s437622 kenlm

run.py

@@ -1,73 +1,118 @@
-from nltk import trigrams, word_tokenize
-from collections import defaultdict, Counter
+#!/usr/bin/env python
+# coding: utf-8
+
+# In[1]:
+
+
 import pandas as pd
 import csv
 import regex as re
+import kenlm
+from english_words import english_words_alpha_set
+from math import log10
+from nltk import trigrams, word_tokenize
+
+
+# In[2]:
+
 
 default_pred = 'to:0.02 be:0.02 the:0.02 or:0.01 not:0.01 and:0.01 a:0.01 :0.9'
 
 
+# In[3]:
+
+
 def preprocess(text):
     text = text.lower().replace('-\\n', '').replace('\\n', ' ')
     return re.sub(r'\p{P}', '', text)
 
 
-class Model():
-    def __init__(self, alpha, test_file_name):
-        train_data = pd.read_csv('train/in.tsv.xz', sep='\t', on_bad_lines='skip', header=None, quoting=csv.QUOTE_NONE,
-                                 nrows=20000)
-        train_labels = pd.read_csv('train/expected.tsv', sep='\t', on_bad_lines='skip', header=None,
-                                   quoting=csv.QUOTE_NONE, nrows=20000)
-        train_data = train_data[[6, 7]]
-        train_data = pd.concat([train_data, train_labels], axis=1)
-        train_data['line'] = train_data[6] + train_data[0] + train_data[7]
+# In[4]:
 
-        self.file = train_data[['line']]
-        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):
-            text = preprocess(str(row['line']))
-            words = word_tokenize(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):
-            count = 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(count + self.alpha + len(word_2))
+train_data = pd.read_csv('train/in.tsv.xz', sep='\t', on_bad_lines='skip', header=None, quoting=csv.QUOTE_NONE,
+                         nrows=20000)
+train_labels = pd.read_csv('train/expected.tsv', sep='\t', on_bad_lines='skip', header=None,
+                           quoting=csv.QUOTE_NONE, nrows=20000)
+data = pd.concat([train_data, train_labels], axis=1)
+data=train_data[6] + train_data[0] + train_data[7]
+data = data.apply(preprocess)
 
-    def predict(self, before, after):
-        prediction = dict(Counter(dict(self.model[before, after])).most_common(5))
-        result = []
-        prob = 0.0
-        for key, value in prediction.items():
-            prob += value
-            result.append(f'{key}:{value} ')
-        if prob == 0.0:
-            return default_pred
-        result.append(f':{max(1 - prob, 0.01)}')
-        return ''.join(result)
+with open("train_file.txt", "w+") as f:
+    for text in data:
+        f.write(text + "\n")
 
-    def make_prediction(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:
-            for _, row in data.iterrows():
-                before, after = word_tokenize(preprocess(str(row[6]))), word_tokenize(preprocess(str(row[7])))
-                if len(before) < 3 or len(after) < 3:
-                    prediction = default_pred
+
+# In[5]:
+
+
+KENLM_BUILD_PATH='../kenlm/kenlm/build'
+get_ipython().system('$KENLM_BUILD_PATH/bin/lmplz -o 4 < train_file.txt > model.arpa')
+get_ipython().system('rm train_file.txt')
+
+
+# In[6]:
+
+
+model = kenlm.Model("model.arpa")
+
+
+# In[7]:
+
+
+def predict(before, after):
+    best_scores = []
+    for word in english_words_alpha_set:
+        text = ' '.join([before, word, after])
+        text_score = model.score(text, bos=False, eos=False)
+        if len(best_scores) < 12:
+            best_scores.append((word, text_score))
+        else:
+            is_better = False
+            worst_score = None
+            for score in best_scores:
+                if not worst_score:
+                    worst_score = score
                 else:
-                    prediction = self.predict(before[-1], after[0])
-                file_out.write(prediction + '\n')
+                    if worst_score[1] > score[1]:
+                        worst_score = score
+            if worst_score[1] < text_score:
+                best_scores.remove(worst_score)
+                best_scores.append((word, text_score))
+    probs = sorted(best_scores, key=lambda tup: tup[1], reverse=True)
+    pred_str = ''
+    for word, prob in probs:
+        pred_str += f'{word}:{prob} '
+    pred_str += f':{log10(0.99)}'
+    return pred_str
+
+
+# In[8]:
+
+
+def make_prediction(path, result_path):
+    data = pd.read_csv(path, sep='\t', on_bad_lines='skip', header=None, quoting=csv.QUOTE_NONE)
+    with open(result_path, 'w', encoding='utf-8') as file_out:
+        for _, row in data.iterrows():
+            before, after = word_tokenize(preprocess(str(row[6]))), word_tokenize(preprocess(str(row[7])))
+            if len(before) < 2 or len(after) < 2:
+                prediction = default_pred
+            else:
+                prediction = predict(before[-1], after[0])
+            file_out.write(prediction + '\n')
+
+
+# In[9]:
+
+
+make_prediction("dev-0/in.tsv.xz", "dev-0/out.tsv")
+
+
+# In[10]:
+
+
+make_prediction("test-A/in.tsv.xz", "test-A/out.tsv")
+
+
 
 
-alpha = 0.1
-model = Model(alpha, 'test-A')
-model.train()
-model.make_prediction()