278 move mask (#283)

* resolves #278

scripts/pipeline.py

@@ -1,10 +1,8 @@
 import bz2
 import pickle
-from os import makedirs, path
 
+from os import makedirs, path
 from pandas import read_pickle
-from tqdm import tqdm
-
 import scripts.data_factory as datafactory
 import scripts.output_factory as output_factory
 import scripts.utils.date_utils
@@ -19,6 +17,7 @@
 from scripts.vandv.emergence_labels import map_prediction_to_emergence_label, report_predicted_emergence_labels_html
 from scripts.vandv.graphs import report_prediction_as_graphs_html
 from scripts.vandv.predictor import evaluate_prediction
+from tqdm import tqdm
 
 
 class Pipeline(object):
@@ -43,6 +42,8 @@ def __init__(self, data_filename, docs_mask_dict, pick_method='sum', ngram_range
                                                ngram_range=ngram_range,
                                                max_document_frequency=max_df,
                                                tokenizer=LemmaTokenizer())
+            tfidf_mask_obj = TfidfMask(self.__tfidf_obj, ngram_range=ngram_range, uni_factor=0.8, unbias=True)
+            self.__tfidf_obj.apply_weights(tfidf_mask_obj.tfidf_mask)
 
             if prefilter_terms != 0:
                 tfidf_reduce_obj = TfidfReduce(self.__tfidf_obj.tfidf_matrix, self.__tfidf_obj.feature_names)

scripts/tfidf_mask.py

@@ -2,7 +2,7 @@
 
 
 class TfidfMask(object):
-    def __init__(self, tfidf_obj, ngram_range=(2, 3), uni_factor=0.8):
+    def __init__(self, tfidf_obj, ngram_range=(2, 3), uni_factor=0.8, unbias=False):
         self.__tfidf_matrix = tfidf_obj.tfidf_matrix
         self.__feature_names = tfidf_obj.feature_names
         self.__tfidf_mask = self.__tfidf_matrix.copy()
@@ -11,12 +11,13 @@ def __init__(self, tfidf_obj, ngram_range=(2, 3), uni_factor=0.8):
         self.__uni_factor = uni_factor
         self.__idf = tfidf_obj.idf
 
-        # do unigrams
-        if ngram_range[0] == 1:
-            self.__clean_unigrams(self.__max_bigram())
+        if unbias:
+            # do unigrams
+            if ngram_range[0] == 1:
+                self.__clean_unigrams(self.__max_bigram())
 
-        for i in range(ngram_range[0], ngram_range[1]):
-            self.__unbias_ngrams(i + 1)
+            for i in range(ngram_range[0], ngram_range[1]):
+                self.__unbias_ngrams(i + 1)
 
     @property
     def tfidf_mask(self):

scripts/tfidf_wrapper.py

@@ -56,6 +56,11 @@ def __trigger_transformer(self):
             self.__l2_norm = utils.l2normvec(self.__tfidf_matrix)
         self.__tfidf_matrix = utils.apply_l2normvec(self.__tfidf_matrix, self.__l2_norm)
 
+    def apply_weights(self, weights_matrix):
+        self.__count_matrix = self.__count_matrix.multiply(weights_matrix)
+        self.__tfidf_matrix = self.__tfidf_matrix.multiply(weights_matrix)
+        self.__count_matrix.data = np.array([np.uint8(round(x)) for x in self.__count_matrix.data])
+
     @property
     def l2_norm(self):
         return self.__l2_norm
@@ -82,4 +87,5 @@ def vocabulary(self):
 
     @property
     def feature_names(self):
-        return self.__feature_names
+        return self.__feature_names
+

tests/test_pygrams.py

@@ -399,9 +399,9 @@ def test_unibitri_reduction_output_termcounts(self, mock_path_isfile, mock_maked
         pygrams.main(args)
 
         def assert_outputs(term_counts_per_week, feature_names, number_of_documents_per_week, week_iso_dates):
-            self.assertListEqual(feature_names, ['abstract', 'extra', 'extra stuff', 'patent', 'stuff', 'with'])
+            self.assertListEqual(feature_names, ['abstract',  'extra stuff', 'patent', 'with'])
             term_counts_as_lists = term_counts_per_week.todense().tolist()
-            self.assertListEqual(term_counts_as_lists[0], [1, 0, 1, 1, 0, 1])
+            self.assertListEqual(term_counts_as_lists[0], [1,  1, 1,  1])
             self.assertListEqual(number_of_documents_per_week, [1])
             self.assertListEqual(week_iso_dates, [200052])
 

tests/test_terms_graph.py

@@ -33,7 +33,7 @@ def setUpClass(cls):
         filter_output_obj = FilterTerms(tfidf_obj.feature_names, None, None)
         term_weights = filter_output_obj.ngram_weights_vec
 
-        tfidf_mask_obj = TfidfMask(tfidf_obj, ngram_range=ngram_range)
+        tfidf_mask_obj = TfidfMask(tfidf_obj, ngram_range=ngram_range, unbias=True)
         tfidf_mask_obj.update_mask(doc_weights, term_weights)
         tfidf_mask = tfidf_mask_obj.tfidf_mask
 

tests/test_tfidf_mask.py

@@ -55,7 +55,7 @@ def init_mask(self, cpc, min_n, uni_factor=0.8):
         filter_output_obj = FilterTerms(self.__tfidf_obj.feature_names, None)
         term_weights = filter_output_obj.ngram_weights_vec
 
-        tfidf_mask_obj = TfidfMask(self.__tfidf_obj, ngram_range=(min_n, self.__max_n), uni_factor=uni_factor)
+        tfidf_mask_obj = TfidfMask(self.__tfidf_obj, ngram_range=(min_n, self.__max_n), uni_factor=uni_factor, unbias=True)
         tfidf_mask_obj.update_mask(doc_filters, term_weights)
         self.__tfidf_mask = tfidf_mask_obj.tfidf_mask