Fix error in CompleteFullGraph class.

Author: YNedderhoff

create_screens_german.sh

@@ -0,0 +1,7 @@
+#!/bin/bash
+
+screen -dmS "cle_1" ./start-cle_automated_german.sh 1 m_e-1 p_e-1 e_e-1
+screen -dmS "cle_2" ./start-cle_automated_german.sh 10 m_e-10 p_e-10 e_e-10
+screen -dmS "cle_3" ./start-cle_automated_german.sh 10 m_e-10_da p_e-10_da e_e-10_da -decrease-alpha
+screen -dmS "cle_4" ./start-cle_automated_german.sh 10 m_e-10_ss p_e-10_ss e_e-10_ss -shuffle-sentences
+screen -dmS "cle_5" ./start-cle_automated_german.sh 10 m_e-10_da_ss p_e-10_da_ss e_e-10_da_ss -decrease-alpha -shuffle-sentences

dependency-parser.py

@@ -41,7 +41,6 @@ def save(file_name, model):
 
 
 def train(args):
-
     start = time.time()
     print "\tCreating feature map..."
 
@@ -90,7 +89,7 @@ def train(args):
         for head in sparse_graph:
             for arc in sparse_graph[head]:
                 if arc.feat_vec:
-                    if int(arc.head) == 0:
+                    if arc.head == 0:
                         if len(arc.feat_vec) != 10:
                             print "Length of arc feature vector is wrong."
                             print arc.feat_vec
@@ -150,7 +149,6 @@ def train(args):
     stop = time.time()
     print "\t\tDone, " + str(stop - start) + " sec"
 
-
 def test(args):
 
     # load classifier vectors (model) and feature vector from file:
@@ -200,11 +198,6 @@ def test(args):
     stop = time.time()
     print "\t\tDone, " + str(stop - start) + " sec"
 
-def write_to_file(token, file_obj):
-    print >> file_obj, str(token.id) + "\t" + str(token.form) + "\t" + str(token.lemma) + "\t" + str(
-        token.pos) + "\t" + "_" + "\t" + "_" + "\t" + str(token.head) + "\t" + str(token.rel) + "\t" + "_" + "\t" + "_"
-
-
 if __name__ == '__main__':
 
     t0 = time.time()

modules/featmap.py

@@ -8,7 +8,7 @@ def fm (infile):
     for sentence in sentences(codecs.open(infile,encoding='utf-8')):
         local_features=[]
         for token1 in sentence:
-            if int(token1.head) == 0:
+            if token1.head == 0:
                 # at this point ROOT is the head and token1 is the dependent
 
                 # unigram features

modules/graphs.py

@@ -3,8 +3,8 @@
 
 class SparseArc:  # sparse representation of an arc
     def __init__(self, head, dependent, s=0.0):
-        self.head = int(head)
-        self.dependent = int(dependent)
+        self.head = head
+        self.dependent = dependent
         self.score = s
         self.feat_vec = []
 
@@ -15,8 +15,8 @@ def __init__(self, head, dependent, s=0.0):
 class FullArc:  # full representation of an arc
     def __init__(self, head, dependent, head_form, dependent_form, head_lemma, dependent_lemma, head_pos,
                  dependent_pos, dependent_rel, s=0.0):
-        self.head = int(head)
-        self.dependent = int(dependent)
+        self.head = head
+        self.dependent = dependent
         self.score = s
         self.feat_vec = []
 
@@ -59,7 +59,7 @@ def __init__(self, tokens):
                 new_arc = SparseArc(token1.id, token2.id)
                 dependents.append(new_arc)
             if dependents:
-                self.heads[int(token1.id)] = dependents
+                self.heads[token1.id] = dependents
 
 
 class FullGraph:  # full representation of a graph (keys: heads, values: FullArc objects)
@@ -77,7 +77,7 @@ def __init__(self, tokens):
                                   token1.pos, token2.pos, token2.rel)
                 dependents.append(new_arc)
             if dependents:
-                self.heads[int(token1.id)] = dependents
+                self.heads[token1.id] = dependents
 
 
 class CompleteFullGraph:  # complete, full representation of a graph (keys: heads, values: FullArc objects)
@@ -90,12 +90,12 @@ def __init__(self, tokens):
             self.heads[0].append(new_arc)
         for token1 in tokens:
             dependents = []
-            for token2 in tokens:
+            for token2 in (token2 for token2 in tokens if token2.head == token1.id):
                 new_arc = FullArc(token1.id, token2.id, token1.form, token2.form, token1.lemma, token2.lemma, \
                                   token1.pos, token2.pos, token2.rel)
                 dependents.append(new_arc)
             if dependents:
-                self.heads[int(token1.id)] = dependents
+                self.heads[token1.id] = dependents
 
 
 def reverse_head_graph(graph):  # reverses a normal graph to a graph where the dependents are the keys

modules/perceptron.py

@@ -42,6 +42,7 @@ def structured_perceptron(graph, feat_map, rev_feat_map, weight_vector, correct,
             errors += 1
 
         if make_graph_compareable(y_predicted) == make_graph_compareable(y_gold):
+            print "yes"
             if tmp_errors < errors:
                 print "Correct but error, this is not possible"
             correct +=1

start-cle.sh

@@ -11,10 +11,11 @@ EVALUATIONS="evaluations"
 #python -u dependency-parser.py -train -i $CORPORA/train/wsj_train.first-5k.conll06 -m model -e 1 -decrease-alpha -shuffle-sentences
 
 #train
-#python -u dependency-parser.py -train -i ../dependency-parsing-files/data/english/train/wsj_train.conll06 -m model -e 2 -decrease-alpha -shuffle-sentences
+python -u dependency-parser.py -train -i ../dependency-parsing-files/data/english/train/wsj_train.conll06 -m $MODELS/m_e-10_da_ss -e 2 -decrease-alpha -shuffle-sentences
 
 #test
-python -u dependency-parser.py -test -i $CORPORA/dev/wsj_dev.conll06 -m model -o predicted.conll06
+#python -u dependency-parser.py -test -i $CORPORA/dev/wsj_dev.conll06 -m model -o predicted.conll06
+#python -u dependency-parser.py -test -i $CORPORA/train/wsj_train.conll06 -m $MODELS/m_e-10_da_ss -o predicted.conll06
 
 #python -u dependency-parser.py -ev -i predicted.conll06 -g $CORPORA/dev/wsj_dev.conll06 -o evaluation_sentence.txt
 

start-cle_automated.sh

@@ -7,7 +7,7 @@ PREDICTIONS="predictions"
 EVALUATIONS="evaluations"
 
 #train
-python -u dependency-parser.py -train -i ${CORPORA}/train/wsj_train.conll06 -m ${MODELS}/$2 -e $1 $5 $6
+#python -u dependency-parser.py -train -i ${CORPORA}/train/wsj_train.conll06 -m ${MODELS}/$2 -e $1 $5 $6
 
 #test
 python -u dependency-parser.py -test -i ${CORPORA}/dev/wsj_dev.conll06 -m ${MODELS}/$2 -o ${PREDICTIONS}/$3".conll06"

start-cle_automated_german.sh

@@ -0,0 +1,19 @@
+#!/bin/bash
+
+CORPORA="../dependency-parsing-files/data/german"
+
+MODELS="models"
+PREDICTIONS="predictions"
+EVALUATIONS="evaluations"
+
+#train
+python -u dependency-parser.py -train -i ${CORPORA}/train/tiger-2.2.train.conll06 -m ${MODELS}/$2 -e $1 $5 $6
+
+#test
+python -u dependency-parser.py -test -i ${CORPORA}/dev/tiger-2.2.dev.conll06 -m ${MODELS}/$2 -o ${PREDICTIONS}/$3".conll06"
+
+# sentence based evaluation
+python -u dependency-parser.py -ev -i ${PREDICTIONS}/$3".conll06" -g ${CORPORA}/dev/tiger-2.2.dev.conll06 -o ${EVALUATIONS}/$4"_sentence".txt
+
+#evaluate
+./eval07.pl -g ${CORPORA}/dev/tiger-2.2.dev.conll06 -s ${PREDICTIONS}/$3".conll06" >> ${EVALUATIONS}/$4".txt" 2>&1

test-dependency-parser.py

@@ -3,7 +3,7 @@
 
 from copy import deepcopy
 
-"""
+
 a = ManualSparseGraph()
 a.add_arc(0, SparseArc(0, 1, 10.0))
 a.add_arc(0, SparseArc(0, 2, 9.0))
@@ -46,7 +46,7 @@
     for arc in y[head]:
         print str(head) + " --> " + str(arc.dependent) + ", " + str(arc.score)
 print "--"
-
+"""
 c = ManualSparseGraph()
 c.add_arc(33, SparseArc(33, 32, 0))
 c.add_arc(4, SparseArc(4, 3, 0))
@@ -85,7 +85,7 @@
 c.add_arc(29, SparseArc(29, 30, 0))
 
 print cycle(c.heads)
-"""
+
 d = ManualSparseGraph()
 # d.add_arc(0, SparseArc(0, 7, 0))
 d.add_arc(16, SparseArc(16, 1, 0))
@@ -112,4 +112,5 @@
 d.add_arc(20, SparseArc(20, 16, 0))
 d.add_arc(7, SparseArc(7, 22, 0))
 
-print cycle(d.heads)
+print cycle(d.heads)
+"""