added adjective handling (appending _ to headwords)

corpusparser/Parser.py

@@ -3,6 +3,7 @@ import re
 import json
 from lxml import etree
 import logging
+import time
 
 logging.basicConfig(level=logging.INFO)
 
@@ -10,17 +11,13 @@ logging.basicConfig(level=logging.INFO)
 # Create an iterator that outputs resulting sentences (python dict format).  
 class Parser():
 
-    def __init__(self, corpus, infiles, logger=None):
-
-        if corpus == "kres":
-            self.kres_folder = Path(infiles[0])
-            self.kres_srl_folder = Path(infiles[1])
-        elif corpus == "ssj":
-            self.ssj_file = Path(infiles[0])
-        else:
-            raise ValueError("Argument corpus should be 'ssj' or 'kres'.")
+    def __init__(self, corpus, kres_srl_folder=None, logger=None):
 
         self.corpus = corpus
+
+        if self.corpus == "kres":
+            self.kres_srl_folder = kres_srl_folder
+
         self.W_TAGS = ['w']
         self.C_TAGS = ['c']
         self.S_TAGS = ['S', 'pc']
@@ -30,6 +27,10 @@ class Parser():
             "missing_srl": []
         }
 
+        # for logging output
+        self.n_kres_files = -1
+        self.nth_kres_file = -1
+
     def parse_jos_links(self, sent_el):
         if self.corpus == "kres":
             return self.parse_jos_links_kres(sent_el)
@@ -90,14 +91,34 @@ class Parser():
 
     def sentence_generator(self):
         # Using generators so we don't copy a whole corpu around in memory.
+        # Use parse_xml_file() instead for pre-file processing (parallelism?)
         if self.corpus == "kres":
+
+            # some logging output
+            if self.n_kres_files == -1:
+                self.n_kres_files = len(list(Path(self.kres_folder).glob('*')))
+
             for xml_file in self.kres_folder.iterdir():
-                # self.parse_xml_file(xml_file)
-                yield from self.parse_xml_file(xml_file)
+                self.nth_kres_file += 1
+                self.logger.info("{} ({}/{})".format(
+                    xml_file, self.nth_kres_file, self.n_kres_files))
+                yield from self.xml_file_to_generator(xml_file)
         else:
-            yield from self.parse_xml_file(self.ssj_file)
+            yield from self.xml_file_to_generator(self.ssj_file)
 
     def parse_xml_file(self, xml_file):
+        # tstart = time.time()
+        file_data = []
+        for tpl in self.xml_file_to_generator(xml_file):
+            file_data += [tpl[1]]
+        tend = time.time()
+        # self.logger.info("Parsed {} in {:.4f} s".format(xml_file, tend - tstart))
+        return file_data
+
+    def xml_file_to_generator(self, xml_file):
+        # for separate srl links, it will guess the srl file based on 
+        # self.kres_srl_folder
+
         srl_from_json = {}
         if self.corpus == "kres":  
             # in case of kres, read the SRL links form a separate json file
@@ -190,7 +211,7 @@ class Parser():
                         "text": sentence_text,
                         "tokens": sentence_tokens,
                         "jos_links": jos_links,
-                        "srl_links": srl_links_parsed
+                        "srl_links": srl_links_parsed,
                     }
                     self.stats["parsed_count"] += 1
                     yield (xml_file, sentence_entry)

corpusparser/__init__.py

@@ -1 +1,2 @@
 from corpusparser.Parser import Parser
+from corpusparser.main import enriched_lemma

corpusparser/main.py

@@ -4,74 +4,93 @@ import argparse
 import logging
 import json
 from pymongo import MongoClient
+import pymongo
 import sys
+from multiprocessing import Pool
+import time
 
 logging.basicConfig(level=logging.INFO)
 logger = logging.getLogger(__name__)
 
-## Main handles command line arguments and writing to files / DB.  
+n_kres_files = -1  # for logging
 
-def ssj_to_json_file(sentence_generator, outfolder):
-    # this funciton is based on the fact that files are parsed sequentially
-    outfolder = Path(outfolder)
-    outfolder.mkdir(parents=True, exist_ok=True)
-    outfile = outfolder / "ssj500k.json"
 
-    data_buffer = []
-    for s in sentence_generator:
-        sdata = s[1]
-        data_buffer += [sdata]
+def enriched_lemma(token):
+    return (token["lemma"] if token["msd"][0] == "G" else token["lemma"] + "_")
 
-    # outfile = Path(outfile)
-    with outfile.open("w") as fp:
-        logger.info("Writing to {}".format(outfile))
-        json.dump(data_buffer, fp)
 
-def kres_to_json_files(sentence_generator, outfolder):
-    outfolder = Path(outfolder) / "kres_json"
-    outfolder.mkdir(parents=True, exist_ok=True)
+def _helper_tid_to_token(tid, tokens):
+    for t in tokens:
+        if t["tid"] == tid:
+            return t
+    return None
 
-    def write_buffer_to_file(outfile, outfile_buffer):
-        logger.info("Writing file: {}".format(outfile))
-        with outfile.open("w") as fp:
-            json.dump(outfile_buffer, fp)
 
-    outfile_buffer = None
-    current_outfile = None
-    for s in sentence_generator:
-        infile = s[0]
-        outfile = outfolder / Path(infile.name.split(".")[0]).with_suffix(".json")
+def _db_preprocess(e):
+    if e["srl_links"] is None:
+        e["headwords"] = []
+        e["functors"] = []
+    else:
+        hw_tids = list(set([x["from"] for x in e["srl_links"]]))
+        hw_tokens = [_helper_tid_to_token(tid, e["tokens"]) for tid in hw_tids]
+        headwords = [enriched_lemma(t) for t in hw_tokens]
+        e["headwords"] = headwords
 
-        # parser sequentially parses files; when we're done with a file, write it out
-        if current_outfile is None:
-            current_outfile = outfile
-            outfile_buffer = [] 
-        elif outfile != current_outfile:
-            write_buffer_to_file(current_outfile, outfile_buffer)
-            current_outfile = outfile
-            outfile_buffer = []
+        functors = list(set([x["afun"] for x in e["srl_links"]]))
+        e["functors"] = functors
+    return e
 
-        # update buffer
-        sdata = s[1]
-        outfile_buffer += [sdata]
-    write_buffer_to_file(current_outfile, outfile_buffer)
 
-def data_to_valdb(sentence_generator, dbaddr, username, password, collection_name):
-    logger.info("Connecting to: {}".format(dbaddr))
-    client = MongoClient(
-        "mongodb://{}".format(dbaddr),
-        username=username,
-        password=password,
+# handler for concurrency
+def _handle_kres_file_tpl(kres_file_tpl):
+    tstart = time.time()
+    kres_file_idx = kres_file_tpl[0]
+    kres_file = kres_file_tpl[1]
+    kres_data = kres_parser.parse_xml_file(kres_file)
+    if args.output == "file":
+        kres_outdir = outdir / "kres_json"
+        kres_outdir.mkdir(parents=True, exist_ok=True)
+        kres_outfile = kres_outdir / Path(kres_file.name.split(".")[0]).with_suffix(".json")
+        with kres_outfile.open("w") as fp:
+            json.dump(kres_data, fp)
+    elif args.output == "db":
+        # mongoclient needs to be created after forking
+        dbclient = MongoClient(
+            "mongodb://{}".format(args.dbaddr),
+            username=args.dbuser,
+            password=args.dbpass,
             authSource="valdb",
-        authMechanism='SCRAM-SHA-256'
+            authMechanism='SCRAM-SHA-1'
         )
-    valdb = client.valdb
-    logger.info("Writing data to {}.".format(collection_name))
-    col = valdb[collection_name]
-    for s in sentence_generator:
-        sdata = s[1]
-        col.insert_one(sdata)
+        valdb = dbclient.valdb
+        kres_col = valdb["kres"]
 
+        # HUUUUGE BOTTLENECK
+        """
+        for sentence in kres_data:
+            kres_col.update({"sid": sentence["sid"]}, sentence, upsert=True)
+        """
+
+        # skip if one of the sentences is already in DB
+        if kres_col.find({"sid": kres_data[0]["sid"]}).count() > 0:
+            logging.info("File {} already in DB ({}/{})".format(
+                kres_file, kres_file_idx, n_kres_files))
+            return
+
+        kres_data_1 = [_db_preprocess(x) for x in kres_data]
+        kres_col.insert_many(kres_data_1)  # much much better (just make sure sid has a unique index)
+        logging.info("Inserted data from {} ({}/{}) in {:.2f} s".format(
+            kres_file, kres_file_idx, n_kres_files, time.time() - tstart))
+
+def _get_dbclient(args):
+    dbclient = MongoClient(
+        "mongodb://{}".format(args.dbaddr),
+        username=args.dbuser,
+        password=args.dbpass,
+        authSource="valdb",
+        authMechanism='SCRAM-SHA-1'
+    )
+    return dbclient
 
 if __name__ == "__main__":
     parser = argparse.ArgumentParser(description="Parsing corpora kres and ssj500k.")
@@ -83,41 +102,51 @@ if __name__ == "__main__":
     parser.add_argument('--dbaddr', required=False, default=None)
     parser.add_argument('--dbuser', required=False, default=None)
     parser.add_argument('--dbpass', required=False, default=None)
+    parser.add_argument('--cores', required=False, default=1)
     args = parser.parse_args()
 
-
-    # parse ssj
-    logger.info("Parsing ssj500k: {}".format(args.ssj_file))
-    ssj_parser = Parser(
-        corpus="ssj",
-        infiles=[args.ssj_file],
-    )
-    # ssj to json
+    outdir = None
     if args.output == "file":
-        ssj_to_json_file(ssj_parser.sentence_generator(), args.outdir)
+        outdir = Path(args.outdir)
+        outdir.mkdir(parents=True, exist_ok=True)
     elif args.output == "db":
-        data_to_valdb(
-            ssj_parser.sentence_generator(),
-            args.dbaddr,
-            args.dbuser,
-            args.dbpass,
-            collection_name="ssj"
-        )
+        # Force unique sid
+        dbclient = _get_dbclient(args)
+        for corpus in ["kres", "ssj"]:
+            dbclient.valdb[corpus].ensure_index([("sid", pymongo.ASCENDING)])
+            dbclient.valdb[corpus].ensure_index([("headwords", pymongo.ASCENDING)])
+            dbclient.valdb[corpus].ensure_index([("functors", pymongo.ASCENDING)])
 
-    # parse kres
-    logger.info("Parsing Kres: {}".format(args.ssj_file))
+    # SSJ
+    logger.info("Parsing Ssj: {}".format(args.ssj_file))
+    ssj_parser = Parser(corpus="ssj")
+    ssj_data = ssj_parser.parse_xml_file(Path(args.ssj_file))
+    if args.output == "file":
+        ssj_outfile = outdir / "ssj500k.json"
+        with ssj_outfile.open("w") as fp:
+            json.dump(ssj_data, fp)
+    elif args.output == "db":
+        dbclient = _get_dbclient(args)
+        valdb = dbclient.valdb
+        ssj_col = valdb["ssj"]
+        for sentence in ssj_data:
+            sentence = _db_preprocess(sentence)
+            ssj_col.update({"sid": sentence["sid"]}, sentence, upsert=True)
+
+
+    # Kres
+    logger.info("Parsing Kres: {}".format(args.kres_folder))
     kres_parser = Parser(
         corpus="kres",
-        infiles=[args.kres_folder, args.kres_srl_folder],
-    )
-    # kres to json
-    if args.output == "file":
-        kres_to_json_files(kres_parser.sentence_generator(), args.outdir)
-    elif args.output == "db":
-        data_to_valdb(
-            kres_parser.sentence_generator(),
-            args.dbaddr,
-            args.dbuser,
-            args.dbpass,
-            collection_name="kres"
+        kres_srl_folder=args.kres_srl_folder
     )
+
+    # [(idx, filepath)]
+    kres_files = [x for x in Path(args.kres_folder).iterdir()]
+    kres_files = [x for x in enumerate(kres_files)]
+    n_kres_files = len(kres_files)
+
+    p = Pool(int(args.cores))
+    p.map(_handle_kres_file_tpl, kres_files)
+
+    logger.info("Finished parsing.")