285 lines
8.3 KiB
Plaintext
285 lines
8.3 KiB
Plaintext
{
|
|
"cells": [
|
|
{
|
|
"cell_type": "code",
|
|
"execution_count": 1,
|
|
"metadata": {},
|
|
"outputs": [],
|
|
"source": [
|
|
"import lzma\n",
|
|
"import pickle\n",
|
|
"from collections import Counter "
|
|
]
|
|
},
|
|
{
|
|
"cell_type": "code",
|
|
"execution_count": 2,
|
|
"metadata": {},
|
|
"outputs": [],
|
|
"source": [
|
|
"def clean_line(line):\n",
|
|
" prefix = line.split('\\t')[6].replace(r'\\n', ' ')\n",
|
|
" suffix = line.split('\\t')[7].replace(r'\\n', ' ')\n",
|
|
" return f'{prefix} {suffix}'\n"
|
|
]
|
|
},
|
|
{
|
|
"cell_type": "code",
|
|
"execution_count": 3,
|
|
"metadata": {},
|
|
"outputs": [],
|
|
"source": [
|
|
"def get_words(filename):\n",
|
|
" with lzma.open(filename, mode='rt', encoding='utf-8') as file:\n",
|
|
" count = 1\n",
|
|
" print('Words')\n",
|
|
" for line in file:\n",
|
|
" print(f'\\rProgress: {(count / 432022 * 100):2f}%', end='')\n",
|
|
" text = clean_line(line)\n",
|
|
" for word in text.split():\n",
|
|
" yield word\n",
|
|
" count += 1\n",
|
|
" print()"
|
|
]
|
|
},
|
|
{
|
|
"cell_type": "code",
|
|
"execution_count": 4,
|
|
"metadata": {},
|
|
"outputs": [],
|
|
"source": [
|
|
"def get_bigrams(filename, V):\n",
|
|
" with lzma.open(filename, mode='rt', encoding='utf-8') as file:\n",
|
|
" count = 1\n",
|
|
" print('Bigrams')\n",
|
|
" for line in file:\n",
|
|
" print(f'\\rProgress: {(count / 432022 * 100):2f}%', end='')\n",
|
|
" text = clean_line(line)\n",
|
|
" first_word = ''\n",
|
|
" for second_word in text.split():\n",
|
|
" if V.get(second_word) is None:\n",
|
|
" second_word = 'UNK'\n",
|
|
" if second_word:\n",
|
|
" yield first_word, second_word\n",
|
|
" first_word = second_word\n",
|
|
" count += 1\n",
|
|
" print()\n",
|
|
"\n",
|
|
"def get_trigrams(filename, V):\n",
|
|
" with lzma.open(filename, mode='rt', encoding='utf-8') as file:\n",
|
|
" count = 1\n",
|
|
" print('Trigrams')\n",
|
|
" for line in file:\n",
|
|
" print(f'\\rProgress: {(count / 432022 * 100):2f}%', end='')\n",
|
|
" text = clean_line(line)\n",
|
|
" first_word = ''\n",
|
|
" second_word = ''\n",
|
|
" for third_word in text.split():\n",
|
|
" if V.get(third_word) is None:\n",
|
|
" third_word = 'UNK'\n",
|
|
" if first_word:\n",
|
|
" yield first_word, second_word, third_word\n",
|
|
" first_word = second_word\n",
|
|
" second_word = third_word\n",
|
|
" count += 1\n",
|
|
" print()"
|
|
]
|
|
},
|
|
{
|
|
"cell_type": "code",
|
|
"execution_count": 5,
|
|
"metadata": {},
|
|
"outputs": [
|
|
{
|
|
"name": "stdout",
|
|
"output_type": "stream",
|
|
"text": [
|
|
"Words\n",
|
|
"Progress: 100.000000%\n"
|
|
]
|
|
}
|
|
],
|
|
"source": [
|
|
"\n",
|
|
"WORD_LIMIT = 3000\n",
|
|
"V = Counter(get_words('train/in.tsv.xz'))\n",
|
|
"V_common_dict = dict(V.most_common(WORD_LIMIT))\n",
|
|
"UNK = 0\n",
|
|
"for key, value in V.items():\n",
|
|
" if V_common_dict.get(key) is None:\n",
|
|
" UNK += value\n",
|
|
"V_common_dict['UNK'] = UNK\n",
|
|
"with open('V.pickle', 'wb') as handle:\n",
|
|
" pickle.dump(V_common_dict, handle, protocol=pickle.HIGHEST_PROTOCOL)\n",
|
|
" \n",
|
|
"with open('V.pickle', 'rb') as handle:\n",
|
|
" V_common_dict = pickle.load(handle)\n",
|
|
"\n",
|
|
"total = sum(V_common_dict.values())"
|
|
]
|
|
},
|
|
{
|
|
"cell_type": "code",
|
|
"execution_count": 6,
|
|
"metadata": {},
|
|
"outputs": [
|
|
{
|
|
"name": "stdout",
|
|
"output_type": "stream",
|
|
"text": [
|
|
"Bigrams\n",
|
|
"Progress: 100.000000%\n"
|
|
]
|
|
}
|
|
],
|
|
"source": [
|
|
"V2 = Counter(get_bigrams('train/in.tsv.xz', V_common_dict))\n",
|
|
"V2_dict = dict(V2)\n",
|
|
"with open('V2.pickle', 'wb') as handle:\n",
|
|
" pickle.dump(V2_dict, handle, protocol=pickle.HIGHEST_PROTOCOL)\n",
|
|
"\n",
|
|
"with open('V2.pickle', 'rb') as handle:\n",
|
|
" V2_dict = pickle.load(handle)"
|
|
]
|
|
},
|
|
{
|
|
"cell_type": "code",
|
|
"execution_count": 7,
|
|
"metadata": {},
|
|
"outputs": [
|
|
{
|
|
"name": "stdout",
|
|
"output_type": "stream",
|
|
"text": [
|
|
"Trigrams\n",
|
|
"Progress: 100.000000%\n"
|
|
]
|
|
}
|
|
],
|
|
"source": [
|
|
"V3 = Counter(get_trigrams('train/in.tsv.xz', V_common_dict))\n",
|
|
"V3_dict = dict(V3)\n",
|
|
"with open('V3.pickle', 'wb') as handle:\n",
|
|
" pickle.dump(V3_dict, handle, protocol=pickle.HIGHEST_PROTOCOL)\n",
|
|
"\n",
|
|
"with open('V3.pickle', 'rb') as handle:\n",
|
|
" V3_dict = pickle.load(handle)\n"
|
|
]
|
|
},
|
|
{
|
|
"cell_type": "code",
|
|
"execution_count": 8,
|
|
"metadata": {},
|
|
"outputs": [],
|
|
"source": [
|
|
"\n",
|
|
"def calculate_probability(first_word, second_word=None, third_word=None):\n",
|
|
" try:\n",
|
|
" if second_word is None:\n",
|
|
" return V_common_dict[first_word] / total\n",
|
|
" if third_word is None:\n",
|
|
" return V2_dict[(first_word, second_word)] / V_common_dict[first_word]\n",
|
|
" else:\n",
|
|
" return V3_dict[(first_word, second_word, third_word)] / V2_dict[(first_word, second_word)]\n",
|
|
" except KeyError:\n",
|
|
" return 0"
|
|
]
|
|
},
|
|
{
|
|
"cell_type": "code",
|
|
"execution_count": 9,
|
|
"metadata": {},
|
|
"outputs": [],
|
|
"source": [
|
|
"def smoothed(trigrams):\n",
|
|
" first, second, third = trigrams\n",
|
|
" return 0.6 * calculate_probability(first, second, third) + 0.25 * calculate_probability(second, third) + 0.15 * calculate_probability(\n",
|
|
" third)\n",
|
|
"\n",
|
|
"\n",
|
|
"def candidates(left_context, right_context):\n",
|
|
" cand = {}\n",
|
|
" first, second= left_context\n",
|
|
" fourth, fifth = right_context\n",
|
|
" for word in V_common_dict:\n",
|
|
" p1 = smoothed((first, second, word))\n",
|
|
" p2 = smoothed((second, word, fourth))\n",
|
|
" p3 = smoothed((word, fourth,fifth))\n",
|
|
" cand[word] = p1 * p2 * p3 \n",
|
|
" cand = sorted(list(cand.items()), key=lambda x: x[1], reverse=True)[:5]\n",
|
|
" norm = [(x[0], float(x[1]) / sum([y[1] for y in cand])) for x in cand]\n",
|
|
" for index, elem in enumerate(norm):\n",
|
|
" unk = None\n",
|
|
" if 'UNK' in elem:\n",
|
|
" unk = norm.pop(index)\n",
|
|
" norm.append(('', unk[1]))\n",
|
|
" break\n",
|
|
" if unk is None:\n",
|
|
" norm[-1] = ('', norm[-1][1])\n",
|
|
" return ' '.join([f'{x[0]}:{x[1]}' for x in norm])"
|
|
]
|
|
},
|
|
{
|
|
"cell_type": "code",
|
|
"execution_count": 12,
|
|
"metadata": {},
|
|
"outputs": [],
|
|
"source": [
|
|
"def create_outputs(folder_name):\n",
|
|
" print(f'Creating outputs in {folder_name}')\n",
|
|
" with lzma.open(f'{folder_name}/in.tsv.xz', mode='rt', encoding='utf-8') as fid:\n",
|
|
" with open(f'{folder_name}/out.tsv', 'w', encoding='utf-8') as f:\n",
|
|
" for line in fid:\n",
|
|
" separated = line.split('\\t')\n",
|
|
" prefix = separated[6].replace(r'\\n', ' ').split()\n",
|
|
" suffix = separated[7].replace(r'\\n', ' ').split()\n",
|
|
" left_context = [x if V_common_dict.get(x) else 'UNK' for x in prefix[-2:]]\n",
|
|
" right_context = [x if V_common_dict.get(x) else 'UNK' for x in suffix[:2]]\n",
|
|
" w = candidates(left_context, right_context)\n",
|
|
" f.write(w + '\\n')\n"
|
|
]
|
|
},
|
|
{
|
|
"cell_type": "code",
|
|
"execution_count": 15,
|
|
"metadata": {},
|
|
"outputs": [
|
|
{
|
|
"name": "stdout",
|
|
"output_type": "stream",
|
|
"text": [
|
|
"Creating outputs in dev-0\n",
|
|
"Creating outputs in test-A\n"
|
|
]
|
|
}
|
|
],
|
|
"source": [
|
|
"create_outputs('dev-0')\n",
|
|
"create_outputs('test-A')"
|
|
]
|
|
}
|
|
],
|
|
"metadata": {
|
|
"kernelspec": {
|
|
"display_name": "base",
|
|
"language": "python",
|
|
"name": "python3"
|
|
},
|
|
"language_info": {
|
|
"codemirror_mode": {
|
|
"name": "ipython",
|
|
"version": 3
|
|
},
|
|
"file_extension": ".py",
|
|
"mimetype": "text/x-python",
|
|
"name": "python",
|
|
"nbconvert_exporter": "python",
|
|
"pygments_lexer": "ipython3",
|
|
"version": "3.9.7"
|
|
},
|
|
"orig_nbformat": 4
|
|
},
|
|
"nbformat": 4,
|
|
"nbformat_minor": 2
|
|
}
|