation = "" for idx, _comp in enumerate(components): idx = str(idx + 1) word = m[idx] if idx in m else None print(rprsnt) rep = rprsnt[idx] if idx in rprsnt else None to_write.extend(self.from_word(word, rep)) representation += " " + to_write[-2] # make them equal size to_write.extend([""] * (MAX_NUM_COMPONENTS * self.length() - len(to_write))) to_write = [structure_id] + to_write + [cid] if not self.all: representation = re.sub(' +', ' ', representation) to_write.append(representation.strip()) to_write.append(str(freq)) rows.append(to_write) if len(rows) > 0: rows = self.sorted_rows(rows) file_handler.write("\n".join([", ".join(row) for row in rows]) + "\n") file_handler.flush() def write_out(self, structures, colocation_ids): def fp_close(fp_): if fp_ != sys.stdout: fp_.close() def fp_open(snum=None): if self.output_file is None: return sys.stdout elif snum is None: return open(self.output_file, "w") else: return open("{}.{}".format(self.output_file, snum), "w") if not self.multiple_output: fp = fp_open() self.write_header(fp) for s in structures: if self.multiple_output: fp=fp_open(s.id) self.write_header(fp) self.write_out_worker(fp, s.id, s.components, colocation_ids) if self.multiple_output: fp_close(fp) if not self.multiple_output: fp_close(fp) class StructureMatch: def __init__(self, match_id, structure_id): self.match_id = match_id self.structure_id = structure_id self.matches = [] self.representations = {} def append(self, match): self.matches.append(match) def __len__(self): return len(self.matches) class ColocationIds: def __init__(self): self.data = {} self.min_frequency = args.min_freq def _add_match(self, key, sid, match): if key not in self.data: self.data[key] = StructureMatch(str(len(self.data) + 1), sid) self.data[key].append(match) def get(self, key, n): return self.data[key][n] def add_matches(self, matches): for sid, nms in matches.items(): for nm in nms: self._add_match(nm[1], sid, nm[0]) def get_matches_for(self, structure_id, group): for _cid_tup, sm in self.data.items(): if sm.structure_id != structure_id: continue for words in sm.matches: yield (sm.match_id, words, len(sm), sm.representations) if group: break def set_representations(self, structures, word_renderer): components_dict = {structure.id: structure for structure in structures} for _1, sm in self.data.items(): ComponentRendition.set_representations(sm, components_dict[sm.structure_id], word_renderer) def match_file(words, structures): matches = {s.id: [] for s in structures} for idx, s in enumerate(structures): logging.info("{}/{}: {:7s}".format(idx, len(structures), s.id)) for w in words: mhere = s.match(w) logging.debug(" GOT: {}".format(len(mhere))) for match in mhere: colocation_id = [(idx, w.lemma) for idx, w in match.items()] colocation_id = [s.id] + list(sorted(colocation_id, key=lambda x:x[0])) colocation_id = tuple(colocation_id) matches[s.id].append((match, colocation_id)) return matches def main(input_file, structures_file, args): structures = build_structures(structures_file) for s in structures: logging.debug(str(s)) colocation_ids = ColocationIds() word_renderer = WordMsdRenderer() if args.parallel: num_parallel = int(args.parallel) # make temporary directory to hold temporary files with tempfile.TemporaryDirectory() as tmpdirname: cmd = sys.argv for inpt in args.input: if inpt in cmd: cmd.remove(inpt) # remove "--parallel X" pidx = cmd.index('--parallel') del cmd[pidx] del cmd[pidx] def func(n): cmdn = [sys.executable] + cmd + [args.input[n], "--match-to-file", "{}/{}.p".format(tmpdirname, n)] subprocess.check_call(cmdn) return n # use ThreadPoolExecuter to run subprocesses in parallel using py threads with concurrent.futures.ThreadPoolExecutor(max_workers=num_parallel) as executor: # fancy interface to wait for threads to finish for id_input in executor.map(func, [i for i, _ in enumerate(args.input)]): with open("{}/{}.p".format(tmpdirname, id_input), "rb") as fp: words, matches = pickle.load(fp) colocation_ids.add_matches(matches) word_renderer.add_words(words) else: for words in load_files(args): matches = match_file(words, structures) # just save to temporary file, used for children of a parallel process # MUST NOT have more than one file if args.match_to_file is not None: with open(args.match_to_file, "wb") as fp: pickle.dump((words, matches), fp) return else: colocation_ids.add_matches(matches) word_renderer.add_words(words) # get word renders for lemma/msd word_renderer.generate_renders() # figure out representations! colocation_ids.set_representations(structures, word_renderer) if args.all: Writer.make_all_writer(args).write_out(structures, colocation_ids) Writer.make_output_writer(args).write_out(structures, colocation_ids) logging.debug([(k, len(v)) for k, v in matches.items()]) logging.debug(sum(len(v) for _, v in matches.items())) if __name__ == '__main__': parser = argparse.ArgumentParser(description='Extract structures from a parsed corpus.') parser.add_argument('structures', help='Structures definitions in xml file') parser.add_argument('input', help='input xml file in `ssj500k form`, can list more than one', nargs='+') parser.add_argument('--output', help='Output file (if none given, then output to stdout)') parser.add_argument('--all', help='Additional output file, writes more data') parser.add_argument('--no-msd-translate', help='MSDs are translated from slovene to english by default', action='store_true') parser.add_argument('--skip-id-check', help='Skips checks for ids of and , if they are in correct format', action='store_true') parser.add_argument('--min_freq', help='Minimal frequency in output', type=int, default=0, const=1, nargs='?') parser.add_argument('--verbose', help='Enable verbose output to stderr', choices=["warning", "info", "debug"], default="info", const="info", nargs='?') parser.add_argument('--count-files', help="Count files: more verbose output", action='store_true') parser.add_argument('--multiple-output', help='Generate one output for each syntactic structure', action='store_true') parser.add_argument('--sort-by', help="Sort by a this column (index)", type=int, default=-1) parser.add_argument('--sort-reversed', help="Sort in reversed ored", action='store_true') parser.add_argument('--pc-tag', help='Tag for separators, usually pc or c', default="pc") parser.add_argument('--parallel', help='Run in multiple processes, should speed things up') parser.add_argument('--match-to-file', help='Do not use!') args = parser.parse_args() logging.basicConfig(stream=sys.stderr, level=args.verbose.upper()) start = time.time() main(args.input, args.structures, args) logging.info("TIME: {}".format(time.time() - start))