A figure from Zhihu as the guidance for the classification when entering level of 300-500 questions. Please follow LeetCode tag and sort in frequency. If solved before, review and check the highlights. Otherwise, resolve it.
In details, the daily exercises include
- Review the topic and template of this classification.
- Move on LeetCode list in current tag. Review the solved questions, and try to finish unsolved questions in timely manner.
- 2 in the morning, 4 in the afternoon. After break time, 2 more. Evening 2 with Ming in the manner of mocking interview.
Follow several principles,
- Figure the frame out in 10 minutes or refer to the solution.
- Speak loud when thinking, framing up the problem, asking potential questions, and asking for the hints. (自己跟自己说虽然比较傻...)
- Watch out the time elapsed everytime. Pretent to have a real inverview during the night time.
- If stalled, go and take a look at OOP and BQ instead of schrolling iPhone......
树上的问题, 检查树, 翻转树, 遍历树, 重建树等问题. Refer to topic 1 and 2 in Pattern folder.
- BST 是考察的重点. BST深入理解in-order traversal的应用.
- BST 当中predecessor and successor的概念. node的succesor在node.right.left.left...., node的predecessor在node.left.right.right....
- Lowest common ancestor (LCA) 问题, 可以从p, q起始, 也可以从root出发.
- !!树上问题通常都要切分成子问题, 然后理解子问题form up以及transit的过程. 绝大多数树上问题都要找到能够切分的子问题.
图上连接, 以及拓扑排序的应用问题. Refer to topic 1, 2, and 11 in Pattern folder.
- 涉及到图上的问题, 都需要构建邻接表 adjacent list 或邻接矩阵
- 遍历邻接表的基本算法输入DFS BFS问题，常见需要使用visited存储已访问节点
- Unweighted, undirected 问题一般使用DFS比较好判断链接状态, 例如有多少个独立节点, 是否有cycle, 一定需要visited记录访问过的节点. 如果非backtracking形式, 及返回recursion时未消除visited, 则需要有单独判断是否为parent node的语句.
- Tarjan's algorithm: unweighted, undirected, find the cycle in graph. 专用在1192上. 需要为每个节点设置一个rank. 使用dfs, 当邻接点是已访问过的, 且rank高于自己的点时, 检测到cycle. ranks同样扮演visited角色, 记录已访问过的节点.
- Kahn's algorithm: unweighted, directed, can be cyclic, topological sort to find the linear dependence of all nodes. 用在201 207, 310上, 需要使用in-degree和adjacent list, 邻接表反写graph[des] = [src1, src2...] 从入度为0的node起始, 每次减少其相邻的nodes的入度.
- Bipartition algorithm: 染色问题, 用在unweighted, undirected图上. LC 886, 785. visited记录染色情况, 0未着色, 1 -1代表相反着色. 递归函数记录当前点应着颜色. 最外层遍历所有未着色节点并调用递归. BFS 使用方法类似.
- Kruskal's algorithm: undirected, weighted graph, find the minimum spanning tree (MST). 从图中生成带权重的最小生成树算法. 题1135. 按权重排序, 然后从最小权重的边开始链接, 并检查是否成环. 直到node-1个环安排上. 需要借助一个parent表, 在每次添加连接时, 更新parent表指向当前的根节点. 如果两个城市的根节点一致的情况下, 连接两个城市势必造成环.
https://www.paincker.com/graph-theory
回溯问题现在发现尽管都是dfs模板上手, 但对于边界条件, 变量传递等, 每道题的实现都没法达成模板统一化, 而且问题框架的设定, 说明白一道问题的能力都很有考研.
新的数据结构, 在第一轮时没有遇到过. 使用场景用在 字符串, 搜索树, 字符前缀, 这些关键概念上. 关键题目是 140 Word Break II, 212 Word Search II, 用于匹配字符串.
- Trie is implemented by hash map and represented in tree structure. A trie is a dictionary of a group of words. The searching and inserting time complexity are O(k) where k is the length of the word (or the height of the tree).
- Strings in the same path have same prefix. Trie node may have a KEY_WORD flag. When iterate to a node, who has a key_word, it denotes a complete word from root to this location. Usually store this word at this key.
- 212题揭示了, 我们创建的trie不一定需要有search方法, 可能需要根据遍历的路径来iterate trie pointer. (trie search is mixed with recursive step, a parent node is passed)
- Trie 常见的 O(k) 操作是, insert, search, 还有需要搭配用到DFS的search_prefix, 即找到从prefix起始的所有符合条件的match.