470628 kenlm

run.py

@@ -1,10 +1,14 @@
 from nltk.tokenize import word_tokenize
 from nltk import trigrams
 from collections import defaultdict, Counter
+from english_words import english_words_alpha_set
 import pandas as pd
 import csv
 import regex as re
 import sys
+import os
+import kenlm
+from math import log10
 
 
 DEFAULT_PREDICTION = 'the:0.2 be:0.2 to:0.2 of:0.1 and:0.1 a:0.1 :0.1'
@@ -15,54 +19,41 @@ def preprocess(text):
 
 
 
-class Model():
-    def __init__(self, alpha, train_file_name, test_file_name):
+class TrainModel():
+    def __init__(self, train_file_name):
         
-        file_expected = pd.read_csv(f'{train_file_name}/expected.tsv', sep='\t', on_bad_lines='skip', header=None, quoting=csv.QUOTE_NONE, nrows=200000)
-        file_in = pd.read_csv(f'{train_file_name}/in.tsv.xz', sep='\t', on_bad_lines='skip', header=None, quoting=csv.QUOTE_NONE, nrows=200000)
+        file_expected = pd.read_csv(f'{train_file_name}/expected.tsv', sep='\t', on_bad_lines='skip', header=None, quoting=csv.QUOTE_NONE)
+        file_in = pd.read_csv(f'{train_file_name}/in.tsv.xz', sep='\t', on_bad_lines='skip', header=None, quoting=csv.QUOTE_NONE)
         file_in = file_in[[6, 7]]
         file_concat = pd.concat([file_in, file_expected], axis=1)
         file_concat['text'] = file_concat[6] + file_concat[0] + file_concat[7]
-        
         self.file = file_concat[['text']]
-        self.test_file_name = test_file_name
-        self.alpha = alpha;
-        self.model = defaultdict(lambda: defaultdict(lambda: 0))
         
-    def train(self):
-        rows = self.file.iterrows()
-        rows_len = len(self.file)
-        for index, (_, row) in enumerate(rows):
-            if index % 1000 == 0:
-                print(f'uczenie modelu: {index / rows_len}')
-            words = word_tokenize(preprocess(str(row['text'])))
-            for word_1, word_2, word_3 in trigrams(words, pad_right=True, pad_left=True):
-                if word_1 and word_2 and word_3:
-                    self.model[(word_1, word_3)][word_2] += 1
-        model_len = len(self.model)
-        for index, words_1_3 in enumerate(self.model):
-            if index % 100000 == 0:
-                print(f'normalizacja i wygładzanie: {index / model_len}')
-            occurrences = sum(self.model[words_1_3].values())
-            for word_2 in self.model[words_1_3]:
-                self.model[words_1_3][word_2] += self.alpha
-                self.model[words_1_3][word_2] /= float(occurrences + self.alpha + len(word_2))
+    
+    def build_kenlm_model(self):
+        with open('text.txt', 'w+', encoding='utf-8') as file:
+            for _, row in self.file.iterrows():
+                file.write(str(row['text']) + '\n')
+        os.system('./kenlm/build/bin/lmplz -o 4 < text.txt > model.arpa --skip_symbols')
+        os.system('./kenlm/build/bin/build_binary model.arpa model.binary')
+        os.system('rm text.txt')
+        self.kenlm_model = kenlm.Model('model.binary')
     
     def predict_row(self, word_before, word_after):
-        prediction = dict(Counter(dict(self.model[word_before, word_after])).most_common(6))
-        result = []
-        prob = 0.0
-        for key, value in prediction.items():
-            prob += value
-            result.append(f'{key}:{value}')
-        if prob == 0.0:
-            return DEFAULT_PREDICTION
-        result.append(f':{max(1 - prob, 0.01)}')
-        return ' '.join(result)
+        scores = {}
+        for word in english_words_alpha_set:
+            score = self.kenlm_model.score(word_before + ' ' + word +  ' ' + word_after, bos=False, eos=False)
+            scores[word] = score
+        prediction = ''
+        top_probs = Counter(scores).most_common(6)
+        for word, prob in top_probs:
+            prediction += f'{word}:{prob} '
+        prediction += f':{log10(0.01)}'
+        return prediction
     
-    def predict(self):
-        data = pd.read_csv(f'{self.test_file_name}/in.tsv.xz', sep='\t', on_bad_lines='skip', header=None, quoting=csv.QUOTE_NONE)
-        with open(f'{self.test_file_name}/out.tsv', 'w', encoding='utf-8') as file_out:
+    def predict(self, test_file_name):
+        data = pd.read_csv(f'{test_file_name}/in.tsv.xz', sep='\t', on_bad_lines='skip', header=None, quoting=csv.QUOTE_NONE)
+        with open(f'{test_file_name}/out.tsv', 'w', encoding='utf-8') as file_out:
             for _, row in data.iterrows():
                 words_before, words_after = word_tokenize(preprocess(str(row[6]))), word_tokenize(preprocess(str(row[7])))
                 if len(words_before) < 3 or len(words_after) < 3:
@@ -73,8 +64,9 @@ class Model():
         
 
 
-alpha = float(sys.argv[1])
-print(f'alfa: {alpha}')
-model = Model(alpha, 'train', sys.argv[2])
-model.train()
-model.predict()
+training = TrainModel('train')
+training.build_kenlm_model()
+training.predict('test-A')
+training.predict('dev-0')
+
+