solution by 442333 based on kenlm

.gitignore

@@ -0,0 +1,8 @@
+
+*~
+*.swp
+*.bak
+*.pyc
+*.o
+.DS_Store
+.token

README.md

@@ -0,0 +1,9 @@
+Challenging America word-gap prediction
+===================================
+
+Guess a word in a gap.
+
+Evaluation metric
+-----------------
+
+LikelihoodHashed is the metric

config.txt

@@ -0,0 +1 @@
+--metric PerplexityHashed --precision 2  --in-header in-header.tsv  --out-header out-header.tsv

in-header.tsv

@@ -0,0 +1 @@
+FileId	Year	LeftContext	RightContext

out-header.tsv

@@ -0,0 +1 @@
+Word

run.py

@@ -0,0 +1,84 @@
+import multiprocessing as mp
+import nltk
+from tqdm import tqdm
+from functools import partial
+import kenlm
+import regex as re
+from tqdm import tqdm
+from collections import Counter
+
+from english_words import get_english_words_set
+words = get_english_words_set(['web2'], lower=True, alpha=True)
+
+path = 'model_5.binary'
+
+language_model = kenlm.Model(path)
+
+def clean(text):
+	text = text.replace('-\\n', '').replace('\\n', ' ').replace('\\t', ' ').replace('<s>','s')
+	while '  ' in text:
+		text = text.replace('  ',' ')
+
+	return re.sub(r'\p{P}', '', text)
+
+def generate_file(input_path, expected_path, output_path):
+	with open(input_path) as input_file, open(expected_path) as expected_file, open(output_path, 'w', encoding='utf-8') as output_file:
+		for line, word in zip(input_file, expected_file):
+			columns = line.split('\t')
+			prefix = clean(columns[6])
+			suffix = clean(columns[7])
+
+			train_line = f"{prefix.strip()} {word.strip()} {suffix.strip()}"
+
+			output_file.write(train_line)
+
+#generate_file('train/in.tsv', 'train/expected.tsv', 'train/train.txt')
+
+def predict(prefix):
+	scores = {}
+
+	for word in words:
+		candidate = f"{prefix} {word}".strip()
+		score = language_model.score(candidate, bos=False, eos=False)
+
+		score_step_lower = language_model.score(f"{prefix.strip()}", bos=False, eos=False)
+
+		scores[word] = score - score_step_lower
+
+	highest_probs = Counter(scores).most_common(10)
+
+	output = ''
+	probs = 0
+	for word, logprob in highest_probs:
+		prob = 10 ** logprob
+		probs += prob
+		output += f"{word}:{prob} "
+	output += f":{1 - probs}"
+
+	return output
+
+def parse_line(line):
+    columns = line.split('\t')
+    prefix = clean(columns[6])
+    
+    prefix = nltk.tokenize.word_tokenize(prefix)
+    
+    prefix_input = prefix[-4] + " " + prefix[-3] + " " + prefix[-2] + " " + prefix[-1]
+    
+    result = predict(prefix_input)
+    
+    return result
+
+def parse(input_path, output_path='out.tsv'):
+    with open(input_path) as f:
+        lines = f.readlines()
+    
+    with open(output_path, 'w', encoding="utf-8") as output_file:
+        pool = mp.Pool()
+        
+        results = list(tqdm(pool.imap(parse_line, lines), total=len(lines)))
+        
+        for result in results:
+            output_file.write(result + '\n')
+
+parse('test-A/in.tsv', output_path="test-A/out.tsv")

train.sh

@@ -0,0 +1,2 @@
+kenlm-master/build/bin/lmplz -o 4 --skip_symbols -T temp < train/train.txt > model_train_4.arpa
+kenlm-master/build/bin/build_binary -s model_train_4.arpa model_train_4.binary