import collections
import math
import re
import random
import torch
from torch import nn
from torch.nn import functional as F
from d2l import torch as d2ldef read_txt():# 读取文本数据with open('./A Study in Drowning.txt', 'r', encoding='utf-8') as f:# 读取每一行lines = f.readlines()# 将不是英文字符的转换为空格，全部变为小写字符返回return [re.sub('[^A-Za-z]+', ' ', line).strip().lower() for line in lines]def tokenize(lines, token='word'):# 将文本以空格进行分割成词if token == 'word':return [line.split() for line in lines]# 将文本分割成字符elif token == 'char':return [list(line) for line in lines]else:print('错误：未知令牌类型：' + token)def count_corpus(tokens):# 如果tokens长度为0或tokens[0]是listif len(tokens) == 0 or isinstance(tokens[0], list):# 将[[,,,],[,,,]]多层结构变为一层结构[,,,,]tokens = [token for line in tokens for token in line]# 统计可迭代对象中元素出现的次数，并返回一个字典（key-value）key 表示元素，value 表示各元素 key 出现的次数return collections.Counter(tokens)# idx_to_token  是一个list  由token作为元素构成['<unk>', ' ', 'e', 't', 'a', 'o', 'h', 'n', 'i', 's', 'r', 'd', 'l', 'u', 'f', 'w', 'g', 'm', 'y', 'c', 'p', 'b', 'k', 'v', 'j', 'x', 'z', 'q']
# token_freqs   是一个list  由token和该token出现的次数构成的元组作为元素构成[(' ', 94824), ('e', 54804), ('t', 38742), ('a', 33172), ('o', 30656), ('h', 29047), ('n', 28667), ('i', 28093), ('s', 27922), ('r', 26121), ('d', 20394), ('l', 17755), ('u', 12267), ('f', 11033), ('w', 10033), ('g', 9837), ('m', 9258), ('y', 9251), ('c', 8872), ('p', 6998), ('b', 6620), ('k', 4817), ('v', 3574), ('j', 500), ('x', 372), ('z', 308), ('q', 285)]
# token_to_idx  是一个dict  由token作为key token在idx_to_token的索引作为value构成{' ': 1, '<unk>': 0, 'a': 4, 'b': 21, 'c': 19, 'd': 11, 'e': 2, 'f': 14, 'g': 16, 'h': 6, 'i': 8, 'j': 24, 'k': 22, 'l': 12, 'm': 17, 'n': 7, 'o': 5, 'p': 20, 'q': 27, 'r': 10, 's': 9, 't': 3, 'u': 13, 'v': 23, 'w': 15, 'x': 25, 'y': 18, 'z': 26}
class Vocab:def __init__(self, tokens=None, min_freq=0, reserved_tokens=None):# 处理特殊情况if tokens is None:tokens = []# 处理特殊情况if reserved_tokens is None:reserved_tokens = []# counter为一个字典（key-value）key 表示元素，value 表示各元素 key 出现的次数counter = count_corpus(tokens)# 排序# iterable：待排序的序列counter.items()# key：排序规则lambda x: x[1]从小到大# reverse：指定排序的方式，默认值False，即升序排列，这是True也就是降序self.token_freqs = sorted(counter.items(), key=lambda x: x[1], reverse=True)# 初始化self.unk, uniq_tokens = 0, ['<unk>'] + reserved_tokens# 初始化  token_freqs中 key不在 uniq_tokens中  且  value大于min_freq 返回token放入uniq_tokensuniq_tokens += [token for token, freq in self.token_freqsif freq >= min_freq and token not in uniq_tokens]# 初始化self.idx_to_token, self.token_to_idx = [], dict()# 赋值for token in uniq_tokens:self.idx_to_token.append(token)self.token_to_idx[token] = len(self.idx_to_token) - 1def __len__(self):return len(self.idx_to_token)def __getitem__(self, tokens):if not isinstance(tokens, (list, tuple)):return self.token_to_idx.get(tokens, self.unk)return [self.__getitem__(token) for token in tokens]def to_tokens(self, indices):if not isinstance(indices, (list, tuple)):return self.idx_to_token[indices]return [self.idx_to_token[index] for index in indices]def load_corpus_time_machine(max_tokens=-1):# 将文本处理成行lines = read_txt()# print(lines)# 将行tokens化tokens = tokenize(lines, 'char')# print(tokens)# 构建字典表vocab = Vocab(tokens)# vocab的格式为{list:524222}[5, 7, 2, 5, 7, 2, 8, 3, ......, 1, 18, 5, 13]#print(vocab)corpus = [vocab[token] for line in tokens for token in line]if max_tokens > 0:corpus = corpus[:max_tokens]return corpus, vocab# 随机地生成一个小批量数据的特征和标签以供读取。 在随机采样中，每个样本都是在原始的长序列上任意捕获的子序列
def seq_data_iter_random(corpus, batch_size, num_steps):"""使用随机抽样生成一个小批量子序列。"""corpus = corpus[random.randint(0, num_steps - 1):]num_subseqs = (len(corpus) - 1) // num_stepsinitial_indices = list(range(0, num_subseqs * num_steps, num_steps))random.shuffle(initial_indices)def data(pos):return corpus[pos:pos + num_steps]num_batches = num_subseqs // batch_sizefor i in range(0, batch_size * num_batches, batch_size):initial_indices_per_batch = initial_indices[i:i + batch_size]X = [data(j) for j in initial_indices_per_batch]Y = [data(j + 1) for j in initial_indices_per_batch]yield torch.tensor(X), torch.tensor(Y)# 保证两个相邻的小批量中的子序列在原始序列上也是相邻的
def seq_data_iter_sequential(corpus, batch_size, num_steps):"""使用顺序分区生成一个小批量子序列。"""offset = random.randint(0, num_steps)num_tokens = ((len(corpus) - offset - 1) // batch_size) * batch_sizeXs = torch.tensor(corpus[offset:offset + num_tokens])Ys = torch.tensor(corpus[offset + 1:offset + 1 + num_tokens])Xs, Ys = Xs.reshape(batch_size, -1), Ys.reshape(batch_size, -1)num_batches = Xs.shape[1] // num_stepsfor i in range(0, num_steps * num_batches, num_steps):X = Xs[:, i:i + num_steps]Y = Ys[:, i:i + num_steps]yield X, Yclass SeqDataLoader:"""加载序列数据的迭代器。"""def __init__(self, batch_size, num_steps, use_random_iter, max_tokens):if use_random_iter:self.data_iter_fn = seq_data_iter_randomelse:self.data_iter_fn = seq_data_iter_sequentialself.corpus, self.vocab = load_corpus_time_machine(max_tokens)self.batch_size, self.num_steps = batch_size, num_stepsdef __iter__(self):return self.data_iter_fn(self.corpus, self.batch_size, self.num_steps)def load_data_time_machine(batch_size, num_steps,use_random_iter=False, max_tokens=10000):"""返回时光机器数据集的迭代器和词汇表。"""data_iter = SeqDataLoader(batch_size, num_steps, use_random_iter,max_tokens)return data_iter, data_iter.vocab# 初始化模型参数
def get_params(vocab_size, num_hiddens, device):# 输入等于输出等于字典大小num_inputs = num_outputs = vocab_size# 均值为0方差为1的随机张量*0.01def normal(shape):return torch.randn(size=shape, device=device) * 0.01# 输入到隐藏层边缘的WW_xh = normal((num_inputs, num_hiddens))# 隐藏层的WW_hh = normal((num_hiddens, num_hiddens))b_h = torch.zeros(num_hiddens, device=device)# 隐藏层到输出的WW_hq = normal((num_hiddens, num_outputs))b_q = torch.zeros(num_outputs, device=device)params = [W_xh, W_hh, b_h, W_hq, b_q]for param in params:param.requires_grad_(True)return params# 初始化隐藏状态
def init_rnn_state(batch_size, num_hiddens, device):# 批量大小，隐藏层大小的全0张量return (torch.zeros((batch_size, num_hiddens), device=device),)# 计算输出
def rnn(inputs, state, params):W_xh, W_hh, b_h, W_hq, b_q = paramsH, = stateoutputs = []for X in inputs:# 激活函数是tanh H为初始化隐藏状态H = torch.tanh(torch.mm(X, W_xh) + torch.mm(H, W_hh) + b_h)Y = torch.mm(H, W_hq) + b_qoutputs.append(Y)# H为当前隐藏状态return torch.cat(outputs, dim=0), (H,)class RNNModelScratch:"""从零开始实现的循环神经网络模型"""def __init__(self, vocab_size, num_hiddens, device, get_params,init_state, forward_fn):self.vocab_size, self.num_hiddens = vocab_size, num_hiddensself.params = get_params(vocab_size, num_hiddens, device)self.init_state, self.forward_fn = init_state, forward_fndef __call__(self, X, state):X = F.one_hot(X.T, self.vocab_size).type(torch.float32)return self.forward_fn(X, state, self.params)def begin_state(self, batch_size, device):return self.init_state(batch_size, self.num_hiddens, device)# 推理测试
def predict_ch8(prefix, num_preds, net, vocab, device):"""在`prefix`后面生成新字符。"""state = net.begin_state(batch_size=1, device=device)outputs = [vocab[prefix[0]]]get_input = lambda: torch.tensor([outputs[-1]], device=device).reshape((1, 1))for y in prefix[1:]:_, state = net(get_input(), state)outputs.append(vocab[y])for _ in range(num_preds):y, state = net(get_input(), state)outputs.append(int(y.argmax(dim=1).reshape(1)))return ''.join([vocab.idx_to_token[i] for i in outputs])# 梯度剪裁
def grad_clipping(net, theta):"""裁剪梯度。"""if isinstance(net, nn.Module):params = [p for p in net.parameters() if p.requires_grad]else:params = net.paramsnorm = torch.sqrt(sum(torch.sum((p.grad**2)) for p in params))if norm > theta:for param in params:param.grad[:] *= theta / norm# 训练函数
def train_epoch_ch8(net, train_iter, loss, updater, device, use_random_iter):"""训练模型一个迭代周期（定义见第8章）。"""state = Nonemetric = d2l.Accumulator(2)for X, Y in train_iter:if state is None or use_random_iter:state = net.begin_state(batch_size=X.shape[0], device=device)else:if isinstance(net, nn.Module) and not isinstance(state, tuple):state.detach_()else:for s in state:s.detach_()y = Y.T.reshape(-1)X, y = X.to(device), y.to(device)y_hat, state = net(X, state)l = loss(y_hat, y.long()).mean()if isinstance(updater, torch.optim.Optimizer):updater.zero_grad()l.backward()grad_clipping(net, 1)updater.step()else:l.backward()grad_clipping(net, 1)updater(batch_size=1)metric.add(l * y.numel(), y.numel())return math.exp(metric[0] / metric[1])def train_ch8(net, train_iter, vocab, lr, num_epochs, device,use_random_iter=False):"""训练模型（定义见第8章）。"""loss = nn.CrossEntropyLoss()if isinstance(net, nn.Module):updater = torch.optim.SGD(net.parameters(), lr)else:updater = lambda batch_size: d2l.sgd(net.params, lr, batch_size)predict = lambda prefix: predict_ch8(prefix, 50, net, vocab, device)for epoch in range(num_epochs):ppl = train_epoch_ch8(net, train_iter, loss, updater, device,use_random_iter)if (epoch + 1) % 10 == 0:print(predict('But'))print(f'困惑度 {ppl:.1f},  {str(device)}')print(predict('But'))# 批量大小为32  时序序列的长度为35 隐藏层大小512
batch_size, num_steps, num_hiddens = 32, 35, 512
# 获取迭代数据和字典
train_iter, vocab = load_data_time_machine(batch_size, num_steps)
# 定义网络
net = RNNModelScratch(len(vocab), num_hiddens, torch.device('cpu'), get_params,init_rnn_state, rnn)
# 训练500轮 学习率为1
num_epochs, lr = 50, 1
# 训练
train_ch8(net, train_iter, vocab, lr, num_epochs, torch.device('cpu'),use_random_iter=True)

训练结果

<unk>ut the the the the the the the the the the the the t
<unk>uthe the the the the the the the the the the the the
<unk>uthe sher and the sher and the sher and the sher and
<unk>uthe sher and the sher and the sher and the sher and
<unk>uthe sher and he her sher and her sher and her sher 
困惑度 8.8,  cpu
<unk>uthe sher and he her sher and her sher and her sher Process finished with exit code 0