lstm from scratch

前言

pytorch官网做的是名字生成的任务。
tutorial里是自定义的rnn，我自定义了一个最简单的lstm。
lstm模型参考的是Understanding LSTM Networks

完整实验过程

import torch
import torch.nn as nn

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 + " .,;'-"
n_letters = len(all_letters) + 1 # Plus EOS marker

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

# 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
    )

# 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]

# Build the category_lines dictionary, a list of lines per category
category_lines = {
    
    }
all_categories = []
for filename in findFiles('data/name_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)

if n_categories == 0:
    raise RuntimeError('Data not found. Make sure that you downloaded data '
        'from https://download.pytorch.org/tutorial/data.zip and extract it to '
        'the current directory.')

print('# categories:', n_categories, all_categories)

# categories: 18 ['Arabic', 'Chinese', 'Czech', 'Dutch', 'English', 'French', 'German', 'Greek', 'Irish', 'Italian', 'Japanese', 'Korean', 'Polish', 'Portuguese', 'Russian', 'Scottish', 'Spanish', 'Vietnamese']

# 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)

import random

# Random item from a list
def randomChoice(l):
    return l[random.randint(0, len(l) - 1)]
#     return l[0]

# Get a random category and random line from that category
def randomTrainingPair():
    category = randomChoice(all_categories)
    
    line = randomChoice(category_lines[category])
    return category, line

# One-hot vector for category
def categoryTensor(category):
    li = all_categories.index(category)
    tensor = torch.zeros(1, n_categories)
    tensor[0][li] = 1
    return tensor

# One-hot matrix of first to last letters (not including EOS) for input
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

# LongTensor of second letter to end (EOS) for target
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)

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

class LSTm(nn.Module):
    def __init__(self, input_size, hidden_size, output_size):
        super(LSTm, self).__init__()
        self.hidden_size = hidden_size 
        self.input_size = input_size
        self.output_size = output_size
    
        self.w_f = torch.nn.Linear(self.input_size, hidden_size)
        nn.init.xavier_normal_(self.w_f.weight,gain=1)
        self.w_i = torch.nn.Linear(self.input_size, hidden_size)
        nn.init.xavier_normal_(self.w_i.weight,gain=1)
        self.w_c = torch.nn.Linear(self.input_size, hidden_size)
        nn.init.xavier_normal_(self.w_c.weight,gain=1)
        self.w_o = torch.nn.Linear(self.input_size, hidden_size)
        nn.init.xavier_normal_(self.w_o.weight,gain=1)
        
        self.sigmoid = torch.nn.Sigmoid()
        self.tanh = torch.nn.Tanh()
        self.outlayer = torch.nn.Linear(hidden_size, output_size)
        self.softmax = nn.LogSoftmax(dim=1)
    
    def forward(self, category, x, h, c):
        
        x = torch.cat([category, x], dim = 1)
        
        ft = self.sigmoid(self.w_f(torch.cat([h, x], dim=1)))
        
        it = self.sigmoid(self.w_i(torch.cat([h, x], dim=1)))
        
        ct_hat = self.tanh(self.w_c(torch.cat([h, x], dim=1)))
        
        ct = torch.mul(ft, c) + torch.mul(it, ct_hat)
        
        ot = self.sigmoid(self.w_o(torch.cat([h, x], dim=1)))
        
        ht = torch.mul(ot, ct)

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

hidden_size = 128
out_size = n_letters
criterion = nn.CrossEntropyLoss()
LStm = LSTm(n_letters+n_categories+hidden_size, hidden_size, out_size)
optimizer = torch.optim.Adam(LStm.parameters(), lr = 1e-3)

def train(category_tensor, input_line_tensor, target_line_tensor):
    target_line_tensor.unsqueeze_(-1)
    hidden = LStm.initHidden()
    cell = LStm.initCell()
    optimizer.zero_grad()

    loss = 0

    for i in range(input_line_tensor.size(0)):
        hidden, cell, output = LStm(category_tensor, hidden, input_line_tensor[i], cell)
        l = criterion(output, target_line_tensor[i])
        loss += l
    loss.backward()
    optimizer.step()

    return output, loss.item() / input_line_tensor.size(0)

import time
import math

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

# rnn = RNN(n_letters, 128, n_letters)
n_iters = 100000
print_every = 5000
plot_every = 500
all_losses = []
total_loss = 0 # Reset every plot_every iters

start = time.time()

for iter in range(1, n_iters + 1):
    output, loss = train(*randomTrainingExample())
    total_loss += loss

    if iter % print_every == 0:
#         print(output)
        print('%s (%d %d%%) %.4f' % (timeSince(start), iter, iter / n_iters * 100, loss))


    if iter % plot_every == 0:
        all_losses.append(total_loss / plot_every)
        total_loss = 0

0m 25s (5000 5%) 2.5861
0m 51s (10000 10%) 3.0035
1m 20s (15000 15%) 2.2847
1m 48s (20000 20%) 2.0237
2m 15s (25000 25%) 2.9981
2m 39s (30000 30%) 2.7145
3m 4s (35000 35%) 2.7130
3m 28s (40000 40%) 2.2639
3m 52s (45000 45%) 2.0530
4m 17s (50000 50%) 2.8611
4m 41s (55000 55%) 3.0305
5m 7s (60000 60%) 1.9455
5m 32s (65000 65%) 2.3996
5m 57s (70000 70%) 2.2090
6m 31s (75000 75%) 1.8699
7m 7s (80000 80%) 2.2690
7m 35s (85000 85%) 2.3699
8m 1s (90000 90%) 1.6409
8m 26s (95000 95%) 2.3424
8m 50s (100000 100%) 2.1979

import matplotlib.pyplot as plt
import matplotlib.ticker as ticker
plt.figure("train loss")
plt.plot(all_losses)
plt.title("Lstm train loss")
plt.ylabel("loss")
plt.xlabel("iter")

hidden_size与out_size不同，增加了一个线性层(这个不对，输出维度设置错了):

这个输出维度没错，
增加训练轮数:

max_length = 20

# Sample from a category and starting letter
def sample(category, start_letter='A'):
    with torch.no_grad():  # no need to track history in sampling
        category_tensor = categoryTensor(category)
        input = inputTensor(start_letter)
        hidden = LStm.initHidden()
        cell = LStm.initCell()

        output_name = start_letter
        # category_tensor, hidden, input_line_tensor[i], cell
        for i in range(max_length):
            hidden, cell, output = LStm(category_tensor, hidden, input[0], cell)
            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

# Get multiple samples from one category and multiple starting letters
def samples(category, start_letters='ABC'):
    for start_letter in start_letters:
        print(sample(category, start_letter))
samples('Russian', 'abcdefghijklmn')

anera
balev
chalon
daranov
erinon
falin
galov
halanov
intono
jareva
kolovak
loukha
malov
nolouki

结论与思考

• 看官网自定义的RNN效果也不是很好，下面是它的loss曲线:
  
• 对线性层的参数用nn.init.xavier_normal_进行初始化之后，loss下降的比较快
• input_size, hidden_size, output_size那里还有待思考
• 用了crossentropy就不要用logsoftmax了，cross是 softmax+log+nllloss的结合

参考

Pytorch详解NLLLoss和CrossEntropyLoss
Understanding LSTM Networks
pytorch官网