from __future__ import unicode_literals, print_function, division
from io import open
import unicodedata
import re
import random

import torch
import torch.nn as nn
from torch import optim
import torch.nn.functional as F

import numpy as np
from torch.utils.data import TensorDataset, DataLoader, RandomSampler
%matplotlib inline  

device = torch.device("cuda" if torch.cuda.is_available() else "cpu")

# 定义句子开始和结束的标记符（Start Of Sentence 和 End Of Sentence）
SOS_token = 0 
EOS_token = 1   

class Lang:
    def __init__(self, name):
        self.name = name  # 语言名称（如"English"或"French"）
        self.word2index = {}  # 单词到索引的映射字典（如{"hello": 2}）
        self.word2count = {}  # 单词出现次数的统计字典（如{"hello": 5}）
        self.index2word = {0: "SOS", 1: "EOS"}  # 索引到单词的映射字典（保留0和1给SOS/EOS）
        self.n_words = 2  # 当前词汇表大小（初始为2，因为已经包含SOS和EOS）

    def addSentence(self, sentence):
        # 将句子分割成单词，并逐个添加到词汇表中
        for word in sentence.split(' '):  # 按空格分割句子
            self.addWord(word)  # 调用addWord方法添加每个单词

    def addWord(self, word):
        # 添加单个单词到词汇表
        if word not in self.word2index:  # 如果单词是新词（未在词汇表中）
            self.word2index[word] = self.n_words  # 为新词分配下一个可用索引
            self.word2count[word] = 1  # 初始化该单词的计数为1
            self.index2word[self.n_words] = word  # 建立反向映射（索引→单词）
            self.n_words += 1  # 增加词汇表计数
        else:  # 如果单词已存在
            self.word2count[word] += 1  # 只增加该单词的计数

# 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'
    )

# Lowercase, trim, and remove non-letter characters
def normalizeString(s):
    s = unicodeToAscii(s.lower().strip())
    s = re.sub(r"([.!?])", r" \1", s)
    s = re.sub(r"[^a-zA-Z!?]+", r" ", s)
    return s.strip()

def readLangs(lang1, lang2, reverse=False):
    """读取双语文本文件，处理并返回语言对象和句子对"""
    
    print("Reading lines...")

    # 读取文件并分割成行
    lines = open('data/%s-%s.txt' % (lang1, lang2), encoding='utf-8').read().strip().split('\n')

    # 将每行分割成句子对并进行标准化处理
    pairs = [[normalizeString(s) for s in l.split('\t')] for l in lines]

    # 根据reverse参数决定语言方向
    if reverse:
        pairs = [list(reversed(p)) for p in pairs]
        input_lang = Lang(lang2)
        output_lang = Lang(lang1)
    else:
        input_lang = Lang(lang1)
        output_lang = Lang(lang2)

    return input_lang, output_lang, pairs

MAX_LENGTH = 10

eng_prefixes = (
    "i am ", "i m ",
    "he is", "he s ",
    "she is", "she s ",
    "you are", "you re ",
    "we are", "we re ",
    "they are", "they re "
)

# 返回过滤句子对：确保两个句子长度都小于MAX_LENGTH且目标句子以特定前缀开头,
def filterPair(p):
    return len(p[0].split(' ')) < MAX_LENGTH and \
        len(p[1].split(' ')) < MAX_LENGTH and \
        p[1].startswith(eng_prefixes)

# 过滤所有句子对，只保留符合filterPair条件的
def filterPairs(pairs):
    return [pair for pair in pairs if filterPair(pair)]

def prepareData(lang1, lang2, reverse=False):
    """准备训练数据：读取数据、过滤句子对、构建词汇表"""
    # 1. 读取原始双语数据
    input_lang, output_lang, pairs = readLangs(lang1, lang2, reverse)
    print("Read %s sentence pairs" % len(pairs))
    
    # 2. 过滤过长的句子和不符合格式的句子
    pairs = filterPairs(pairs)
    print("Trimmed to %s sentence pairs" % len(pairs))
    
    # 3. 构建两种语言的词汇表
    print("Counting words...")
    for pair in pairs:
        input_lang.addSentence(pair[0])  # 将源语言句子加入词汇表
        output_lang.addSentence(pair[1])  # 将目标语言句子加入词汇表
    
    # 4. 打印词汇表信息
    print("Counted words:")
    print(input_lang.name, input_lang.n_words)
    print(output_lang.name, output_lang.n_words)
    
    return input_lang, output_lang, pairs

# 使用示例：准备英法翻译数据（反向模式）
input_lang, output_lang, pairs = prepareData('eng', 'fra', True)
print(random.choice(pairs))  # 随机打印一个预处理后的句子对

class EncoderRNN(nn.Module):
    def __init__(self, input_size, hidden_size, dropout_p=0.1):
        super(EncoderRNN, self).__init__()
        self.hidden_size = hidden_size

        self.embedding = nn.Embedding(input_size, hidden_size)
        self.gru = nn.GRU(hidden_size, hidden_size, batch_first=True)
        self.dropout = nn.Dropout(dropout_p)

    def forward(self, input):
        embedded = self.dropout(self.embedding(input))
        output, hidden = self.gru(embedded)
        return output, hidden

class DecoderRNN(nn.Module):
    """基于GRU的RNN解码器，用于序列生成任务（如机器翻译）"""
    
    def __init__(self, hidden_size, output_size):
        """
        初始化解码器
        :param hidden_size: 隐藏层维度（与编码器保持一致）
        :param output_size: 输出词汇表大小（目标语言词表大小）
        """
        super(DecoderRNN, self).__init__()
        # 词嵌入层：将目标语言的单词索引映射为稠密向量
        self.embedding = nn.Embedding(output_size, hidden_size)
        # GRU层：处理序列生成（输入和隐藏维度相同）
        self.gru = nn.GRU(hidden_size, hidden_size, batch_first=True)
        # 输出层：将隐藏状态映射到目标词表空间
        self.out = nn.Linear(hidden_size, output_size)

    def forward(self, encoder_outputs, encoder_hidden, target_tensor=None):
        """
        前向传播（生成目标序列）
        :param encoder_outputs: 编码器输出（通常未直接使用）
        :param encoder_hidden: 编码器的最终隐藏状态（作为解码器初始状态）
        :param target_tensor: 目标序列（用于Teacher Forcing训练模式）
        :return: (输出序列概率, 最终隐藏状态, None)
        """
        # 初始化：输入为<SOS>标记，隐藏状态继承自编码器
        batch_size = encoder_outputs.size(0)
        decoder_input = torch.empty(batch_size, 1, dtype=torch.long, device=device).fill_(SOS_token)
        decoder_hidden = encoder_hidden
        decoder_outputs = []  # 保存每一步的输出

        # 循环生成序列（最多MAX_LENGTH步）
        for i in range(MAX_LENGTH):
            # 单步解码：输入当前词和隐藏状态，得到输出和新状态
            decoder_output, decoder_hidden = self.forward_step(decoder_input, decoder_hidden)
            decoder_outputs.append(decoder_output)

            # 决定下一步的输入（训练和预测模式不同）
            if target_tensor is not None:
                # Teacher Forcing模式：使用真实目标词作为下一步输入
                decoder_input = target_tensor[:, i].unsqueeze(1)
            else:
                # 预测模式：使用模型预测的概率最高词作为下一步输入
                _, topi = decoder_output.topk(1)  # 取概率最大的词索引
                decoder_input = topi.squeeze(-1).detach()  # 切断梯度反向传播

        # 后处理：拼接所有输出并计算log softmax
        decoder_outputs = torch.cat(decoder_outputs, dim=1)
        decoder_outputs = F.log_softmax(decoder_outputs, dim=-1)
        return decoder_outputs, decoder_hidden, None  # 返回None保持训练接口统一

    def forward_step(self, input, hidden):
        """单步解码过程"""
        output = self.embedding(input)  # 词嵌入
        output = F.relu(output)  # 激活函数
        output, hidden = self.gru(output, hidden)  # GRU处理
        output = self.out(output)  # 映射到词表空间
        return output, hidden

embedded = self.dropout(self.embedding(input))

query = hidden.permute(1, 0, 2)
context, attn_weights = self.attention(query, encoder_outputs)

input_gru = torch.cat((embedded, context), dim=2)

output, hidden = self.gru(input_gru, hidden)
output = self.out(output)

return output, hidden, attn_weights

class BahdanauAttention(nn.Module):
    """Bahdanau注意力机制(加性注意力)"""
    
    def __init__(self, hidden_size):
        """
        初始化注意力层
        :param hidden_size: 隐藏层维度(必须与编码器/解码器保持一致)
        """
        super(BahdanauAttention, self).__init__()
        # 三个线性变换层(无偏置)
        self.Wa = nn.Linear(hidden_size, hidden_size)  # 解码器隐藏状态变换
        self.Ua = nn.Linear(hidden_size, hidden_size)  # 编码器输出变换
        self.Va = nn.Linear(hidden_size, 1)  # 计算注意力分数

    def forward(self, query, keys):
        """
        计算注意力上下文向量
        :param query: 解码器当前隐藏状态 [batch_size, 1, hidden_size]
        :param keys: 编码器所有输出 [batch_size, seq_len, hidden_size]
        :return: (上下文向量, 注意力权重)
        """
        # 计算注意力分数
        scores = self.Va(torch.tanh(self.Wa(query) + self.Ua(keys)))  # [batch_size, seq_len, 1]
        scores = scores.squeeze(2).unsqueeze(1)  # 调整维度 [batch_size, 1, seq_len]
        
        # 计算注意力权重和上下文向量
        weights = F.softmax(scores, dim=-1)  # 沿序列维度归一化
        context = torch.bmm(weights, keys)  # 加权求和 [batch_size, 1, hidden_size]
        
        return context, weights


class AttnDecoderRNN(nn.Module):
    """带注意力机制的RNN解码器"""
    
    def __init__(self, hidden_size, output_size, dropout_p=0.1):
        """
        初始化解码器
        :param hidden_size: 隐藏层维度
        :param output_size: 输出词表大小
        :param dropout_p: dropout概率
        """
        super(AttnDecoderRNN, self).__init__()
        self.embedding = nn.Embedding(output_size, hidden_size)  # 词嵌入层
        self.attention = BahdanauAttention(hidden_size)  # 注意力层
        # GRU输入维度为hidden_size*2(因为拼接了嵌入向量和上下文向量)
        self.gru = nn.GRU(2 * hidden_size, hidden_size, batch_first=True)
        self.out = nn.Linear(hidden_size, output_size)  # 输出层
        self.dropout = nn.Dropout(dropout_p)  # Dropout层

    def forward(self, encoder_outputs, encoder_hidden, target_tensor=None):
        """
        前向传播(带注意力机制的序列生成)
        :param encoder_outputs: 编码器所有输出 [batch_size, seq_len, hidden_size]
        :param encoder_hidden: 编码器最终隐藏状态 [num_layers, batch_size, hidden_size]
        :param target_tensor: 目标序列(用于Teacher Forcing)
        :return: (输出序列概率, 最终隐藏状态, 注意力权重矩阵)
        """
        batch_size = encoder_outputs.size(0)
        # 初始化解码器输入(<SOS>标记)和隐藏状态
        decoder_input = torch.empty(batch_size, 1, dtype=torch.long, device=device).fill_(SOS_token)
        decoder_hidden = encoder_hidden
        decoder_outputs = []  # 保存所有输出
        attentions = []  # 保存所有注意力权重

        # 逐步生成序列(最多MAX_LENGTH步)
        for i in range(MAX_LENGTH):
            decoder_output, decoder_hidden, attn_weights = self.forward_step(
                decoder_input, decoder_hidden, encoder_outputs
            )
            decoder_outputs.append(decoder_output)
            attentions.append(attn_weights)

            # 决定下一步输入(Teacher Forcing或自回归)
            if target_tensor is not None:
                decoder_input = target_tensor[:, i].unsqueeze(1)  # Teacher Forcing
            else:
                _, topi = decoder_output.topk(1)
                decoder_input = topi.squeeze(-1).detach()  # 自回归生成

        # 后处理
        decoder_outputs = torch.cat(decoder_outputs, dim=1)  # 拼接所有输出 [batch_size, seq_len, output_size]
        decoder_outputs = F.log_softmax(decoder_outputs, dim=-1)  # 计算log概率
        attentions = torch.cat(attentions, dim=1)  # 拼接注意力权重 [batch_size, tgt_len, src_len]
        
        return decoder_outputs, decoder_hidden, attentions

    def forward_step(self, input, hidden, encoder_outputs):
        """单步解码过程"""
        embedded = self.dropout(self.embedding(input))  # 词嵌入+dropout [batch_size, 1, hidden_size]
        
        # 计算注意力上下文向量
        query = hidden.permute(1, 0, 2)  # 调整hidden形状匹配注意力层 [batch_size, 1, hidden_size]
        context, attn_weights = self.attention(query, encoder_outputs)
        
        # 拼接嵌入向量和上下文向量作为GRU输入
        input_gru = torch.cat((embedded, context), dim=2)  # [batch_size, 1, hidden_size*2]
        
        # GRU处理和输出映射
        output, hidden = self.gru(input_gru, hidden)
        output = self.out(output)  # [batch_size, 1, output_size]
        
        return output, hidden, attn_weights

def indexesFromSentence(lang, sentence):
    """将句子中的每个单词转换为对应的索引列表"""
    # lang: 包含word2index词典的语言对象
    # sentence: 输入的字符串句子
    return [lang.word2index[word] for word in sentence.split(' ')]

def tensorFromSentence(lang, sentence):
    """将句子转换为张量形式，并添加EOS结束标记"""
    indexes = indexesFromSentence(lang, sentence)  # 获取单词索引列表
    indexes.append(EOS_token)  # 添加句子结束标记
    # 转换为PyTorch张量，并调整形状为[1, seq_len]
    return torch.tensor(indexes, dtype=torch.long, device=device).view(1, -1)

def tensorsFromPair(pair):
    """将语言对(输入句子,目标句子)转换为输入和目标张量"""
    input_tensor = tensorFromSentence(input_lang, pair[0])  # 输入语言张量
    target_tensor = tensorFromSentence(output_lang, pair[1])  # 目标语言张量
    return (input_tensor, target_tensor)

def get_dataloader(batch_size):
    """创建数据加载器(DataLoader)用于批量训练"""
    # 1. 准备数据：读取、过滤并构建词汇表
    input_lang, output_lang, pairs = prepareData('eng', 'fra', True)

    n = len(pairs)
    # 初始化填充后的数组(用0填充短句子)
    input_ids = np.zeros((n, MAX_LENGTH), dtype=np.int32)
    target_ids = np.zeros((n, MAX_LENGTH), dtype=np.int32)

    # 2. 将每个句子对转换为固定长度的数字序列
    for idx, (inp, tgt) in enumerate(pairs):
        # 处理输入句子
        inp_ids = indexesFromSentence(input_lang, inp)
        inp_ids.append(EOS_token)
        input_ids[idx, :len(inp_ids)] = inp_ids  # 前面填充实际内容，后面保持0
        
        # 处理目标句子
        tgt_ids = indexesFromSentence(output_lang, tgt)
        tgt_ids.append(EOS_token)
        target_ids[idx, :len(tgt_ids)] = tgt_ids

    # 3. 创建PyTorch数据集和数据加载器
    train_data = TensorDataset(
        torch.LongTensor(input_ids).to(device),  # 输入序列 [n_samples, MAX_LENGTH]
        torch.LongTensor(target_ids).to(device)   # 目标序列 [n_samples, MAX_LENGTH]
    )

    train_sampler = RandomSampler(train_data)  # 随机采样器
    train_dataloader = DataLoader(
        train_data, 
        sampler=train_sampler, 
        batch_size=batch_size  # 按指定批量大小加载数据
    )
    return input_lang, output_lang, train_dataloader

def train_epoch(dataloader, encoder, decoder, encoder_optimizer,
          decoder_optimizer, criterion):

    total_loss = 0
    for data in dataloader:
        input_tensor, target_tensor = data

        encoder_optimizer.zero_grad()
        decoder_optimizer.zero_grad()

        encoder_outputs, encoder_hidden = encoder(input_tensor)
        decoder_outputs, _, _ = decoder(encoder_outputs, encoder_hidden, target_tensor)

        loss = criterion(
            decoder_outputs.view(-1, decoder_outputs.size(-1)),
            target_tensor.view(-1)
        )
        loss.backward()

        encoder_optimizer.step()
        decoder_optimizer.step()

        total_loss += loss.item()

    return total_loss / len(dataloader)

import time
import math

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

def timeSince(since, percent):
    now = time.time()
    s = now - since
    es = s / (percent)
    rs = es - s
    return '%s (- %s)' % (asMinutes(s), asMinutes(rs))

def train(train_dataloader, encoder, decoder, n_epochs, learning_rate=0.001,
               print_every=100, plot_every=100):
    start = time.time()
    plot_losses = []
    print_loss_total = 0  # Reset every print_every
    plot_loss_total = 0  # Reset every plot_every

    encoder_optimizer = optim.Adam(encoder.parameters(), lr=learning_rate)
    decoder_optimizer = optim.Adam(decoder.parameters(), lr=learning_rate)
    criterion = nn.NLLLoss()

    for epoch in range(1, n_epochs + 1):
        loss = train_epoch(train_dataloader, encoder, decoder, encoder_optimizer, decoder_optimizer, criterion)
        print_loss_total += loss
        plot_loss_total += loss

        if epoch % print_every == 0:
            print_loss_avg = print_loss_total / print_every
            print_loss_total = 0
            print('%s (%d %d%%) %.4f' % (timeSince(start, epoch / n_epochs),
                                        epoch, epoch / n_epochs * 100, print_loss_avg))

        if epoch % plot_every == 0:
            plot_loss_avg = plot_loss_total / plot_every
            plot_losses.append(plot_loss_avg)
            plot_loss_total = 0

    showPlot(plot_losses)

import matplotlib.pyplot as plt
plt.switch_backend('agg')
import matplotlib.ticker as ticker
import numpy as np

def showPlot(points):
    plt.figure()
    fig, ax = plt.subplots()
    # this locator puts ticks at regular intervals
    loc = ticker.MultipleLocator(base=0.2)
    ax.yaxis.set_major_locator(loc)
    plt.plot(points)

def evaluate(encoder, decoder, sentence, input_lang, output_lang):
    with torch.no_grad():
        input_tensor = tensorFromSentence(input_lang, sentence)

        encoder_outputs, encoder_hidden = encoder(input_tensor)
        decoder_outputs, decoder_hidden, decoder_attn = decoder(encoder_outputs, encoder_hidden)

        _, topi = decoder_outputs.topk(1)
        decoded_ids = topi.squeeze()

        decoded_words = []
        for idx in decoded_ids:
            if idx.item() == EOS_token:
                decoded_words.append('<EOS>')
                break
            decoded_words.append(output_lang.index2word[idx.item()])
    return decoded_words, decoder_attn
    
def evaluateRandomly(encoder, decoder, n=10):
    for i in range(n):
        pair = random.choice(pairs)
        print('>', pair[0])
        print('=', pair[1])
        output_words, _ = evaluate(encoder, decoder, pair[0], input_lang, output_lang)
        output_sentence = ' '.join(output_words)
        print('<', output_sentence)
        print('')

hidden_size = 128
batch_size = 32

input_lang, output_lang, train_dataloader = get_dataloader(batch_size)

encoder = EncoderRNN(input_lang.n_words, hidden_size).to(device)
decoder = AttnDecoderRNN(hidden_size, output_lang.n_words).to(device)

train(train_dataloader, encoder, decoder, 80, print_every=5, plot_every=5)

encoder.eval()
decoder.eval()
evaluateRandomly(encoder,decoder)

def showAttention(input_sentence, output_words, attentions):
    fig = plt.figure()
    ax = fig.add_subplot(111)
    cax = ax.matshow(attentions.cpu().numpy(), cmap='bone')
    fig.colorbar(cax)

    # 设置x轴
    x_labels = [''] + input_sentence.split(' ') + ['<EOS>']
    ax.set_xticks(range(len(x_labels)))  # 先设置固定数量的刻度
    ax.set_xticklabels(x_labels, rotation=90)
    
    # 设置y轴
    y_labels = [''] + output_words
    ax.set_yticks(range(len(y_labels)))  # 先设置固定数量的刻度
    ax.set_yticklabels(y_labels)

    plt.show()

def evaluateAndShowAttention(input_sentence):
    output_words, attentions = evaluate(encoder, decoder, input_sentence,input_lang, output_lang)
    print('input = ', input_sentence)
    print('output = ',' '.join(output_words))
    showAttention(input_sentence,output_words,attentions[0,:len(output_words),:])

evaluateAndShowAttention('il n est pas aussi grand que son pere')

evaluateAndShowAttention('je suis trop fatigue pour conduire')

evaluateAndShowAttention('je suis desole si c est une question idiote')

evaluateAndShowAttention('je suis reellement fiere de vous')