René's URL Explorer Experiment

    {
       "@context": "https://schema.org",
       "@type": "Article",
       "name": "\uae30\ucd08\ubd80\ud130 \uc2dc\uc791\ud558\ub294 NLP: \ubb38\uc790-\ub2e8\uc704 RNN\uc73c\ub85c \uc774\ub984 \uc0dd\uc131\ud558\uae30",
       "headline": "\uae30\ucd08\ubd80\ud130 \uc2dc\uc791\ud558\ub294 NLP: \ubb38\uc790-\ub2e8\uc704 RNN\uc73c\ub85c \uc774\ub984 \uc0dd\uc131\ud558\uae30",
       "description": "PyTorch Documentation. Explore PyTorch, an open-source machine learning library that accelerates the path from research prototyping to production deployment. Discover tutorials, API references, and guides to help you build and deploy deep learning models efficiently.",
       "url": "/intermediate/char_rnn_generation_tutorial.html",
       "articleBody": "\ucc38\uace0 Go to the end to download the full example code. \uae30\ucd08\ubd80\ud130 \uc2dc\uc791\ud558\ub294 NLP: \ubb38\uc790-\ub2e8\uc704 RNN\uc73c\ub85c \uc774\ub984 \uc0dd\uc131\ud558\uae30# Author: Sean Robertson\ubc88\uc5ed: \ud669\uc131\uc218 \uc774 \ud29c\ud1a0\ub9ac\uc5bc\uc740 3\ubd80\ub85c \uad6c\uc131\ub41c \uc2dc\ub9ac\uc988\uc758 \uc77c\ubd80\uc785\ub2c8\ub2e4: \uae30\ucd08\ubd80\ud130 \uc2dc\uc791\ud558\ub294 NLP: \ubb38\uc790-\ub2e8\uc704 RNN\uc73c\ub85c \uc774\ub984 \ubd84\ub958\ud558\uae30 \uae30\ucd08\ubd80\ud130 \uc2dc\uc791\ud558\ub294 NLP: \ubb38\uc790-\ub2e8\uc704 RNN\uc73c\ub85c \uc774\ub984 \uc0dd\uc131\ud558\uae30 \uae30\ucd08\ubd80\ud130 \uc2dc\uc791\ud558\ub294 NLP: Sequence to Sequence \ub124\ud2b8\uc6cc\ud06c\uc640 Attention\uc744 \uc774\uc6a9\ud55c \ubc88\uc5ed \uc774 \ud29c\ud1a0\ub9ac\uc5bc\uc740 3\uac1c\ub85c \uc774\ub904\uc9c4 \u201c\uae30\ucd08\ubd80\ud130 \uc2dc\uc791\ud558\ub294 NLP\u201d\uc758 2\ubc88\uc9f8 \ud29c\ud1a0\ub9ac\uc5bc\uc785\ub2c8\ub2e4. \uccab\ubc88\uc9f8 \ud29c\ud1a0\ub9ac\uc5bc\uc778 \uae30\ucd08\ubd80\ud130 \uc2dc\uc791\ud558\ub294 NLP: \ubb38\uc790-\ub2e8\uc704 RNN\uc73c\ub85c \uc774\ub984 \ubd84\ub958\ud558\uae30 \uc5d0\uc11c\ub294 RNN\uc744 \uc0ac\uc6a9\ud558\uc5ec \uc8fc\uc5b4\uc9c4 \uc774\ub984\uc774 \uc5b4\ub5a0\ud55c \uc5b8\uc5b4\uc778\uc9c0\ub97c \ubd84\ub958\ud588\uc2b5\ub2c8\ub2e4. \uc774\ubc88\uc5d0\ub294 \ubc18\ub300\ub85c \uc5b8\uc5b4\ub85c\ubd80\ud130 \uc774\ub984\uc744 \uc0dd\uc131\ud560 \uc608\uc815\uc785\ub2c8\ub2e4. \u003e python sample.py Russian RUS Rovakov Uantov Shavakov \u003e python sample.py German GER Gerren Ereng Rosher \u003e python sample.py Spanish SPA Salla Parer Allan \u003e python sample.py Chinese CHI Chan Hang Iun \uc6b0\ub9ac\ub294 \uba87 \uac1c\uc758 \uc120\ud615 \uacc4\uce35\uc73c\ub85c \uc791\uc740 RNN\uc744 \uc9c1\uc811 \ub9cc\ub4e4\uace0 \uc788\uc2b5\ub2c8\ub2e4. \uc774\uc804 \ud29c\ud1a0\ub9ac\uc5bc\uc778 \uc774\ub984\uc744 \uc77d\uc740 \ud6c4 \uadf8 \uc5b8\uc5b4\ub97c \uc608\uce21\ud558\ub294 \uac83\uacfc\uc758 \ud070 \ucc28\uc774\uc810\uc740 \uc5b8\uc5b4\ub97c \uc785\ub825\ud558\uace0 \ud55c \ubc88\uc5d0 \ud55c \uae00\uc790\ub97c \uc0dd\uc131\ud558\uc5ec \ucd9c\ub825\ud558\ub294 \uac83\uc785\ub2c8\ub2e4. \uc5b8\uc5b4 \ud615\uc131(\ub2e8\uc5b4 \ub610\ub294 \ub2e4\ub978 \uace0\ucc28\uc6d0 \uad6c\uc870\ub85c\ub3c4 \uc218\ud589\ub420 \uc218 \uc788\uc74c)\uc744 \uc704\ud574 \ubb38\uc790\ub97c \ubc18\ubcf5\uc801\uc73c\ub85c \uc608\uce21\ud558\ub294 \uac83\uc744 \u201c\uc5b8\uc5b4 \ubaa8\ub378\u201d \uc774\ub77c\uace0 \ud569\ub2c8\ub2e4. \ucd94\ucc9c \uc790\ub8cc: Pytorch\ub97c \uc124\uce58\ud588\uace0, Python\uc744 \uc54c\uace0, Tensor\ub97c \uc774\ud574\ud55c\ub2e4\uace0 \uac00\uc815\ud569\ub2c8\ub2e4: https://pytorch.org/ \uc124\uce58 \uc548\ub0b4 PyTorch\ub85c \ub525\ub7ec\ub2dd\ud558\uae30: 60\ubd84\ub9cc\uc5d0 \ub05d\uc7a5\ub0b4\uae30 PyTorch \uc2dc\uc791\ud558\uae30 \uc608\uc81c\ub85c \ubc30\uc6b0\ub294 \ud30c\uc774\ud1a0\uce58(PyTorch) \ub113\uace0 \uae4a\uc740 \ud1b5\ucc30\uc744 \uc704\ud55c \uc790\ub8cc Torch \uc0ac\uc6a9\uc790\ub97c \uc704\ud55c PyTorch \uc774\uc804 Lua Torch \uc0ac\uc6a9\uc790\ub97c \uc704\ud55c \uc790\ub8cc RNN\uacfc \uc791\ub3d9 \ubc29\uc2dd\uc744 \uc544\ub294 \uac83 \ub610\ud55c \uc720\uc6a9\ud569\ub2c8\ub2e4: The Unreasonable Effectiveness of Recurrent Neural Networks \uc2e4\uc0dd\ud65c \uc608\uc81c\ub97c \ubcf4\uc5ec \uc90d\ub2c8\ub2e4. Understanding LSTM Networks LSTM\uc5d0 \uad00\ud55c \uac83\uc774\uc9c0\ub9cc RNN\uc5d0 \uad00\ud574\uc11c\ub3c4 \uc720\uc775\ud569\ub2c8\ub2e4. \uc774\uc804 \ud29c\ud1a0\ub9ac\uc5bc\ub3c4 \ucd94\ucc9c\ud569\ub2c8\ub2e4. \uae30\ucd08\ubd80\ud130 \uc2dc\uc791\ud558\ub294 NLP: \ubb38\uc790-\ub2e8\uc704 RNN\uc73c\ub85c \uc774\ub984 \ubd84\ub958\ud558\uae30 \ub370\uc774\ud130 \uc900\ube44# \ucc38\uace0 \uc5ec\uae30 \uc5d0\uc11c \ub370\uc774\ud130\ub97c \ub2e4\uc6b4 \ubc1b\uace0, \ud604\uc7ac \ub514\ub809\ud1a0\ub9ac\uc5d0 \uc555\ucd95\uc744 \ud478\uc2ed\uc2dc\uc624. \uc774 \uacfc\uc815\uc758 \ub354 \uc790\uc138\ud55c \uc0ac\ud56d\uc740 \uc9c0\ub09c \ud29c\ud1a0\ub9ac\uc5bc\uc744 \ubcf4\uc2ed\uc2dc\uc624. \uc694\uc57d\ud558\uba74, \uc904\ub9c8\ub2e4 \uc774\ub984\uc774 \uc801\ud78c \ud14d\uc2a4\ud2b8 \ud30c\uc77c data/names/[Language].txt \uc788\uc2b5\ub2c8\ub2e4. \uc774\uac83\uc744 array\ub85c \ubd84\ub9ac\ud558\uace0, Unicode\ub97c ASCII\ub85c \ubcc0\uacbd\ud558\uace0, \uc0ac\uc804 {language: [names ...]} \uc744 \ub9cc\ub4e4\uc5b4\uc11c \ub9c8\ubb34\ub9ac\ud569\ub2c8\ub2e4. from __future__ import unicode_literals, print_function, division from io import open import glob import os import unicodedata import string all_letters = string.ascii_letters + \" .,;\u0027-\" n_letters = len(all_letters) + 1 # EOS(end of sentence) \uae30\ud638 \ucd94\uac00 def findFiles(path): return glob.glob(path) # \uc720\ub2c8\ucf54\ub4dc \ubb38\uc790\uc5f4\uc744 ASCII\ub85c \ubcc0\ud658, https://stackoverflow.com/a/518232/2809427 def unicodeToAscii(s): return \u0027\u0027.join( c for c in unicodedata.normalize(\u0027NFD\u0027, s) if unicodedata.category(c) != \u0027Mn\u0027 and c in all_letters ) # \ud30c\uc77c\uc744 \uc77d\uace0 \uc904 \ub2e8\uc704\ub85c \ubd84\ub9ac def readLines(filename): with open(filename, encoding=\u0027utf-8\u0027) as some_file: return [unicodeToAscii(line.strip()) for line in some_file] # \uac01 \uc5b8\uc5b4\uc758 \uc774\ub984 \ubaa9\ub85d\uc778 category_lines \uc0ac\uc804 \uc0dd\uc131 category_lines = {} all_categories = [] for filename in findFiles(\u0027data/names/*.txt\u0027): category = os.path.splitext(os.path.basename(filename))[0] all_categories.append(category) lines = readLines(filename) category_lines[category] = lines n_categories = len(all_categories) if n_categories == 0: raise RuntimeError(\u0027Data not found. Make sure that you downloaded data \u0027 \u0027from https://download.pytorch.org/tutorial/data.zip and extract it to \u0027 \u0027the current directory.\u0027) print(\u0027# categories:\u0027, n_categories, all_categories) print(unicodeToAscii(\"O\u0027N\u00e9\u00e0l\")) # categories: 18 [\u0027English\u0027, \u0027Dutch\u0027, \u0027Spanish\u0027, \u0027Japanese\u0027, \u0027Portuguese\u0027, \u0027Arabic\u0027, \u0027Czech\u0027, \u0027Vietnamese\u0027, \u0027Irish\u0027, \u0027French\u0027, \u0027German\u0027, \u0027Russian\u0027, \u0027Italian\u0027, \u0027Chinese\u0027, \u0027Korean\u0027, \u0027Greek\u0027, \u0027Polish\u0027, \u0027Scottish\u0027] O\u0027Neal \ub124\ud2b8\uc6cc\ud06c \uc0dd\uc131# \uc774 \ub124\ud2b8\uc6cc\ud06c\ub294 \uc9c0\ub09c \ud29c\ud1a0\ub9ac\uc5bc\uc758 RNN \uc774 \ub2e4\ub978 \uc785\ub825\ub4e4\uacfc \uc5f0\uacb0\ub418\ub294 category tensor\ub97c \ucd94\uac00 \uc778\uc790\ub85c \uac00\uc9c0\uac8c \ud655\uc7a5\ud569\ub2c8\ub2e4. category tensor\ub294 \ubb38\uc790 \uc785\ub825\uacfc \ub9c8\ucc2c\uac00\uc9c0\ub85c one-hot \ubca1\ud130\uc785\ub2c8\ub2e4. \uc5ed\uc790\uc8fc: \uae30\uc874 \uc785\ub825\uacfc category tensor\ub97c \uacb0\ud569\ud558\uc5ec \uc785\ub825\uc73c\ub85c \uc0ac\uc6a9\ud558\uae30 \ub54c\ubb38\uc5d0 \uc785\ub825\uc758 \uc0ac\uc774\uc988\uac00 n_categories \ub9cc\ud07c \ucee4\uc9d1\ub2c8\ub2e4. \uc6b0\ub9ac\ub294 \ucd9c\ub825\uc744 \ub2e4\uc74c \ubb38\uc790\uc758 \ud655\ub960\ub85c \ud574\uc11d\ud569\ub2c8\ub2e4. \uc0d8\ud50c\ub9c1 \ud560 \ub54c, \uac00\uc7a5 \ud655\ub960\uc774 \ub192\uc740 \ubb38\uc790\uac00 \ub2e4\uc74c \uc785\ub825 \ubb38\uc790\ub85c \uc0ac\uc6a9\ub429\ub2c8\ub2e4. \ub354 \ub098\uc740 \ub3d9\uc791\uc744 \uc704\ud574 \ub450 \ubc88\uc9f8 \uc120\ud615 \ub808\uc774\uc5b4 o2o (\uc740\ub2c9\uacfc \ucd9c\ub825\uc744 \uacb0\ud569\ud55c \ud6c4) \ub97c \ucd94\uac00\ud588\uc2b5\ub2c8\ub2e4 . \ub610\ud55c Drop-out \uacc4\uce35\uc774 \uc788\uc2b5\ub2c8\ub2e4. \uc774 \uacc4\uce35\uc740 \uc8fc\uc5b4\uc9c4 \ud655\ub960(\uc5ec\uae30\uc11c\ub294 0.1)\ub85c \ubb34\uc791\uc704\ub85c \uc785\ub825\uc744 0 # \uc73c\ub85c \ub9cc\ub4ed\ub2c8\ub2e4. \uc77c\ubc18\uc801\uc73c\ub85c \uc785\ub825\uc744 \ud750\ub9ac\uac8c \ud574\uc11c \uacfc\uc801\ud569\uc744 \ub9c9\ub294 \ub370 \uc0ac\uc6a9\ub429\ub2c8\ub2e4. \uc5ec\uae30\uc11c \uc6b0\ub9ac\ub294 \uace0\uc758\ub85c \uc77c\ubd80 \ud63c\ub3c8\uc744 \ucd94\uac00\ud558\uace0 \uc0d8\ud50c\ub9c1 \ub2e4\uc591\uc131\uc744 \ub192\uc774\uae30 \uc704\ud574 \ub124\ud2b8\uc6cc\ud06c\uc758 \ub9c8\uc9c0\ub9c9\uc5d0 \uc774\uac83\uc744 \uc0ac\uc6a9\ud569\ub2c8\ub2e4. import torch import torch.nn as nn class RNN(nn.Module): def __init__(self, input_size, hidden_size, output_size): super(RNN, self).__init__() self.hidden_size = hidden_size self.i2h = nn.Linear(n_categories + input_size + hidden_size, hidden_size) self.i2o = nn.Linear(n_categories + input_size + hidden_size, output_size) self.o2o = nn.Linear(hidden_size + output_size, output_size) self.dropout = nn.Dropout(0.1) self.softmax = nn.LogSoftmax(dim=1) def forward(self, category, input, hidden): input_combined = torch.cat((category, input, hidden), 1) hidden = self.i2h(input_combined) output = self.i2o(input_combined) output_combined = torch.cat((hidden, output), 1) output = self.o2o(output_combined) output = self.dropout(output) output = self.softmax(output) return output, hidden def initHidden(self): return torch.zeros(1, self.hidden_size) \ud559\uc2b5# \ud559\uc2b5 \uc900\ube44# \uc81c\uc77c \uba3c\uc800 (category, line)\uc758 \ubb34\uc791\uc704 \uc30d\uc744 \uc5bb\ub294 \ud568\uc218: import random # \ubaa9\ub85d\uc5d0\uc11c \ubb34\uc791\uc704 \uc544\uc774\ud15c \ubc18\ud658 def randomChoice(l): return l[random.randint(0, len(l) - 1)] # \uc784\uc758\uc758 category \ubc0f \uadf8 category\uc5d0\uc11c \ubb34\uc791\uc704 \uc904(\uc774\ub984) \uc5bb\uae30 def randomTrainingPair(): category = randomChoice(all_categories) line = randomChoice(category_lines[category]) return category, line \uac01 \uc2dc\uac04 \ub2e8\uacc4 \ub9c8\ub2e4 (\uc989, \ud559\uc2b5 \ub2e8\uc5b4\uc758 \uac01 \ubb38\uc790 \ub9c8\ub2e4) \ub124\ud2b8\uc6cc\ud06c\uc758 \uc785\ub825\uc740 (\uc5b8\uc5b4, \ud604\uc7ac \ubb38\uc790, \uc740\ub2c9 \uc0c1\ud0dc) \uac00 \ub418\uace0, \ucd9c\ub825\uc740 (\ub2e4\uc74c \ubb38\uc790, \ub2e4\uc74c \uc740\ub2c9 \uc0c1\ud0dc) \uac00 \ub41c\ub2e4. \ub530\ub77c\uc11c \uac01 \ud559\uc2b5 \uc138\ud2b8 \ub9c8\ub2e4 \uc5b8\uc5b4, \uc785\ub825 \ubb38\uc790\uc758 \uc138\ud2b8, \ucd9c\ub825/\ubaa9\ud45c \ubb38\uc790\uc758 \uc138\ud2b8\uac00 \ud544\uc694\ud558\ub2e4. \uac01 \uc2dc\uac04 \ub2e8\uacc4\ub9c8\ub2e4 \ud604\uc7ac \ubb38\uc790\uc5d0\uc11c \ub2e4\uc74c \ubb38\uc790\ub97c \uc608\uce21\ud558\uae30 \ub54c\ubb38\uc5d0, \ubb38\uc790 \uc30d\uc740 \ud55c \uc904(\ud558\ub098\uc758 \uc774\ub984)\uc5d0\uc11c \uc5f0\uc18d\ub41c \ubb38\uc790 \uadf8\ub8f9\uc785\ub2c8\ub2e4. - \uc608\ub97c \ub4e4\uc5b4 \"ABCD\u003cEOS\u003e\" \ub294 (\u201cA\u201d, \u201cB\u201d), (\u201cB\u201d, \u201cC\u201d), (\u201cC\u201d, \u201cD\u201d), (\u201cD\u201d, \u201cEOS\u201d) \ub85c \uc0dd\uc131\ud569\ub2c8\ub2e4. Category(\uc5b8\uc5b4) Tensor\ub294 \u003c1 x n_categories\u003e \ud06c\uae30\uc758 One-hot Tensor \uc785\ub2c8\ub2e4. \ud559\uc2b5\uc2dc\uc5d0 \ubaa8\ub4e0 \uc2dc\uac04 \ub2e8\uacc4\uc5d0\uc11c \ub124\ud2b8\uc6cc\ud06c\uc5d0 \uc774\uac83\uc744 \uc804\ub2ec\ud569\ub2c8\ub2e4. - \uc774\uac83\uc740 \uc124\uacc4 \uc120\ud0dd\uc0ac\ud56d\uc73c\ub85c, \ucd08\uae30 \uc740\ub2c9 \uc0c1\ud0dc \ub610\ub294 \ub610 \ub2e4\ub978 \uc804\ub7b5\uc758 \ubd80\ubd84\uc73c\ub85c \ud3ec\ud568\ub420 \uc218 \uc788\uc2b5\ub2c8\ub2e4. # Category\ub97c \uc704\ud55c One-hot \ubca1\ud130 def categoryTensor(category): li = all_categories.index(category) tensor = torch.zeros(1, n_categories) tensor[0][li] = 1 return tensor # \uc785\ub825\uc744 \uc704\ud55c \ucc98\uc74c\ubd80\ud130 \ub9c8\uc9c0\ub9c9 \ubb38\uc790(EOS \uc81c\uc678)\uae4c\uc9c0\uc758 One-hot \ud589\ub82c def inputTensor(line): tensor = torch.zeros(len(line), 1, n_letters) for li in range(len(line)): letter = line[li] tensor[li][0][all_letters.find(letter)] = 1 return tensor # \ubaa9\ud45c\ub97c \uc704\ud55c \ub450\ubc88\uc9f8 \ubb38\uc790 \ubd80\ud130 \ub9c8\uc9c0\ub9c9(EOS)\uae4c\uc9c0\uc758 ``LongTensor`` def targetTensor(line): letter_indexes = [all_letters.find(line[li]) for li in range(1, len(line))] letter_indexes.append(n_letters - 1) # EOS return torch.LongTensor(letter_indexes) \ud559\uc2b5 \ub3d9\uc548 \ud3b8\uc758\ub97c \uc704\ud574 \ubb34\uc791\uc704\ub85c (category[\uc5b8\uc5b4], line[\uc774\ub984])\uc744 \uac00\uc838\uc624\uace0 \uadf8\uac83\uc744 \ud544\uc694\ud55c \ud615\ud0dc (category[\uc5b8\uc5b4], input[\ud604\uc7ac \ubb38\uc790], target[\ub2e4\uc74c \ubb38\uc790]) Tensor\ub85c \ubc14\uafb8\ub294 randomTrainingExample \ud568\uc218\ub97c \ub9cc\ub4e4 \uc608\uc815\uc785\ub2c8\ub2e4. # \uc784\uc758\uc758 Category\uc5d0\uc11c Category, Input, Target Tensor\ub97c \ub9cc\ub4ed\ub2c8\ub2e4. def randomTrainingExample(): category, line = randomTrainingPair() category_tensor = categoryTensor(category) input_line_tensor = inputTensor(line) target_line_tensor = targetTensor(line) return category_tensor, input_line_tensor, target_line_tensor \ub124\ud2b8\uc6cc\ud06c \ud559\uc2b5# \ub9c8\uc9c0\ub9c9 \ucd9c\ub825\ub9cc \uc0ac\uc6a9\ud558\ub294 \ubd84\ub958\uc640 \ub2ec\ub9ac, \ubaa8\ub4e0 \ub2e8\uacc4\uc5d0\uc11c \uc608\uce21\uc744 \uc218\ud589\ud558\ubbc0\ub85c \ubaa8\ub4e0 \ub2e8\uacc4\uc5d0\uc11c \uc190\uc2e4\uc744 \uacc4\uc0b0\ud569\ub2c8\ub2e4. Autograd\uc758 \ub9c8\ubc95\uc774 \uac01 \ub2e8\uacc4\uc758 \uc190\uc2e4\ub4e4\uc744 \uac04\ub2e8\ud558\uac8c \ud569\ud558\uace0 \ub9c8\uc9c0\ub9c9\uc5d0 \uc5ed\uc804\ud30c\ub97c \ud638\ucd9c\ud558\uac8c \ud574\uc90d\ub2c8\ub2e4. criterion = nn.NLLLoss() learning_rate = 0.0005 def train(category_tensor, input_line_tensor, target_line_tensor): target_line_tensor.unsqueeze_(-1) hidden = rnn.initHidden() rnn.zero_grad() loss = torch.Tensor([0]) # \ub610\ub294 \uadf8\ub0e5 ``loss = 0`` \uc744 \uc0ac\uc6a9\ud574\ub3c4 \ub429\ub2c8\ub2e4. for i in range(input_line_tensor.size(0)): output, hidden = rnn(category_tensor, input_line_tensor[i], hidden) l = criterion(output, target_line_tensor[i]) loss += l loss.backward() for p in rnn.parameters(): p.data.add_(p.grad.data, alpha=-learning_rate) return output, loss.item() / input_line_tensor.size(0) \ud559\uc2b5\uc5d0 \uac78\ub9ac\ub294 \uc2dc\uac04\uc744 \ucd94\uc801\ud558\uae30 \uc704\ud574 \uc0ac\ub78c\uc774 \uc77d\uc744 \uc218 \uc788\ub294 \ubb38\uc790\uc5f4\uc744 \ubc18\ud658\ud558\ub294``timeSince (timestamp)`` \ud568\uc218\ub97c \ucd94\uac00\ud569\ub2c8\ub2e4: import time import math def timeSince(since): now = time.time() s = now - since m = math.floor(s / 60) s -= m * 60 return \u0027%dm %ds\u0027 % (m, s) \ud559\uc2b5\uc740 \uc77c\uc0c1\uc801\uc778 \uc77c\uc785\ub2c8\ub2e4. - \uba87 \ubc88 train() \uc744 \ud638\ucd9c\ud558\uace0, \uba87 \ubd84 \uc815\ub3c4 \uae30\ub2e4\ub838\ub2e4\uac00 print_every \ub9c8\ub2e4 \ud604\uc7ac \uc2dc\uac04\uacfc \uc190\uc2e4\uc744 \ucd9c\ub825\ud558\uace0, \ub098\uc911\uc5d0 \ub3c4\uc2dd\ud654\ub97c \uc704\ud574 plot_every \ub9c8\ub2e4 all_losses \uc5d0 \ud3c9\uade0 \uc190\uc2e4\uc744 \uc800\uc7a5\ud569\ub2c8\ub2e4. rnn = RNN(n_letters, 128, n_letters) n_iters = 100000 print_every = 5000 plot_every = 500 all_losses = [] total_loss = 0 # ``plot_every`` \ub9c8\ub2e4 \ucd08\uae30\ud654 start = time.time() for iter in range(1, n_iters + 1): output, loss = train(*randomTrainingExample()) total_loss += loss if iter % print_every == 0: print(\u0027%s (%d %d%%) %.4f\u0027 % (timeSince(start), iter, iter / n_iters * 100, loss)) if iter % plot_every == 0: all_losses.append(total_loss / plot_every) total_loss = 0 0m 15s (5000 5%) 3.0273 0m 26s (10000 10%) 2.8095 0m 38s (15000 15%) 3.2219 0m 49s (20000 20%) 1.5323 1m 1s (25000 25%) 2.5673 1m 12s (30000 30%) 2.1202 1m 23s (35000 35%) 2.5683 1m 35s (40000 40%) 2.3383 1m 46s (45000 45%) 2.3752 1m 57s (50000 50%) 2.7899 2m 8s (55000 55%) 1.7466 2m 19s (60000 60%) 2.2184 2m 31s (65000 65%) 3.2277 2m 42s (70000 70%) 2.4214 2m 53s (75000 75%) 2.9422 3m 4s (80000 80%) 2.3035 3m 15s (85000 85%) 2.5719 3m 26s (90000 90%) 3.0298 3m 38s (95000 95%) 3.1836 3m 49s (100000 100%) 2.6175 \uc190\uc2e4 \ub3c4\uc2dd\ud654# all_losses\ub97c \uc774\uc6a9\ud55c \uc190\uc2e4\uc758 \ub3c4\uc2dd\ud654\ub294 \ub124\ud2b8\uc6cc\ud06c\uc758 \ud559\uc2b5 \uc0c1\ud0dc\ub97c \ubcf4\uc5ec\uc90d\ub2c8\ub2e4: import matplotlib.pyplot as plt plt.figure() plt.plot(all_losses) [\u003cmatplotlib.lines.Line2D object at 0x7f0dab102bd0\u003e] \ub124\ud2b8\uc6cc\ud06c \uc0d8\ud50c\ub9c1# \uc0d8\ud50c\ub9c1\uc744 \uc704\ud574\uc11c, \ub124\ud2b8\uc6cc\ud06c\uc5d0 \ud558\ub098\uc758 \uae00\uc790\ub97c \uc8fc\uace0 \ub2e4\uc74c \ubb38\uc790\ub97c \ubb3c\uc5b4\ubcf4\uace0 \uc774\uac83\uc744 \ub2e4\uc74c \ubb38\uc790\ub85c \uc804\ub2ec\ud558\ub294 \uac83\uc744 EOS \ud1a0\ud070\uae4c\uc9c0 \ubc18\ubcf5\ud569\ub2c8\ub2e4. \uc785\ub825 \uce74\ud14c\uace0\ub9ac(\uc5b8\uc5b4), \uc2dc\uc791 \ubb38\uc790, \ube44\uc5b4 \uc788\ub294 \uc740\ub2c9 \uc0c1\ud0dc\ub97c \uc704\ud55c Tensor\ub97c \uc0dd\uc131\ud558\uc2ed\uc2dc\uc624 \uc2dc\uc791 \ubb38\uc790\ub85c output_name \ubb38\uc790\uc5f4\uc744 \uc0dd\uc131\ud558\uc2ed\uc2dc\uc624 \ucd5c\ub300 \ucd9c\ub825 \uae38\uc774\uae4c\uc9c0, \ud604\uc7ac \ubb38\uc790\ub97c \ub124\ud2b8\uc6cc\ud06c\uc5d0 \uc804\ub2ec\ud558\uc2ed\uc2dc\uc624. \uac00\uc7a5 \ub192\uc740 \ucd9c\ub825\uc5d0\uc11c \ub2e4\uc74c \ubb38\uc790\uc640 \ub2e4\uc74c \uc740\ub2c9 \uc0c1\ud0dc\ub97c \uc5bb\uc73c\uc2ed\uc2dc\uc624 \ub9cc\uc77c \ubb38\uc790\uac00 EOS\uba74, \uc5ec\uae30\uc11c \uba48\ucd94\uc2ed\uc2dc\uc624 \ub9cc\uc77c \uc77c\ubc18\uc801\uc778 \ubb38\uc790\ub77c\uba74, output_name \uc5d0 \ucd94\uac00\ud558\uace0 \uacc4\uc18d\ud558\uc2ed\uc2dc\uc624 \ub9c8\uc9c0\ub9c9 \uc774\ub984\uc744 \ubc18\ud658\ud558\uc2ed\uc2dc\uc624 \ucc38\uace0 \uc2dc\uc791 \ubb38\uc790\ub97c \uc8fc\ub294 \uac83 \uc678\uc5d0 \u201c\ubb38\uc790\uc5f4 \uc2dc\uc791\u201d \ud1a0\ud070\uc744 \ud559\uc2b5\uc5d0 \ud3ec\ud568\ub418\uac8c \ud558\uace0 \ub124\ud2b8\uc6cc\ud06c\uac00 \uc790\uccb4\uc801\uc73c\ub85c \uc2dc\uc791 \ubb38\uc790\ub97c \uc120\ud0dd\ud558\uac8c \ud558\ub294 \ub2e4\ub978 \ubc29\ubc95\ub3c4 \uc788\uc2b5\ub2c8\ub2e4. max_length = 20 # \uce74\ud14c\uace0\ub9ac\uc640 \uc2dc\uc791 \ubb38\uc790\ub85c\ubd80\ud130 \uc0d8\ud50c\ub9c1 \ud558\uae30 def sample(category, start_letter=\u0027A\u0027): with torch.no_grad(): # \uc0d8\ud50c\ub9c1\uc5d0\uc11c \ud788\uc2a4\ud1a0\ub9ac\ub97c \ucd94\uc801\ud560 \ud544\uc694 \uc5c6\uc74c category_tensor = categoryTensor(category) input = inputTensor(start_letter) hidden = rnn.initHidden() output_name = start_letter for i in range(max_length): output, hidden = rnn(category_tensor, input[0], hidden) topv, topi = output.topk(1) topi = topi[0][0] if topi == n_letters - 1: break else: letter = all_letters[topi] output_name += letter input = inputTensor(letter) return output_name # \ud558\ub098\uc758 \uce74\ud14c\uace0\ub9ac\uc640 \uc5ec\ub7ec \uc2dc\uc791 \ubb38\uc790\ub4e4\ub85c \uc5ec\ub7ec \uac1c\uc758 \uc0d8\ud50c \uc5bb\uae30 def samples(category, start_letters=\u0027ABC\u0027): for start_letter in start_letters: print(sample(category, start_letter)) samples(\u0027Russian\u0027, \u0027RUS\u0027) samples(\u0027German\u0027, \u0027GER\u0027) samples(\u0027Spanish\u0027, \u0027SPA\u0027) samples(\u0027Chinese\u0027, \u0027CHI\u0027) Rakovak Uantovov Shillov Gerren Eren Rour Salla Pares Allan Chin Han Iun Exercises# Try with a different dataset of category -\u003e line, for example: Fictional series -\u003e Character name Part of speech -\u003e Word Country -\u003e City Use a \u201cstart of sentence\u201d token so that sampling can be done without choosing a start letter Get better results with a bigger and/or better shaped network Try the nn.LSTM and nn.GRU layers \uc0c1\uc704 \uc218\uc900 \ub124\ud2b8\uc6cc\ud06c\ub85c \uc5ec\ub7ec \uac1c\uc758 \uc774\ub7f0 RNN\uc744 \uacb0\ud569\ud574 \ubcf4\uc2ed\uc2dc\uc624 Total running time of the script: (3 minutes 49.522 seconds) Download Jupyter notebook: char_rnn_generation_tutorial.ipynb Download Python source code: char_rnn_generation_tutorial.py Download zipped: char_rnn_generation_tutorial.zip",
       "author": {
         "@type": "Organization",
         "name": "PyTorch Contributors",
         "url": "https://pytorch.org"
       },
       "image": "../_static/img/pytorch_seo.png",
       "mainEntityOfPage": {
         "@type": "WebPage",
         "@id": "/intermediate/char_rnn_generation_tutorial.html"
       },
       "datePublished": "2023-01-01T00:00:00Z",
       "dateModified": "2023-01-01T00:00:00Z"
     }