Merge pull request #338 from NervanaSystems/alon/master_local

Alon/master local

Author: peteriz

examples/cross_doc_coref/cross_doc_coref_sieves.py

@@ -15,8 +15,11 @@
 # ******************************************************************************
 
 import logging
+from typing import List
 
 from nlp_architect import LIBRARY_ROOT
+from nlp_architect.common.cdc.cluster import Clusters
+from nlp_architect.common.cdc.topics import Topics
 from nlp_architect.data.cdc_resources.relations.relation_types_enums import RelationType
 from nlp_architect.models.cross_doc_coref.cdc_config import EventConfig, EntityConfig
 from nlp_architect.models.cross_doc_coref.cdc_resource import CDCResources
@@ -25,7 +28,24 @@
 from nlp_architect.models.cross_doc_sieves import run_event_coref, run_entity_coref
 
 
-def run_example():
+def run_example(cdc_settings):
+    event_mentions = Topics(LIBRARY_ROOT + '/datasets/ecb/ecb_all_event_mentions.json')
+
+    event_clusters = None
+    if cdc_settings.event_config.run_evaluation:
+        logger.info('Running event coreference resolution')
+        event_clusters = run_event_coref(event_mentions, cdc_settings)
+
+    entity_mentions = Topics(LIBRARY_ROOT + '/datasets/ecb/ecb_all_entity_mentions.json')
+    entity_clusters = None
+    if cdc_settings.entity_config.run_evaluation:
+        logger.info('Running entity coreference resolution')
+        entity_clusters = run_entity_coref(entity_mentions, cdc_settings)
+
+    return event_clusters, entity_clusters
+
+
+def create_example_settings():
     event_config = EventConfig()
     event_config.sieves_order = [
         (SieveType.STRICT, RelationType.SAME_HEAD_LEMMA, 0.0),
@@ -34,11 +54,7 @@ def run_example():
         (SieveType.RELAX, RelationType.SAME_HEAD_LEMMA_RELAX, 0.5),
     ]
 
-    event_config.gold_mentions_file = LIBRARY_ROOT + \
-        '/datasets/ecb/ecb_all_event_mentions.json'
-
     entity_config = EntityConfig()
-
     entity_config.sieves_order = [
         (SieveType.STRICT, RelationType.SAME_HEAD_LEMMA, 0.0),
         (SieveType.VERY_RELAX, RelationType.WIKIPEDIA_REDIRECT_LINK, 0.1),
@@ -47,44 +63,42 @@ def run_example():
         (SieveType.VERY_RELAX, RelationType.REFERENT_DICT, 0.5)
     ]
 
-    entity_config.gold_mentions_file = LIBRARY_ROOT + \
-        '/datasets/ecb/ecb_all_entity_mentions.json'
-
     # CDCResources hold default attribute values that might need to be change,
     # (using the defaults values in this example), use to configure attributes
     # such as resources files location, output directory, resources init methods and other.
     # check in class and see if any attributes require change in your set-up
     resource_location = CDCResources()
-    resources = CDCSettings(resource_location, event_config, entity_config)
+    return CDCSettings(resource_location, event_config, entity_config)
 
-    event_clusters = None
-    if event_config.run_evaluation:
-        logger.info('Running event coreference resolution')
-        event_clusters = run_event_coref(resources)
 
-    entity_clusters = None
-    if entity_config.run_evaluation:
-        logger.info('Running entity coreference resolution')
-        entity_clusters = run_entity_coref(resources)
+def print_results(clusters: List[Clusters], type: str):
+    print('-=' + type + ' Clusters=-')
+    for topic_cluster in clusters:
+        print('\n\tTopic=' + topic_cluster.topic_id)
+        for cluster in topic_cluster.clusters_list:
+            cluster_mentions = list()
+            for mention in cluster.mentions:
+                mentions_dict = dict()
+                mentions_dict['id'] = mention.mention_id
+                mentions_dict['text'] = mention.tokens_str
+                cluster_mentions.append(mentions_dict)
+
+            print('\t\tCluster(' + str(cluster.coref_chain) + ') Mentions='
+                  + str(cluster_mentions))
 
-    print('-=Cross Document Coref Results=-')
-    print('-=Event Clusters Mentions=-')
-    for event_cluster in event_clusters.clusters_list:
-        print(event_cluster.coref_chain)
-        for event_mention in event_cluster.mentions:
-            print(event_mention.mention_id)
-            print(event_mention.tokens_str)
 
-    print('-=Entity Clusters Mentions=-')
-    for entity_cluster in entity_clusters.clusters_list:
-        print(entity_cluster.coref_chain)
-        for entity_mention in entity_cluster.mentions:
-            print(entity_mention.mention_id)
-            print(entity_mention.tokens_str)
+def run_cdc_pipeline():
+    cdc_settings = create_example_settings()
+    event_clusters, entity_clusters = run_example(cdc_settings)
+
+    print('-=Cross Document Coref Results=-')
+    print_results(event_clusters, 'Event')
+    print('################################')
+    print_results(entity_clusters, 'Entity')
 
 
 if __name__ == '__main__':
     logging.basicConfig(level=logging.INFO)
     logger = logging.getLogger(__name__)
 
-    run_example()
+    run_cdc_pipeline()

nlp_architect/common/cdc/cluster.py

@@ -65,14 +65,15 @@ def get_cluster_id(self) -> str:
 class Clusters(object):
     cluster_coref_chain = 1000
 
-    def __init__(self, mentions: List[MentionData] = None) -> None:
+    def __init__(self, topic_id: str, mentions: List[MentionData] = None) -> None:
         """
 
         Args:
             mentions: ``list[MentionData]``, required
                 The initial mentions to create the clusters from
         """
         self.clusters_list = []
+        self.topic_id = topic_id
         self.set_initial_clusters(mentions)
 
     def set_initial_clusters(self, mentions: List[MentionData]) -> None:

nlp_architect/common/cdc/topics.py

@@ -37,9 +37,9 @@ def __init__(self, mentions_file_path: str) -> None:
         Args:
             mentions_file_path: this topic mentions json file
         """
-        self.topics_list = self.load_gold_mentions(mentions_file_path)
+        self.topics_list = self.load_gold_mentions_from_file(mentions_file_path)
 
-    def load_gold_mentions(self, mentions_file_path: str) -> List[Topic]:
+    def load_gold_mentions_from_file(self, mentions_file_path: str) -> List[Topic]:
         start_data_load = time.time()
         logger.info('Loading mentions from-%s', mentions_file_path)
         mentions = load_json_file(mentions_file_path)

nlp_architect/data/cdc_resources/wikipedia/wiki_online.py

@@ -15,6 +15,7 @@
 # ******************************************************************************
 
 import os
+import logging
 
 from nlp_architect.data.cdc_resources.data_types.wiki.wikipedia_page import WikipediaPage
 from nlp_architect.data.cdc_resources.data_types.wiki.wikipedia_page_extracted_relations import \
@@ -27,6 +28,8 @@
 DISAMBIGUATE_PAGE = ['wikimedia disambiguation page', 'wikipedia disambiguation page']
 NAME_DESCRIPTIONS = ['given name', 'first name', 'family name']
 
+logger = logging.getLogger(__name__)
+
 
 class WikiOnline(object):
     def __init__(self):
@@ -52,7 +55,7 @@ def get_pages(self, phrase):
                     full_page = self.get_wiki_page_with_items(phrase, page_result)
                     ret_pages.add(WikipediaSearchPageResult(appr, full_page))
             except Exception as e:
-                print(e)
+                logger.error(e)
 
         self.cache[phrase] = ret_pages
         return ret_pages
@@ -73,7 +76,7 @@ def get_wiki_page_with_items(self, phrase, page):
 
         ret_page = WikipediaPage(phrase, None, page_title, None, 0, pageid, description, relations)
 
-        print('Page:' + str(ret_page) + ". Extracted successfully")
+        logger.debug('Page:' + str(ret_page) + ". Extracted successfully")
 
         return ret_page
 

nlp_architect/models/cross_doc_coref/cdc_config.py

@@ -15,7 +15,6 @@
 # ******************************************************************************
 from typing import List, Tuple
 
-from nlp_architect import LIBRARY_ROOT
 from nlp_architect.data.cdc_resources.relations.relation_types_enums import RelationType
 from nlp_architect.models.cross_doc_coref.system.sieves.sieves import SieveType
 
@@ -26,20 +25,19 @@ def __init__(self):
 
         self.__sieves_order = None
         self.__run_evaluation = False
-        self.__gold_mentions_file = None
 
     @property
     def sieves_order(self):
         """
         Sieve definition and Sieve running order
 
-            Tuple[SieveType, RelationType, Threshold(float)] - define sieves to run, were
+        Tuple[SieveType, RelationType, Threshold(float)] - define sieves to run, were
 
-            Strict- Merge clusters only in case all mentions has current relation between them,
-            Relax- Merge clusters in case (matched mentions) / len(cluster_1.mentions)) >= thresh,
-            Very_Relax- Merge clusters in case (matched mentions) / (all possible pairs) >= thresh
+        Strict- Merge clusters only in case all mentions has current relation between them,
+        Relax- Merge clusters in case (matched mentions) / len(cluster_1.mentions)) >= thresh,
+        Very_Relax- Merge clusters in case (matched mentions) / (all possible pairs) >= thresh
 
-            RelationType represent the type of sieve to run.
+        RelationType represent the type of sieve to run.
 
         """
         return self.__sieves_order
@@ -57,15 +55,6 @@ def run_evaluation(self):
     def run_evaluation(self, run_evaluation: bool):
         self.__run_evaluation = run_evaluation
 
-    @property
-    def gold_mentions_file(self):
-        """Mentions file to run against"""
-        return self.__gold_mentions_file
-
-    @gold_mentions_file.setter
-    def gold_mentions_file(self, gold_file):
-        self.__gold_mentions_file = gold_file
-
 
 class EventConfig(CDCConfig):
     def __init__(self):
@@ -91,9 +80,6 @@ def __init__(self):
             (SieveType.STRICT, RelationType.WORDNET_DERIVATIONALLY, 0.0)
         ]
 
-        self.gold_mentions_file = LIBRARY_ROOT + \
-            '/datasets/ecb/ecb_all_event_mentions.json'
-
 
 class EntityConfig(CDCConfig):
     def __init__(self):
@@ -118,6 +104,3 @@ def __init__(self):
             (SieveType.STRICT, RelationType.WORDNET_SAME_SYNSET_ENTITY, 0.0),
             (SieveType.VERY_RELAX, RelationType.REFERENT_DICT, 0.5)
         ]
-
-        self.gold_mentions_file = LIBRARY_ROOT + \
-            '/datasets/ecb/ecb_all_entity_mentions.json'

nlp_architect/models/cross_doc_coref/system/cdc_settings.py

@@ -16,7 +16,6 @@
 
 import logging
 
-from nlp_architect.common.cdc.topics import Topics
 from nlp_architect.data.cdc_resources.relations.computed_relation_extraction import \
     ComputedRelationExtraction
 from nlp_architect.data.cdc_resources.relations.referent_dict_relation_extraction import \
@@ -44,21 +43,12 @@ def __init__(self, resources, event_coref_config, entity_coref_config):
         self.context2vec_model = None
         self.wordnet = None
         self.within_doc = None
-        self.events_topics = None
-        self.entity_topics = None
         self.event_config = event_coref_config
         self.entity_config = entity_coref_config
         self.cdc_resources = resources
 
         self.load_modules()
 
-        if event_coref_config.run_evaluation:
-            self.events_topics = Topics(event_coref_config.gold_mentions_file)
-        if entity_coref_config.run_evaluation:
-            self.entity_topics = Topics(entity_coref_config.gold_mentions_file)
-        if not self.events_topics and not self.entity_topics:
-            raise Exception('No entity or events Gold topics loaded!')
-
     def load_modules(self):
         relations = set()
         for sieve in self.event_config.sieves_order:

nlp_architect/models/cross_doc_coref/system/sieves/run_sieve_system.py

@@ -18,6 +18,8 @@
 import time
 
 from nlp_architect.common.cdc.cluster import Clusters
+from nlp_architect.common.cdc.topics import Topic
+from nlp_architect.models.cross_doc_coref.system.cdc_settings import CDCSettings
 from nlp_architect.models.cross_doc_coref.system.sieves.sieves import get_sieve
 
 logger = logging.getLogger(__name__)
@@ -29,7 +31,7 @@ def __init__(self, topic):
         self.results_dict = dict()
         self.results_ordered = []
         logger.info('loading topic %s, total mentions: %d', topic.topic_id, len(topic.mentions))
-        self.clusters = Clusters(topic.mentions)
+        self.clusters = Clusters(topic.topic_id, topic.mentions)
 
     @staticmethod
     def set_sieves_from_config(config, get_rel_extraction):
@@ -90,3 +92,12 @@ def __init__(self, topic, resources):
         super(RunSystemsEvent, self).__init__(topic)
         self.sieves = self.set_sieves_from_config(resources.event_config,
                                                   resources.get_module_from_relation)
+
+
+def get_run_system(topic: Topic, resource: CDCSettings, eval_type: str):
+    if eval_type.lower() == 'entity':
+        return RunSystemsEntity(topic, resource)
+    elif eval_type.lower() == 'event':
+        return RunSystemsEvent(topic, resource)
+    else:
+        raise AttributeError(eval_type + ' Not supported!')