[FEAT] Adding new feature (core python only) for diachronic specificity.

analysis/diachronic_specificity.py

+import sqlalchemy
+from gargantext_web import api
+from node import models
+from sqlalchemy import create_engine
+from sqlalchemy.sql import func
+import numpy as np
+import collections
+
+NodeType = models.NodeType.sa
+NodeNgram = models.Node_Ngram.sa
+NodeNodeNgram = models.NodeNgramNgram.sa
+Ngram = models.Ngram.sa
+Node = models.Node.sa
+Corpus = models.Corpus.sa
+
+def get_session():
+    import sqlalchemy.orm
+    from django.db import connections
+    from sqlalchemy.orm import sessionmaker
+    from aldjemy.core import get_engine
+    alias = 'default'
+    connection = connections[alias]
+    engine = create_engine("postgresql+psycopg2://alexandre:C8kdcUrAQy66U@localhost/gargandb",
+            use_native_hstore=True)
+    Session = sessionmaker(bind=engine)
+    return Session()
+
+session = get_session()
+
+
+def result2dict(query):
+    results = dict()
+    for result in query:
+        if result[0] is not None:
+            results[result[0]] = result[1]
+    return(results)
+
+
+def diachronic_specificity(corpus_id, string, order=True):
+    ''' 
+    Take as parameter Corpus primary key and text of ngrams.
+    Result is a dictionnary.
+    Keys are period (years for now)
+    Values are measure to indicate diachronic specificity.
+    Nowadays, the measure is rather simple: distance of frequency of period from mean of frequency of all corpus.
+    '''
+    corpus = session.query(Node).get(int(corpus_id))
+    ngram = session.query(Ngram).filter(Ngram.terms == string).first()
+    
+    ngram_frequency_query = session.query(Node.metadata['publication_year'], func.count('*'))                        .join(NodeNgram, Node.id == NodeNgram.node_id)                        .filter( NodeNgram.ngram == ngram)                        .filter(Node.parent_id == corpus.id)                        .group_by(Node.metadata['publication_year']).all()
+
+    document_year_sum_query = session.query(Node.metadata['publication_year'], func.count('*'))            .filter(Node.parent_id == corpus.id)            .group_by(Node.metadata['publication_year']).all()
+            
+            
+    document_filterByngram_year = result2dict(ngram_frequency_query)
+    document_all_year = result2dict(document_year_sum_query)
+    #print(document_all_year)
+    data = dict()
+    
+    for year in document_all_year.keys():
+       data[year] = document_filterByngram_year.get(year, 0) / document_all_year[year]
+    
+    mean = np.mean(list(data.values()))
+    
+    data_dict = dict(zip(data.keys(), list(map(lambda x: x - mean, data.values()))))
+    
+    if order == True:
+        return collections.OrderedDict(sorted(data_dict.items()))
+    else:
+        return data_dict
+
+
+# For tests
+#diachronic_specificity(102750, "bayer", order=True)
+
+