Quiet logging.

Author: attardi

Makefile

@@ -79,7 +79,7 @@ else ifeq ($(LANG), fr)
 else ifeq ($(LANG), it)
   CORPUS=it_isdt
   RES2=Italian-ISDT
-  MODEL = --bert= dbmdz/bert-base-italian-xxl-cased
+  MODEL = --bert=dbmdz/bert-base-italian-xxl-cased
   BERT = dbmdz-xxl
 else ifeq ($(LANG), lt)
   CORPUS=lt_alksnis
@@ -146,9 +146,11 @@ endif
 .PRECIOUS: exp/$(LANG)-$(FEAT)$(VER)/model
 
 # relate LANG to CORPUS
-exp/$(LANG)%: exp/$(CORPUS).$(BERT)%
+exp/$(LANG)%: exp/$(CORPUS).$(BERT)$(VER)%
 	@:
 
+TARGET=exp/$(CORPUS).$(BERT)$(VER)
+
 exp/$(CORPUS).$(BERT)$(VER)/model:
 	python -u -m diaparser.cmds.biaffine_dependency train -d=$(GPU) -p=$@ \
 	   -c=$(CONFIG) $(MODEL) $(ATTN) \
@@ -161,11 +163,6 @@ exp/$(CORPUS).$(BERT)$(VER).test.conllu: exp/$(CORPUS).$(BERT)$(VER)/model
 	   --pred=$@
 	python $(CORPUS_DIR)/fix-root.py $@
 
-exp/$(CORPUS).$(BERT)$(VER).test.time: exp/$(CORPUS).$(BERT)$(VER)/model
-	( time python -m diaparser.cmds.biaffine_dependency predict -d=$(GPU) -p=$< --feat=$(FEAT) --tree  \
-	   $(BLIND_TEST)  \
-	   --pred=/dev/null; ) &> $@
-
 LANGS=ar bg cs en et fi fr it lt lv nl pl ru sk sv ta uk 
 LANGS1=ar bg en et fi sk
 LANGS2=fr it ru ta uk sv
@@ -184,16 +181,16 @@ train:
 # ----------------------------------------------------------------------
 # Evaluation
 
-%.test.nen.conllu: %.test.conllu
+$(TARGET).test.nen.conllu: $(TARGET).test.conllu
 	   perl $(UD_TOOLS)/enhanced_collapse_empty_nodes.pl $< > $@
 
-%.test.eval: %.test.nen.conllu
+$(TARGET).test.eval: $(TARGET).test.nen.conllu
 	python $(UD_TOOLS)/iwpt20_xud_eval.py -v $(UD_TOOLS)/../test-gold/$(LANG).nen.conllu $< > $@
 
-%.test.evalb: %.test.eval
+$(TARGET).test.evalb: $(TARGET).test.eval
 	python $(CORPUS_DIR)/eval.py -g $(GOLD_TEST) -s $@ --evalb
 
-%.test.eval07: %.test.conllu
+$(TARGET).test.eval07: $(TARGET).test.conllu
 	perl $(CORPUS_DIR)/eval07.pl -p -q -g $(GOLD_TEST) -s $< > $@
 
 evaluate:

diaparser/cmds/biaffine_dependency.py

@@ -6,7 +6,7 @@
 
 from ..parsers import BiaffineDependencyParser
 from .cmd import parse
-
+import os
 
 def main():
     parser = argparse.ArgumentParser(description='Create Biaffine Dependency Parser.')
@@ -45,7 +45,7 @@ def main():
     subparser.add_argument('--pred', default='pred.conllx', help='path to predicted result')
     subparser.add_argument('--text', metavar='LANGUAGE', default=None,
                            help='parse plain text in the given language rather than CoNLL-U files.')
-    subparser.add_argument('--cache-dir', default='~/.cache/parser',
+    subparser.add_argument('--cache-dir', default=os.path.expanduser('~/.cache/diaparser'),
                            help='path to saved parser/tokenizer models')
     parse(parser)
 

diaparser/cmds/cmd.py

@@ -22,7 +22,7 @@ def parse(argparser):
     torch.set_num_threads(args.threads)
     torch.manual_seed(args.seed)
     init_device(args.device)
-    init_logger(logger, f"{args.path}.{args.mode}.log")
+    init_logger(logger, f"{args.path}.{args.mode}.log", verbose=args.verbose)
     logger.info('Configuration:\n' + str(args))
 
     if args.mode == 'train':

tokenizer/tokenizer.py

@@ -29,19 +29,20 @@ def format(self, sentences):
         """
         Convert sentences to TSV format.
         """
+        empty_fields = '\t_' * 8
         for i, sentence in enumerate(sentences):
             yield f'# sent_id = {i+1}'
             sent_text = sentence.text.replace("\n", " ")
             yield f'# text = {sent_text}'
             for token in sentence.tokens:
                 # multiword
                 if len(token.words) > 1:
-                    token_ids = '-'.join([str(id) for id in token.id])
-                    yield f'{token_ids}\t{token.text}'
+                    token_range = f'{token.id[0]}-{token.id[-1]}'
+                    yield f'{token_range}\t{token.text + empty_fields}'
                     for word in token.words:
-                        yield f'{word.id}\t{word.text}'
+                        yield f'{word.id}\t{word.text + empty_fields}'
                 else:
-                    yield f'{token.id[0]}\t{token.text}'
+                    yield f'{token.id[0]}\t{token.text + empty_fields}'
             yield ''
 
     def reader(self):