127. 单词接龙

题目
给定两个单词（beginWord 和 endWord）和一个字典，找到从 beginWord 到 endWord 的最短转换序列的长度。转换需遵循如下规则：

每次转换只能改变一个字母。
转换过程中的中间单词必须是字典中的单词。
说明:

如果不存在这样的转换序列，返回 0。
所有单词具有相同的长度。
所有单词只由小写字母组成。
字典中不存在重复的单词。
你可以假设 beginWord 和 endWord 是非空的，且二者不相同。
示例 1:

输入:
beginWord = "hit",
endWord = "cog",
wordList = ["hot","dot","dog","lot","log","cog"]

输出: 5

解释: 一个最短转换序列是 "hit" -> "hot" -> "dot" -> "dog" -> "cog",
     返回它的长度 5。

示例 2:

输入:
beginWord = "hit"
endWord = "cog"
wordList = ["hot","dot","dog","lot","log"]

输出: 0

解释: endWord "cog" 不在字典中，所以无法进行转换。

【中等】
【分析】BFS
想法是这个想法，但是超时了。。。

class Solution:
    def ladderLength(self, beginWord: str, endWord: str, wordList: List[str]) -> int:
        if endWord not in wordList:
            return 0
        queue=[beginWord]
        seen={beginWord:1}
        while queue:
            word=queue.pop(0)
            for i in range(len(word)):
                index=list(range(i))+list(range(i+1,len(word)))
                w0=[word[_] for _ in index]
                for w in wordList:
                    if w!=word:
                        w1=[w[_] for _ in index]
                        if w1==w0 and w not in seen:
                            if w==endWord:
                                return(seen[word]+1)
                            queue.append(w)
                            seen[w]=seen[word]+1
        return 0

需要改进，改进方向可以是单词预处理部分，怎么来处理单词只改变其中一个字母变成wordList中的某单词。

【分析2】单词预处理，举个栗子：
beginWord = “hit”,
endWord = “cog”,
wordList = [“hot”,“dot”,“dog”,“lot”,“log”,“cog”]

对$hot$处理，有三种方式：$\ast ot,h\ast t,ho\ast$，同理，其他单词都有三种处理方式，让$\ast ot,h\ast t,ho\ast$等为key，然后让处理成这种方式的单词作为value。

这一过程就像是在处理无向无权图，最后化为求最短路径问题。

from collections import defaultdict       
class Solution:
    def ladderLength(self, beginWord: str, endWord: str, wordList: List[str]) -> int:
        if endWord not in wordList or not beginWord or not endWord or not wordList:
            return 0
        
        #预处理
        dic=defaultdict(list)
        L=len(wordList[0])
        for word in wordList:
            for i in range(L):
                dic[word[0:i]+"*"+word[i+1::]].append(word)    
        
        queue=[beginWord]
        seen={beginWord:1}
        while queue:
            word=queue.pop(0)
            for i in range(L):
                next_word=dic[word[0:i]+"*"+word[i+1::]]
                for w in next_word:
                    if w==endWord:
                        return(seen[word]+1)
                    if w not in seen:
                        queue.append(w)
                        seen[w]=seen[word]+1
        return 0

时间复杂度：$O(M\times N)O(M\times N)$，其中 M 是单词的长度 N 是单词表中单词的总数。找到所有的变换需要对每个单词做 M 次操作。同时，最坏情况下广度优先搜索也要访问所有的 N个单词。

空间复杂度：$O(M\times N)O(M\times N)$，要在 dic 字典中记录每个单词的 M 个通用状态。访问数组的大小是 N。广搜队列最坏情况下需要存储 N 个单词。
【注】当字典中的value是一个列表时，可以利用collections中的defaultdict模块做到，之前用dict直接做没有这个方便。

【分析3】双向广度优先搜索，即从beginWord和endWord两边同时开始搜索，搜索结束的条件是找到一个单词被两边都搜索过。所以，需要记录两边分别走的搜索步数，还有两边分别访问过的单词。

下面给出官方参考，我懒得再写一遍了，下次自己再考虑写一写

from collections import defaultdict
class Solution(object):
    def __init__(self):
        self.length = 0
        # Dictionary to hold combination of words that can be formed,
        # from any given word. By changing one letter at a time.
        self.all_combo_dict = defaultdict(list)

    def visitWordNode(self, queue, visited, others_visited):
        current_word, level = queue.pop(0)
        for i in range(self.length):
            # Intermediate words for current word
            intermediate_word = current_word[:i] + "*" + current_word[i+1:]

            # Next states are all the words which share the same intermediate state.
            for word in self.all_combo_dict[intermediate_word]:
                # If the intermediate state/word has already been visited from the
                # other parallel traversal this means we have found the answer.
                if word in o thers_visited:
                    return level + others_visited[word]
                if word not in visited:
                    # Save the level as the value of the dictionary, to save number of hops.
                    visited[word] = level + 1
                    queue.append((word, level + 1))
        return None

    def ladderLength(self, beginWord, endWord, wordList):
        """
        :type beginWord: str
        :type endWord: str
        :type wordList: List[str]
        :rtype: int
        """

        if endWord not in wordList or not endWord or not beginWord or not wordList:
            return 0

        # Since all words are of same length.
        self.length = len(beginWord)

        for word in wordList:
            for i in range(self.length):
                # Key is the generic word
                # Value is a list of words which have the same intermediate generic word.
                self.all_combo_dict[word[:i] + "*" + word[i+1:]].append(word)


        # Queues for birdirectional BFS
        queue_begin = [(beginWord, 1)] # BFS starting from beginWord
        queue_end = [(endWord, 1)] # BFS starting from endWord

        # Visited to make sure we don't repeat processing same word
        visited_begin = {beginWord: 1}
        visited_end = {endWord: 1}
        ans = None

        # We do a birdirectional search starting one pointer from begin
        # word and one pointer from end word. Hopping one by one.
        while queue_begin and queue_end:

            # One hop from begin word
            ans = self.visitWordNode(queue_begin, visited_begin, visited_end)
            if ans:
                return ans
            # One hop from end word
            ans = self.visitWordNode(queue_end, visited_end, visited_begin)
            if ans:
                return ans

        return 0