From 1078b0cee54cb0174d3ff38150bcf970dd1f4e49 Mon Sep 17 00:00:00 2001
From: Franck Dary <franck.dary@lis-lab.fr>
Date: Sun, 19 Apr 2020 21:01:45 +0200
Subject: [PATCH] updated tokeparser machines
---
UD_any/tokeparser_base/machine.rm | 46 +++++++++++++++++++++++++++++++
UD_any/tokeparser_incr/machine.rm | 34 +++++++++++++----------
UD_any/tokeparser_seq/machine.rm | 36 +++++++++++++-----------
3 files changed, 85 insertions(+), 31 deletions(-)
create mode 100644 UD_any/tokeparser_base/machine.rm
diff --git a/UD_any/tokeparser_base/machine.rm b/UD_any/tokeparser_base/machine.rm
new file mode 100644
index 0000000..5c4a1d0
--- /dev/null
+++ b/UD_any/tokeparser_base/machine.rm
@@ -0,0 +1,46 @@
+Name : Tokenizer, Tagger, Morpho and Parser Machine
+Classifier : tokeparser
+{
+ Transitions : {data/tokenizer.ts data/tagger.ts data/morpho_parts.ts data/parser.ts}
+ Network type : LSTM
+ Unknown value threshold : 1
+ Buffer context : {-3 -2 -1}
+ Stack context : {2 1 0}
+ Columns : {FORM UPOS}
+ Focused buffer : {-1 0}
+ Focused stack : {2 1 0}
+ Focused columns : {ID EOS FORM FEATS DEPREL}
+ Max nb elements : {1 1 10 10 1}
+ Raw input left window : 5
+ Raw input right window : 5
+ Embeddings size : 128
+ MLP : {2048 0.3}
+ Context LSTM size : 512
+ Focused LSTM size : 256
+ Rawinput LSTM size : 32
+ Split trans LSTM size : 256
+ Num layers : 3
+ BiLSTM : true
+ LSTM dropout : 0.1
+ Total input dropout : 0.3
+ Embeddings dropout : 0.3
+ Dropout 2d : false
+ Tree embedding columns : {DEPREL}
+ Tree embedding buffer : {-1}
+ Tree embedding stack : {0 1 2}
+ Tree embedding nb : {6}
+ Tree embedding size : 128
+ Optimizer : Adam {0.0005 0.9 0.999 0.00000001 0.00001 true}
+}
+Splitwords : data/splitwords.ts
+Predictions : ID FORM UPOS FEATS HEAD DEPREL EOS
+Strategy : sequential
+ tokenizer tagger ENDWORD 1
+ tokenizer tagger SPLIT 1
+ tokenizer tagger 0
+ tagger morpho 1
+ morpho parser NOTHING 1
+ morpho parser 0
+ parser tokenizer SHIFT 1
+ parser tokenizer RIGHT 1
+ parser tokenizer 0
diff --git a/UD_any/tokeparser_incr/machine.rm b/UD_any/tokeparser_incr/machine.rm
index abe911d..5ec3600 100644
--- a/UD_any/tokeparser_incr/machine.rm
+++ b/UD_any/tokeparser_incr/machine.rm
@@ -1,32 +1,36 @@
-Name : Tokenizer, Tagger and Morpho Machine
-Classifier : tokemorpho
+Name : Tokenizer, Tagger, Morpho and Parser Machine
+Classifier : tokeparser
{
Transitions : {data/tokenizer.ts data/tagger.ts data/morpho_parts.ts data/parser.ts}
Network type : LSTM
- Unknown value threshold : -1
- Buffer context : {-5 -4 -3 -2 -1}
- Stack context : {3 2 1 0}
+ Unknown value threshold : 1
+ Buffer context : {-3 -2 -1}
+ Stack context : {2 1 0}
Columns : {FORM UPOS}
Focused buffer : {-1 0}
- Focused stack : {1 0}
- Focused columns : {ID EOS FORM FEATS}
- Max nb elements : {1 1 10 10}
+ Focused stack : {2 1 0}
+ Focused columns : {ID EOS FORM FEATS DEPREL}
+ Max nb elements : {1 1 10 10 1}
Raw input left window : 5
Raw input right window : 5
- Embeddings size : 256
- MLP : {2048 0.3 2048 0.3}
+ Embeddings size : 128
+ MLP : {2048 0.3}
Context LSTM size : 512
Focused LSTM size : 256
- Rawinput LSTM size : 64
+ Rawinput LSTM size : 32
Split trans LSTM size : 256
Num layers : 3
BiLSTM : true
- LSTM dropout : 0.3
- Tree embedding columns : {FORM POS DEPREL}
+ LSTM dropout : 0.1
+ Total input dropout : 0.3
+ Embeddings dropout : 0.3
+ Dropout 2d : false
+ Tree embedding columns : {DEPREL}
Tree embedding buffer : {-1}
- Tree embedding stack : {0 1}
- Tree embedding nb : {6 10}
+ Tree embedding stack : {0 1 2}
+ Tree embedding nb : {6}
Tree embedding size : 128
+ Optimizer : Adam {0.0005 0.9 0.999 0.00000001 0.00001 true}
}
Splitwords : data/splitwords.ts
Predictions : ID FORM UPOS FEATS HEAD DEPREL EOS
diff --git a/UD_any/tokeparser_seq/machine.rm b/UD_any/tokeparser_seq/machine.rm
index e2b2ab0..77947cd 100644
--- a/UD_any/tokeparser_seq/machine.rm
+++ b/UD_any/tokeparser_seq/machine.rm
@@ -1,32 +1,36 @@
-Name : Tokenizer, Tagger and Morpho Machine
-Classifier : tokemorpho
+Name : Tokenizer, Tagger, Morpho and Parser Machine
+Classifier : tokeparser
{
Transitions : {data/tokenizer.ts data/tagger.ts data/morpho_parts.ts data/parser.ts}
Network type : LSTM
- Unknown value threshold : -1
- Buffer context : {-5 -4 -3 -2 -1 1 2}
- Stack context : {3 2 1 0}
+ Unknown value threshold : 1
+ Buffer context : {-3 -2 -1 1 2 3}
+ Stack context : {2 1 0}
Columns : {FORM UPOS}
- Focused buffer : {-1 0}
- Focused stack : {1 0}
- Focused columns : {ID EOS FORM FEATS}
- Max nb elements : {1 1 10 10}
+ Focused buffer : {-1 0 1 2}
+ Focused stack : {2 1 0}
+ Focused columns : {ID EOS FORM FEATS DEPREL}
+ Max nb elements : {1 1 10 10 1}
Raw input left window : 5
Raw input right window : 5
- Embeddings size : 256
- MLP : {2048 0.3 2048 0.3}
+ Embeddings size : 128
+ MLP : {2048 0.3}
Context LSTM size : 512
Focused LSTM size : 256
- Rawinput LSTM size : 64
+ Rawinput LSTM size : 32
Split trans LSTM size : 256
Num layers : 3
BiLSTM : true
- LSTM dropout : 0.3
- Tree embedding columns : {FORM POS DEPREL}
+ LSTM dropout : 0.1
+ Total input dropout : 0.3
+ Embeddings dropout : 0.3
+ Dropout 2d : false
+ Tree embedding columns : {DEPREL}
Tree embedding buffer : {-1}
- Tree embedding stack : {0 1}
- Tree embedding nb : {6 10}
+ Tree embedding stack : {0 1 2}
+ Tree embedding nb : {6}
Tree embedding size : 128
+ Optimizer : Adam {0.0005 0.9 0.999 0.00000001 0.00001 true}
}
Splitwords : data/splitwords.ts
Predictions : ID FORM UPOS FEATS HEAD DEPREL EOS
--
GitLab