diff --git a/UD_any/morpho/machine.rm b/UD_any/morpho/machine.rm
index 5ff55ba71ed460a06373763423f6cf4bd1f34052..eeaeaa5fe3fdb4ed2235fff8a492e1b45c3f2e60 100644
--- a/UD_any/morpho/machine.rm
+++ b/UD_any/morpho/machine.rm
@@ -1,9 +1,9 @@
Name : Feats Machine
Classifier : morpho
{
- Transitions : {data/morpho_parts.ts}
+ Transitions : {morpho,data/morpho_parts.ts}
Network type : LSTM
- Unknown value threshold : -1
+ Unknown value threshold : 1
Buffer context : {-3 -2 -1 0 1 2}
Stack context : {}
Columns : {FORM UPOS}
@@ -13,15 +13,24 @@ Classifier : morpho
Max nb elements : {1 10 10}
Raw input left window : -1
Raw input right window : -1
- Embeddings size : 256
- MLP : {8192 0.3}
- Context LSTM size : 1024
+ 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 : 2
+ Num layers : 3
BiLSTM : true
- LSTM dropout : 0.3
+ LSTM dropout : 0.1
+ Total input dropout : 0.3
+ Embeddings dropout : 0.3
+ Dropout 2d : false
+ Tree embedding columns : {}
+ Tree embedding buffer : {}
+ Tree embedding stack : {}
+ Tree embedding nb : {}
+ Tree embedding size : 0
+ Optimizer : Adam {0.0005 0.9 0.999 0.00000001 0.00001 true}
}
Predictions : FEATS
Strategy : sequential
diff --git a/UD_any/parser/machine.rm b/UD_any/parser/machine.rm
index a7105a59100b7bd4a81e3f62880aae38a11f6ecb..c5fb678697662558d121dc5ced060663e9fd6d37 100644
--- a/UD_any/parser/machine.rm
+++ b/UD_any/parser/machine.rm
@@ -1,27 +1,36 @@
Name : Parser Machine
Classifier : parser
{
- Transitions : {data/parser.ts}
+ Transitions : {parser,data/parser.ts}
Network type : LSTM
- Unknown value threshold : -1
+ Unknown value threshold : 1
Buffer context : {-3 -2 -1 0 1 2}
- Stack context : {}
+ Stack context : {2 1 0}
Columns : {FORM UPOS}
Focused buffer : {-1 0}
- Focused stack : {-1 0}
- Focused columns : {ID FORM FEATS}
- Max nb elements : {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 : -1
Raw input right window : -1
- Embeddings size : 256
- MLP : {8192 0.3}
- Context LSTM size : 1024
+ 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 : 2
+ Num layers : 3
BiLSTM : true
- LSTM dropout : 0.3
+ 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}
}
Predictions : HEAD DEPREL EOS
Strategy : sequential
diff --git a/UD_any/tagger/machine.rm b/UD_any/tagger/machine.rm
index c54c5d81132286a6d4d87d3545bbaa7e6caa498d..cbfd0b71b6e31d52528a852051ecec21fe15789c 100644
--- a/UD_any/tagger/machine.rm
+++ b/UD_any/tagger/machine.rm
@@ -1,9 +1,9 @@
Name : Tagger Machine
Classifier : tagger
{
- Transitions : {data/tagger.ts}
+ Transitions : {tagger,data/tagger.ts}
Network type : LSTM
- Unknown value threshold : -1
+ Unknown value threshold : 1
Buffer context : {-3 -2 -1 0 1 2}
Stack context : {}
Columns : {FORM UPOS}
@@ -13,15 +13,24 @@ Classifier : tagger
Max nb elements : {1 10}
Raw input left window : -1
Raw input right window : -1
- Embeddings size : 256
- MLP : {8192 0.3}
- Context LSTM size : 1024
+ 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 : 2
+ Num layers : 3
BiLSTM : true
- LSTM dropout : 0.3
+ LSTM dropout : 0.1
+ Total input dropout : 0.3
+ Embeddings dropout : 0.3
+ Dropout 2d : false
+ Tree embedding columns : {}
+ Tree embedding buffer : {}
+ Tree embedding stack : {}
+ Tree embedding nb : {}
+ Tree embedding size : 0
+ Optimizer : Adam {0.0005 0.9 0.999 0.00000001 0.00001 true}
}
Predictions : UPOS
Strategy : sequential
diff --git a/UD_any/tagparser_incr/machine.rm b/UD_any/tagparser_incr/machine.rm
deleted file mode 100644
index 334d08ca6bef1af0473708d0929730fde4d46d9f..0000000000000000000000000000000000000000
--- a/UD_any/tagparser_incr/machine.rm
+++ /dev/null
@@ -1,8 +0,0 @@
-Name : Tagger+Parser Machine
-Classifier : any CNN(4,4,2) data/taggerparser.ts
-Predictions : UPOS HEAD DEPREL EOS
-Strategy : incremental
- tagger parser 0
- parser tagger SHIFT 1
- parser tagger RIGHT 1
- parser parser 0
diff --git a/UD_any/tokemorpho_incr/machine.rm b/UD_any/tokemorpho_incr/machine.rm
deleted file mode 100644
index 04abba11d6ae795742d0946b9349ea94aa330c85..0000000000000000000000000000000000000000
--- a/UD_any/tokemorpho_incr/machine.rm
+++ /dev/null
@@ -1,43 +0,0 @@
-Name : Tokenizer, Tagger and Morpho Machine
-Classifier : tokemorpho
-{
- Transitions : {data/tokenizer.ts data/tagger.ts data/morpho_parts.ts}
- Network type : LSTM
- Unknown value threshold : 1
- Buffer context : {-3 -2 -1}
- Stack context : {}
- Columns : {FORM UPOS}
- Focused buffer : {-1 0}
- Focused stack : {}
- Focused columns : {ID FORM FEATS}
- Max nb elements : {1 10 10}
- Raw input left window : 5
- Raw input right window : 5
- Embeddings size : 64
- MLP : {2048 0.3}
- Context LSTM size : 1024
- Focused LSTM size : 256
- Rawinput LSTM size : 16
- Split trans LSTM size : 128
- Num layers : 3
- BiLSTM : true
- LSTM dropout : 0.1
- Total input dropout : 0.3
- Embeddings dropout : 0.3
- Dropout 2d : false
- Tree embedding columns : {}
- Tree embedding buffer : {}
- Tree embedding stack : {}
- Tree embedding nb : {}
- Tree embedding size : 0
- Optimizer : Adam {0.0005 0.9 0.999 0.00000001 0.00001 true}
-}
-Splitwords : data/splitwords.ts
-Predictions : ID FORM UPOS FEATS EOS
-Strategy : incremental
- tokenizer tagger ENDWORD 0
- tokenizer tagger SPLIT 0
- tokenizer tokenizer 0
- tagger morpho 0
- morpho tokenizer NOTHING 1
- morpho morpho 0
diff --git a/UD_any/tokemorpho_seq/machine.rm b/UD_any/tokemorpho_seq/machine.rm
deleted file mode 100644
index 43fcf8bc88b073d09fc4799433e85f4b27408a55..0000000000000000000000000000000000000000
--- a/UD_any/tokemorpho_seq/machine.rm
+++ /dev/null
@@ -1,34 +0,0 @@
-Name : Tokenizer, Tagger and Morpho Machine
-Classifier : tokemorpho
-{
- Transitions : {data/tokenizer.ts data/tagger.ts data/morpho_parts.ts}
- Network type : LSTM
- Unknown value threshold : -1
- Buffer context : {-3 -2 -1 1 2}
- Stack context : {}
- Columns : {FORM UPOS}
- Focused buffer : {-1 0}
- Focused stack : {}
- Focused columns : {ID FORM FEATS}
- Max nb elements : {1 10 10}
- Raw input left window : 5
- Raw input right window : 5
- Embeddings size : 256
- MLP : {8192 0.3}
- Context LSTM size : 1024
- Focused LSTM size : 256
- Rawinput LSTM size : 32
- Split trans LSTM size : 256
- Num layers : 2
- BiLSTM : true
- LSTM dropout : 0.3
-}
-Splitwords : data/splitwords.ts
-Predictions : ID FORM UPOS FEATS EOS
-Strategy : sequential
- tokenizer tagger ENDWORD 1
- tokenizer tagger SPLIT 1
- tokenizer tagger 0
- tagger morpho 1
- morpho tokenizer NOTHING 1
- morpho tokenizer 0
diff --git a/UD_any/tokenizer/machine.rm b/UD_any/tokenizer/machine.rm
index 9b1252e29109806c3a6919c89d2315aef49bbcc8..bc622d2c5fd99e706f7d30d776f98a65d5ed3d90 100644
--- a/UD_any/tokenizer/machine.rm
+++ b/UD_any/tokenizer/machine.rm
@@ -1,9 +1,9 @@
Name : Tokenizer Machine
Classifier : tokenizer
{
- Transitions : {data/tokenizer.ts}
+ Transitions : {tokenizer,data/tokenizer.ts}
Network type : LSTM
- Unknown value threshold : -1
+ Unknown value threshold : 1
Buffer context : {-3 -2 -1}
Stack context : {}
Columns : {FORM}
@@ -13,15 +13,24 @@ Classifier : tokenizer
Max nb elements : {1 10}
Raw input left window : 5
Raw input right window : 5
- Embeddings size : 256
- MLP : {8192 0.3}
- Context LSTM size : 1024
+ 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 : 2
+ Num layers : 3
BiLSTM : true
- LSTM dropout : 0.3
+ LSTM dropout : 0.1
+ Total input dropout : 0.3
+ Embeddings dropout : 0.3
+ Dropout 2d : false
+ Tree embedding columns : {}
+ Tree embedding buffer : {}
+ Tree embedding stack : {}
+ Tree embedding nb : {}
+ Tree embedding size : 0
+ Optimizer : Adam {0.0005 0.9 0.999 0.00000001 0.00001 true}
}
Splitwords : data/splitwords.ts
Predictions : ID FORM EOS
diff --git a/UD_any/toketagger_incr/machine.rm b/UD_any/toketagger_incr/machine.rm
deleted file mode 100644
index eedf496f11763fda6400d3d92c2eff7dcf6b7301..0000000000000000000000000000000000000000
--- a/UD_any/toketagger_incr/machine.rm
+++ /dev/null
@@ -1,9 +0,0 @@
-Name : Tokenizer and Tagger Machine
-Classifier : toketagger LSTM(-1,{-3,-2,-1},{},{FORM,UPOS},{-1,0},{},{ID,FORM},{1,10},5,5) {data/tokenizer.ts data/tagger.ts}
-Splitwords : data/splitwords.ts
-Predictions : ID FORM UPOS EOS
-Strategy : incremental
- tokenizer tagger ENDWORD 0
- tokenizer tagger SPLIT 0
- tokenizer tokenizer 0
- tagger tokenizer 1
diff --git a/UD_any/toketagger_seq/machine.rm b/UD_any/toketagger_seq/machine.rm
deleted file mode 100644
index 92fd15365f9bbb5de6451378ca6b583f20c2b4b8..0000000000000000000000000000000000000000
--- a/UD_any/toketagger_seq/machine.rm
+++ /dev/null
@@ -1,9 +0,0 @@
-Name : Tokenizer and Tagger Machine
-Classifier : toketagger LSTM(-1,{-3,-2,-1,0,1,2},{},{FORM,UPOS},{-1,0},{},{ID,FORM},{1,10},5,5) {data/tokenizer.ts data/tagger.ts}
-Splitwords : data/splitwords.ts
-Predictions : ID FORM UPOS EOS
-Strategy : sequential
- tokenizer tagger ENDWORD 1
- tokenizer tagger SPLIT 1
- tokenizer tagger 0
- tagger tokenizer 1