PyTorch翻譯官網(wǎng)教程-NLP FROM SCRATCH: CLASSIFYING NAMES WITH A CHARACTER-LEVEL RNN

這篇具有很好參考價(jià)值的文章主要介紹了PyTorch翻譯官網(wǎng)教程-NLP FROM SCRATCH: CLASSIFYING NAMES WITH A CHARACTER-LEVEL RNN。希望對(duì)大家有所幫助。如果存在錯(cuò)誤或未考慮完全的地方，請大家不吝賜教，您也可以點(diǎn)擊"舉報(bào)違法"按鈕提交疑問。

官網(wǎng)鏈接

NLP From Scratch: Classifying Names with a Character-Level RNN — PyTorch Tutorials 2.0.1+cu117 documentation

使用CHARACTER-LEVEL RNN 對(duì)名字分類

我們將建立和訓(xùn)練一個(gè)基本的字符級(jí)遞歸神經(jīng)網(wǎng)絡(luò)(RNN)來分類單詞。本教程以及另外兩個(gè)“from scratch”的自然語言處理(NLP)教程 NLP From Scratch: Generating Names with a Character-Level RNN 和 NLP From Scratch: Translation with a Sequence to Sequence Network and Attention,演示如何預(yù)處理數(shù)據(jù)以建立NLP模型。特別是，這些教程沒有使用torchtext的許多便利功能，因此您可以看到如何簡單使用預(yù)處理模型NLP。

字符級(jí)RNN將單詞作為一系列字符來讀取，每一步輸出一個(gè)預(yù)測和“隱藏狀態(tài)”，將之前的隱藏狀態(tài)輸入到下一步。我們把最后的預(yù)測作為輸出，即這個(gè)詞屬于哪個(gè)類。

具體來說，我們將訓(xùn)練來自18種語言的幾千個(gè)姓氏，并根據(jù)拼寫來預(yù)測一個(gè)名字來自哪種語言:

$ python predict.py Hinton
(-0.47) Scottish
(-1.52) English
(-3.57) Irish

$ python predict.py Schmidhuber
(-0.19) German
(-2.48) Czech
(-2.68) Dutch

建議準(zhǔn)備

在開始本教程之前，建議您安裝PyTorch，并對(duì)Python編程語言和張量有基本的了解:

PyTorch 有關(guān)安裝說明
Deep Learning with PyTorch: A 60 Minute Blitz 開始使用PyTorch并學(xué)習(xí)張量的基礎(chǔ)知識(shí)
Learning PyTorch with Examples 使用概述
PyTorch for Former Torch Users 如果您是前Lua Torch用戶

了解rnn及其工作原理也很有用:

The Unreasonable Effectiveness of Recurrent Neural Networks 展示了一些現(xiàn)實(shí)生活中的例子
Understanding LSTM Networks 是專門關(guān)于LSTMs的，但也有關(guān)于RNNs的信息

準(zhǔn)備數(shù)據(jù)

從這里下載數(shù)據(jù)并將其解壓縮到當(dāng)前目錄。here

“data/names”目錄下包含18個(gè)文本文件，文件名為“[Language].txt”。每個(gè)文件包含一堆名稱，每行一個(gè)名稱，大多數(shù)是羅馬化的(但我們?nèi)匀恍枰獜腢nicode轉(zhuǎn)換為ASCII)。

我們最終會(huì)得到一個(gè)包含每種語言名稱列表的字典，{language: [names ...]}。通用變量“category”和“l(fā)ine”(在本例中表示語言和名稱)用于以后的可擴(kuò)展性。

from io import open
import glob
import os

def findFiles(path): return glob.glob(path)

print(findFiles('data/names/*.txt'))

import unicodedata
import string

all_letters = string.ascii_letters + " .,;'"
n_letters = len(all_letters)

# Turn a Unicode string to plain ASCII, thanks to https://stackoverflow.com/a/518232/2809427
def unicodeToAscii(s):
    return ''.join(
        c for c in unicodedata.normalize('NFD', s)
        if unicodedata.category(c) != 'Mn'
        and c in all_letters
    )

print(unicodeToAscii('?lusàrski'))

# Build the category_lines dictionary, a list of names per language
category_lines = {}
all_categories = []

# Read a file and split into lines
def readLines(filename):
    lines = open(filename, encoding='utf-8').read().strip().split('\n')
    return [unicodeToAscii(line) for line in lines]

for filename in findFiles('data/names/*.txt'):
    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)

輸出

['data/names/Arabic.txt', 'data/names/Chinese.txt', 'data/names/Czech.txt', 'data/names/Dutch.txt', 'data/names/English.txt', 'data/names/French.txt', 'data/names/German.txt', 'data/names/Greek.txt', 'data/names/Irish.txt', 'data/names/Italian.txt', 'data/names/Japanese.txt', 'data/names/Korean.txt', 'data/names/Polish.txt', 'data/names/Portuguese.txt', 'data/names/Russian.txt', 'data/names/Scottish.txt', 'data/names/Spanish.txt', 'data/names/Vietnamese.txt']
Slusarski

現(xiàn)在我們有了category_lines，這是一個(gè)將每個(gè)類別(語言)映射到行(名稱)列表的字典。我們還記錄了all_categories(只是一個(gè)語言列表)和n_categories，以供以后參考。

print(category_lines['Italian'][:5])

輸出

['Abandonato', 'Abatangelo', 'Abatantuono', 'Abate', 'Abategiovanni']

把名字變成張量

現(xiàn)在我們已經(jīng)組織好了所有的名字，我們需要把它們變成張量來使用它們。

為了表示單個(gè)字母，我們使用大小為<1 x n_letters> 的 “one-hot vector”。一個(gè)獨(dú)熱向量被0填充，除了當(dāng)前字母所以處是1。例如："b" = <0 1 0 0 0 ...>.

為了組成一個(gè)單詞，我們將一堆這樣的單詞連接到一個(gè)二維矩陣中<line_length x 1 x n_letters>.

額外的1維度是因?yàn)镻yTorch假設(shè)所有的東西都是分批的——我們在這里只是使用1的批大小。

import torch

# Find letter index from all_letters, e.g. "a" = 0
def letterToIndex(letter):
    return all_letters.find(letter)

# Just for demonstration, turn a letter into a <1 x n_letters> Tensor
def letterToTensor(letter):
    tensor = torch.zeros(1, n_letters)
    tensor[0][letterToIndex(letter)] = 1
    return tensor

# Turn a line into a <line_length x 1 x n_letters>,
# or an array of one-hot letter vectors
def lineToTensor(line):
    tensor = torch.zeros(len(line), 1, n_letters)
    for li, letter in enumerate(line):
        tensor[li][0][letterToIndex(letter)] = 1
    return tensor

print(letterToTensor('J'))

print(lineToTensor('Jones').size())

輸出

tensor([[0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,
         0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 1.,
         0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,
         0., 0., 0.]])
torch.Size([5, 1, 57])

創(chuàng)建網(wǎng)絡(luò)

在autograd之前，在Torch中創(chuàng)建循環(huán)神經(jīng)網(wǎng)絡(luò)涉及到在幾個(gè)時(shí)間步上克隆一層的參數(shù)。圖層包含隱藏狀態(tài)和梯度，現(xiàn)在完全由圖形本身處理。這意味著你可以以一種非?！凹兇狻钡姆绞綄?shí)現(xiàn)RNN，作為常規(guī)的前饋層。

這個(gè)RNN模塊(主要是從the PyTorch for Torch users tutorial復(fù)制的)只有2個(gè)線性層，在輸入和隱藏狀態(tài)上操作，在輸出之后有一個(gè)LogSoftmax層。

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(input_size + hidden_size, hidden_size)
        self.h2o = nn.Linear(hidden_size, output_size)
        self.softmax = nn.LogSoftmax(dim=1)

    def forward(self, input, hidden):
        combined = torch.cat((input, hidden), 1)
        hidden = self.i2h(combined)
        output = self.h2o(hidden)
        output = self.softmax(output)
        return output, hidden

    def initHidden(self):
        return torch.zeros(1, self.hidden_size)

n_hidden = 128
rnn = RNN(n_letters, n_hidden, n_categories)

為了運(yùn)行這個(gè)網(wǎng)絡(luò)的一個(gè)步驟，我們需要傳遞一個(gè)輸入(在我們的例子中，是當(dāng)前字母的張量)和一個(gè)先前的隱藏狀態(tài)(我們一開始將其初始化為零)。我們將返回輸出(每種語言的概率)和下一個(gè)隱藏狀態(tài)(我們將其保留到下一步)。

input = letterToTensor('A')
hidden = torch.zeros(1, n_hidden)

output, next_hidden = rnn(input, hidden)

為了提高效率，我們不想為每一步都創(chuàng)建一個(gè)新的張量，所以我們將使用lineToTensor而不是letterToTensor并使用切片。這可以通過預(yù)計(jì)算張量批次來進(jìn)一步優(yōu)化。

input = lineToTensor('Albert')
hidden = torch.zeros(1, n_hidden)

output, next_hidden = rnn(input[0], hidden)
print(output)

輸出

tensor([[-2.9083, -2.9270, -2.9167, -2.9590, -2.9108, -2.8332, -2.8906, -2.8325,
         -2.8521, -2.9279, -2.8452, -2.8754, -2.8565, -2.9733, -2.9201, -2.8233,
         -2.9298, -2.8624]], grad_fn=<LogSoftmaxBackward0>)

正如您所看到的，輸出是一個(gè)<1 x n_categories> 張量，其中每個(gè)項(xiàng)目是該類別的可能性(越高越有可能)。

訓(xùn)練

訓(xùn)練準(zhǔn)備

在開始訓(xùn)練之前，我們應(yīng)該編寫一些輔助函數(shù)。首先是解釋網(wǎng)絡(luò)的輸出，我們知道這是每個(gè)類別的可能性。我們可以用Tensor.topk得到最大值的索引:

def categoryFromOutput(output):
    top_n, top_i = output.topk(1)
    category_i = top_i[0].item()
    return all_categories[category_i], category_i

print(categoryFromOutput(output))

輸出

('Scottish', 15)

我們還需要一種快速獲取訓(xùn)練示例(名稱及其語言)的方法:

import random

def randomChoice(l):
    return l[random.randint(0, len(l) - 1)]

def randomTrainingExample():
    category = randomChoice(all_categories)
    line = randomChoice(category_lines[category])
    category_tensor = torch.tensor([all_categories.index(category)], dtype=torch.long)
    line_tensor = lineToTensor(line)
    return category, line, category_tensor, line_tensor

for i in range(10):
    category, line, category_tensor, line_tensor = randomTrainingExample()
    print('category =', category, '/ line =', line)

輸出

category = Chinese / line = Hou
category = Scottish / line = Mckay
category = Arabic / line = Cham
category = Russian / line = V'Yurkov
category = Irish / line = O'Keeffe
category = French / line = Belrose
category = Spanish / line = Silva
category = Japanese / line = Fuchida
category = Greek / line = Tsahalis
category = Korean / line = Chang

訓(xùn)練網(wǎng)絡(luò)

現(xiàn)在訓(xùn)練這個(gè)網(wǎng)絡(luò)所需要做的就是給它看一堆例子，讓它猜測，然后告訴它是否錯(cuò)了。

對(duì)于損失函數(shù)nn.NLLLoss是合適的，因?yàn)镽NN的最后一層是nn.LogSoftmax.。

criterion = nn.NLLLoss()

每個(gè)訓(xùn)練循環(huán)將:

創(chuàng)建輸入張量和目標(biāo)張量
創(chuàng)建一個(gè)零初始隱藏狀態(tài)
讀取每個(gè)字母

- 為下一個(gè)字母保存隱藏狀態(tài)

將最終輸出與目標(biāo)進(jìn)行比較
反向傳播
返回輸出和損失

learning_rate = 0.005 # If you set this too high, it might explode. If too low, it might not learn

def train(category_tensor, line_tensor):
    hidden = rnn.initHidden()

    rnn.zero_grad()

    for i in range(line_tensor.size()[0]):
        output, hidden = rnn(line_tensor[i], hidden)

    loss = criterion(output, category_tensor)
    loss.backward()

    # Add parameters' gradients to their values, multiplied by learning rate
    for p in rnn.parameters():
        p.data.add_(p.grad.data, alpha=-learning_rate)

    return output, loss.item()

現(xiàn)在我們只需要用一堆例子來運(yùn)行它。由于train函數(shù)返回輸出和損失，我們可以打印它的猜測并跟蹤損失以便繪制。由于有1000個(gè)示例，我們只打印每個(gè)print_every示例，并取損失的平均值。

import time
import math

n_iters = 100000
print_every = 5000
plot_every = 1000



# Keep track of losses for plotting
current_loss = 0
all_losses = []

def timeSince(since):
    now = time.time()
    s = now - since
    m = math.floor(s / 60)
    s -= m * 60
    return '%dm %ds' % (m, s)

start = time.time()

for iter in range(1, n_iters + 1):
    category, line, category_tensor, line_tensor = randomTrainingExample()
    output, loss = train(category_tensor, line_tensor)
    current_loss += loss

    # Print ``iter`` number, loss, name and guess
    if iter % print_every == 0:
        guess, guess_i = categoryFromOutput(output)
        correct = '?' if guess == category else '? (%s)' % category
        print('%d %d%% (%s) %.4f %s / %s %s' % (iter, iter / n_iters * 100, timeSince(start), loss, line, guess, correct))

    # Add current loss avg to list of losses
    if iter % plot_every == 0:
        all_losses.append(current_loss / plot_every)
        current_loss = 0

輸出

5000 5% (0m 33s) 2.6379 Horigome / Japanese ?
10000 10% (1m 5s) 2.0172 Miazga / Japanese ? (Polish)
15000 15% (1m 39s) 0.2680 Yukhvidov / Russian ?
20000 20% (2m 12s) 1.8239 Mclaughlin / Irish ? (Scottish)
25000 25% (2m 45s) 0.6978 Banh / Vietnamese ?
30000 30% (3m 18s) 1.7433 Machado / Japanese ? (Portuguese)
35000 35% (3m 51s) 0.0340 Fotopoulos / Greek ?
40000 40% (4m 23s) 1.4637 Quirke / Irish ?
45000 45% (4m 57s) 1.9018 Reier / French ? (German)
50000 50% (5m 30s) 0.9174 Hou / Chinese ?
55000 55% (6m 2s) 1.0506 Duan / Vietnamese ? (Chinese)
60000 60% (6m 35s) 0.9617 Giang / Vietnamese ?
65000 65% (7m 9s) 2.4557 Cober / German ? (Czech)
70000 70% (7m 42s) 0.8502 Mateus / Portuguese ?
75000 75% (8m 14s) 0.2750 Hamilton / Scottish ?
80000 80% (8m 47s) 0.7515 Maessen / Dutch ?
85000 85% (9m 20s) 0.0912 Gan / Chinese ?
90000 90% (9m 53s) 0.1190 Bellomi / Italian ?
95000 95% (10m 26s) 0.0137 Vozgov / Russian ?
100000 100% (10m 59s) 0.7808 Tong / Vietnamese ?

繪制結(jié)果

繪制all_losses的歷史損失圖顯示了網(wǎng)絡(luò)的學(xué)習(xí)情況:

import matplotlib.pyplot as plt
import matplotlib.ticker as ticker

plt.figure()
plt.plot(all_losses)

PyTorch翻譯官網(wǎng)教程-NLP FROM SCRATCH: CLASSIFYING NAMES WITH A CHARACTER-LEVEL RNN,深度學(xué)習(xí),pytorch,自然語言處理,rnn

輸出

[<matplotlib.lines.Line2D object at 0x7f16606095a0>]

評(píng)估結(jié)果

為了了解網(wǎng)絡(luò)在不同類別上的表現(xiàn)如何，我們將創(chuàng)建一個(gè)混淆矩陣，表示網(wǎng)絡(luò)猜測(列)的每種語言(行)。為了計(jì)算混淆矩陣，使用evaluate(),在網(wǎng)絡(luò)中運(yùn)行一堆樣本，這與 train() 去掉反向傳播相同。

# Keep track of correct guesses in a confusion matrix
confusion = torch.zeros(n_categories, n_categories)
n_confusion = 10000

# Just return an output given a line
def evaluate(line_tensor):
    hidden = rnn.initHidden()

    for i in range(line_tensor.size()[0]):
        output, hidden = rnn(line_tensor[i], hidden)

    return output

# Go through a bunch of examples and record which are correctly guessed
for i in range(n_confusion):
    category, line, category_tensor, line_tensor = randomTrainingExample()
    output = evaluate(line_tensor)
    guess, guess_i = categoryFromOutput(output)
    category_i = all_categories.index(category)
    confusion[category_i][guess_i] += 1

# Normalize by dividing every row by its sum
for i in range(n_categories):
    confusion[i] = confusion[i] / confusion[i].sum()

# Set up plot
fig = plt.figure()
ax = fig.add_subplot(111)
cax = ax.matshow(confusion.numpy())
fig.colorbar(cax)

# Set up axes
ax.set_xticklabels([''] + all_categories, rotation=90)
ax.set_yticklabels([''] + all_categories)

# Force label at every tick
ax.xaxis.set_major_locator(ticker.MultipleLocator(1))
ax.yaxis.set_major_locator(ticker.MultipleLocator(1))

# sphinx_gallery_thumbnail_number = 2
plt.show()

PyTorch翻譯官網(wǎng)教程-NLP FROM SCRATCH: CLASSIFYING NAMES WITH A CHARACTER-LEVEL RNN,深度學(xué)習(xí),pytorch,自然語言處理,rnn

輸出

/var/lib/jenkins/workspace/intermediate_source/char_rnn_classification_tutorial.py:445: UserWarning:

FixedFormatter should only be used together with FixedLocator

/var/lib/jenkins/workspace/intermediate_source/char_rnn_classification_tutorial.py:446: UserWarning:

FixedFormatter should only be used together with FixedLocator

你可以從主軸上挑出亮點(diǎn)，顯示它猜錯(cuò)了哪些語言，例如中文猜錯(cuò)了韓語，西班牙語猜錯(cuò)了意大利語。它似乎在希臘語上表現(xiàn)得很好，而在英語上表現(xiàn)得很差(可能是因?yàn)榕c其他語言重疊)。

運(yùn)行用戶輸入

def predict(input_line, n_predictions=3):
    print('\n> %s' % input_line)
    with torch.no_grad():
        output = evaluate(lineToTensor(input_line))

        # Get top N categories
        topv, topi = output.topk(n_predictions, 1, True)
        predictions = []

        for i in range(n_predictions):
            value = topv[0][i].item()
            category_index = topi[0][i].item()
            print('(%.2f) %s' % (value, all_categories[category_index]))
            predictions.append([value, all_categories[category_index]])

predict('Dovesky')
predict('Jackson')
predict('Satoshi')

輸出

> Dovesky
(-0.57) Czech
(-0.97) Russian
(-3.43) English

> Jackson
(-1.02) Scottish
(-1.49) Russian
(-1.96) English

> Satoshi
(-0.42) Japanese
(-1.70) Polish
(-2.74) Italian

in the Practical PyTorch repo中腳本的最終版本將上述代碼拆分為幾個(gè)文件:

data.py (加載文件)
model.py (定義 RNN)
train.py (執(zhí)行訓(xùn)練)
predict.py (運(yùn)行帶有命令行參數(shù)的predict() )
server.py (使用bottle.py作為JSON API提供預(yù)測)

運(yùn)行train.py來訓(xùn)練和保存網(wǎng)絡(luò)。

運(yùn)行predict.py并輸入一個(gè)名稱來查看預(yù)測:

$ python predict.py Hazaki
(-0.42) Japanese
(-1.39) Polish
(-3.51) Czech

運(yùn)行server.py 并訪問http://localhost:5533/Yourname以獲得預(yù)測的JSON輸出。

練習(xí)

嘗試使用不同的數(shù)據(jù)集 -> 類別，例如:

任何單詞->語言
名字->性別
角色名稱->作家
頁面標(biāo)題 -> 博客或社交新聞網(wǎng)站子版塊

使用一個(gè)更大的和/或更好的形狀網(wǎng)絡(luò)，可以獲得更好的結(jié)果文章來源地址http://www.zghlxwxcb.cn/news/detail-651479.html

添加更多線性圖層
試試 nn.LSTM 和 nn.GRU 網(wǎng)絡(luò)層
將這些RNNs組合成一個(gè)更高級(jí)的網(wǎng)絡(luò)

到了這里，關(guān)于PyTorch翻譯官網(wǎng)教程-NLP FROM SCRATCH: CLASSIFYING NAMES WITH A CHARACTER-LEVEL RNN的文章就介紹完了。如果您還想了解更多內(nèi)容，請?jiān)谟疑辖撬阉鱐OY模板網(wǎng)以前的文章或繼續(xù)瀏覽下面的相關(guān)文章，希望大家以后多多支持TOY模板網(wǎng)！