Designing a custom structure means composing existing structures so that each required operation hits its target complexity, letting one structure cover another's weakness. The classic technique is pairing a hash map with an array, heap, or linked list.
Requirements: , , and each in O(1). A hash map alone can't do O(1) random; an array alone can't do O(1) remove. .
insert(x)remove(x)getRandom()import random
class RandomizedSet:
def __init__(self):
self.idx = {} # value -> index in list
self.vals = [] # values for O(1) random pick
def insert(self, x): # O(1)
if x in self.idx: return False
self.idx[x] = len(self.vals)
self.vals.append(x)
return True
def remove(self, x): # O(1): swap with last, pop
if x not in self.idx: return False
i, last = self.idx[x], self.vals[-1]
self.vals[i] = last # move last into the hole
self.idx[last] = i
self.vals.pop()
del self.idx[x]
return True
def getRandom(self): # O(1)
return random.choice(self.vals)
1. List every required operation + its target complexity.
2. For each, name a structure that hits it:
by key O(1) -> hash map
by index O(1) -> array
min/max O(log n) -> heap
ordered O(log n) -> balanced tree / skip list
O(1) splice -> doubly linked list
3. COMBINE them; use the hash map to map keys -> positions/nodes
in the other structure so updates stay O(1).
4. Verify each operation against its target.
| Pattern | Structures combined | Enables |
|---|---|---|
| LRU cache | hash map + doubly linked list | O(1) get/put |
| O(1) randomized set | hash map + array | O(1) insert/remove/random |
| Min in O(1) on a stack | stack + auxiliary min-stack | O(1) getMin |
Real systems rarely need a single textbook structure — they need a composite tuned to a specific access pattern, which is exactly what these designs deliver.
Interviewers use these problems to test whether you can reason from operation costs to a composition, the core skill behind high-performance system design.