rephrase improving versus changing

kermitt2 · Nov 29, 2023 · 194767d · 194767d
1 parent f4c9d89
commit 194767d
Show file tree

Hide file tree

Showing 5 changed files with 9 additions and 9 deletions.
diff --git a/delft/applications/datasetTagger.py b/delft/applications/datasetTagger.py
@@ -306,8 +306,8 @@ def annotate_text(texts, output_format, architecture='BidLSTM_CRF', features=Non
     parser.add_argument("--max-epoch", type=int, default=-1,
                         help="Maximum number of epochs.")
     parser.add_argument("--early-stop", type=t_or_f, default=None,
-                        help="Force training early termination when evaluation scores at the end of "
-                             "n epochs are not changing.")
+                        help="Force early training termination when metrics scores are not improving " + 
+                             "after a number of epochs equals to the patience parameter.")
     parser.add_argument("--multi-gpu", default=False,
                         help="Enable the support for distributed computing (the batch size needs to be set accordingly using --batch-size)",
                         action="store_true")

diff --git a/delft/applications/grobidTagger.py b/delft/applications/grobidTagger.py
@@ -422,8 +422,8 @@ class Tasks:
     parser.add_argument("--max-epoch", type=int, default=-1,
                         help="Maximum number of epochs for training.")
     parser.add_argument("--early-stop", type=t_or_f, default=None,
-                        help="Force training early termination when evaluation scores at the end of "
-                             "n epochs are not changing.")
+                        help="Force early training termination when metrics scores are not improving " + 
+                             "after a number of epochs equals to the patience parameter.")
 
     parser.add_argument("--multi-gpu", default=False,
                         help="Enable the support for distributed computing (the batch size needs to be set accordingly using --batch-size)",

diff --git a/delft/applications/insultTagger.py b/delft/applications/insultTagger.py
@@ -137,8 +137,8 @@ def annotate(texts, output_format, architecture='BidLSTM_CRF', transformer=None,
                         help="Maximum number of epochs.")
     parser.add_argument("--batch-size", type=int, default=-1, help="batch-size parameter to be used.")
     parser.add_argument("--early-stop", type=t_or_f, default=None,
-                        help="Force training early termination when evaluation scores at the end of "
-                             "n epochs are not changing.")
+                        help="Force early training termination when metrics scores are not improving " + 
+                             "after a number of epochs equals to the patience parameter.")
 
     parser.add_argument("--multi-gpu", default=False,
                         help="Enable the support for distributed computing (the batch size needs to be set accordingly using --batch-size)",

diff --git a/delft/applications/nerTagger.py b/delft/applications/nerTagger.py
@@ -635,8 +635,8 @@ def annotate(output_format,
     parser.add_argument("--max-epoch", type=int, default=-1,
                         help="Maximum number of epochs.")
     parser.add_argument("--early-stop", type=t_or_f, default=None,
-                        help="Force training early termination when evaluation scores at the end of "
-                             "n epochs are not changing.")
+                        help="Force early training termination when metrics scores are not improving " + 
+                             "after a number of epochs equals to the patience parameter.")
 
     parser.add_argument("--multi-gpu", default=False,
                         help="Enable the support for distributed computing (the batch size needs to be set accordingly using --batch-size)",

diff --git a/doc/ner.md b/doc/ner.md
@@ -34,7 +34,7 @@ Results with transformer fine-tuning for CoNLL-2003 NER dataset, including a fin
 | ---  | --- | --- | --- |
 | BERT | bert-base-cased     | DeLFT | 91.19 |  
 | BERT_CRF | bert-base-cased +CRF| DeLFT | 91.25 |  
-| BERT_ChainCRF | bert-base-cased +CRF| DeLFT |  |  
+| BERT_ChainCRF | bert-base-cased +CRF| DeLFT | 91.22 |  
 | BERT | roberta-base     | DeLFT | 91.64 |  
 
 Note: DeLFT uses `BERT` as architecture name for transformers in general, but the transformer model could be in principle any transformer variants preset in HuggingFace Hub. DeLFT supports 2 implementations of a CRF layer to be combined with RNN and transformer architectures: `CRF` based on TensorFlow Addons and `ChainCRF` a custom implementation. Both should produce similar accuracy results, but `ChainCRF` is significantly faster and robust.