Just FYI, this repository isn't just README.md, it's also about a thousand solutions here.
Leetcode imports modules as wildcards, so you don't have to specify module names. There are some exceptions:
- Single
bisect()
without a prefix triggersobject is not callable
, usebisect.bisect()
orbisect_left()
. - You have to specify
re.sub
becausesub
without a prefix isoperator.sub
. - Default
pow
is__builtins__['pow']
(supports up to 3 arguments, including the modulus), notmath.pow
.
For example, Leetcode header has import * from itertools
, so we use comb()
instead of itertools.comb()
:
class Solution:
def uniquePaths(self, m: int, n: int) -> int:
return comb(m+n-2, n-1)
You can also use __import__('module').func
for unlisted modules (namely, numpy
, scipy
, and sortedcontainers
).
class Solution:
def checkStraightLine(self, p):
return __import__('numpy').linalg.matrix_rank([[1]+x for x in p])<3
Sometimes you can save on casting of the return type, e.g. Leetcode autoconverts keys and mixed types to lists.
class Solution:
def topKFrequent(self, nums: List[int], k: int) -> List[int]:
return dict(Counter(nums).most_common(k))
It also automatically evaluates generators:
class Solution:
def countBits(self, n: int) -> List[int]:
return map(int.bit_count,range(n+1))
You can also return linked list of values as ListNode('a,b,...')
. This one is really specific, but sometimes useful.
class Solution:
def addTwoNumbers(self, a: Optional[ListNode], b: Optional[ListNode]) -> Optional[ListNode]:
f=lambda n:n and n.val+10*f(n.next)or 0;return ListNode(','.join([*str(f(a)+f(b))][::-1]))
Leetcode also has serialize
and deserialize
functions for lists and trees:
class Solution:
def reverseList(self, h: Optional[ListNode]) -> Optional[ListNode]:
return h and h.deserialize(str(eval(h.serialize(h))[::-1]))
There is also has_sycle
function:
class Solution:
def hasCycle (self, h: Optional[ListNode]) -> bool:
return ListNode.has_cycle(h)
There are also _*_node_to_array
and _array_to_*_node
functions:
class Solution:
def isPalindrome(self, h: ListNode) -> bool:
return(s:=type(h)._list_node_to_array(h))==s[::-1]
You can also dump the entire preprocessed solution file to check all the imports for yourself (see gist):
with open(__file__, 'rt') as f:
print(f.read())
The solution driver code writes all results to the user.out
file, so we can use it like this:
class Solution:
def twoSum(self, nums: List[int], target: int) -> List[int]:
from zlib import decompress
from base64 import b64decode
open('user.out', 'wb').write(decompress(b64decode('eJzdkMEVwCAIQ++dggFyEKi2zuLr/mtItZb63KAc\
kpfwuVAYFK6tCIjNPH1KncodJMuBTqWTYUGe89hNX1Kd/K2Nh1iM3mYbkMlpIaFrvvcCaVwCH+YB3FSHVu5xXDc='))),exit(0)
There is no approved method to get all the test cases for problems in LeetCode. You can, however, leverage the fact that LeetCode reveals the test case that causes your code to fail. The solution above is not very reliable, because tests and environment may change, but it's pretty fast.
You can explore the sandbox using shell commands, e.g. (see gist):
import subprocess
print(subprocess.run(["ls", "-la", "/"]))
Some leetcode problems may be solved at the function declaration level.
class Solution:searchInsert=bisect_left
class Solution:permute=permutations
class Solution:sortArray=sorted
class Solution:bulbSwitch=isqrt
class Solution:search=contains
class Solution:myPow=pow
Note that it only works for the built-in functions, they can omit self
parameter.
It's a built-in CPython feature:
You cannot use your own function like that, without skipping the first argument.
class Solution:reverseWords=lambda _,s:' '.join(w[::-1]for w in s.split())
It's not necessarily shorter, because lambdas can't use semicolons.
In some cases you don't even have to write "class Solution:", e.g:
Codec=TreeNode
Let's consider function declaration is zero lines.
class Solution:
def accountBalanceAfterPurchase(self, x: int) -> int:
return(104-x)//10*10
class Solution:
def majorityElement(self, n: List[int]) -> int:
return mode(n)
class Solution:
def numberOfMatches(self, n: int) -> int:
return~-n
class Solution:
def stoneGame(self, piles: List[int]) -> bool:
return 1
You can also write:
class Solution:stoneGame=truth
class Solution:
def isStrictlyPalindromic(self, n: int) -> bool:
0
Notice no return operator here, can be anything (e.g. pass
), as the function returns None. You can also write:
class Solution:isStrictlyPalindromic=not_
Fictitious (anonymous) lambdas may be nested. E.g. you can use lambdas as parameters:
(lambda a,b,c: code)(a,b,c)
becomes(lambda a,b,c: code)(lambda a: code, lamda b: code, lambda c: code)
You can't unpack lambda tuples in Python 3 since PEP 3113, however, if your lambda is flat, there is an upgrade path:
lambda (x, y): x + y
in Python 2 becomeslambda xy:(lambda x,y: x+y)(*xy)
in Python 3.
You can also unpack multiple tuples as lambda xy,ab:(lambda x,y,a,b: x+y+a+b)(*(xy+ab))
.
class Solution:
def countVowelPermutation(self, n: int) -> int:
return sum(reduce(lambda x,_:(lambda a,e,i,o,u:(e+i+u,a+i,e+o,i,i+o))(*x),[0]*(n-1),[1]*5))
%(10**9+7)
Generator expressions (x for y in z)
are memory efficient since they only require memory for
the one value they yield. If you don't care about memory you can use square brackets to make it a list comprehension that automatically runs the loop.
You can also exhaust a generator using all()
, any()
or sum()
, depending on the return values.
You can also save a few chars using [*g]
syntax instead of list(g)
where g is a generator function.
Generator length len(list(g))
can be calculated in constant memory as sum(1 for _ in g)
.
Generators provide an easy, built-in way to create instances of Iterators.
Iterators are objects that have an __iter__
and a __next__
method.
The iter()
method returns an iterator for the given argument.
Each access iterator advances one step.
May be useful, e.g. this solution would not work without converting a string to an iterator:
class Solution:
def appendCharacters(self, s: str, t: str) -> int:
s=iter(s);return sum(c not in s for c in t)
You can also use iter()
to split a list into chunks. The [iter(s)]*n
trick breaks a list into pieces of size n:
class Solution:
def minChanges(self, s: str) -> int:
return sum(map(ne,s[::2],s[1::2]))
class Solution:
def minChanges(self, s: str) -> int:
return sum(map(ne,s:=iter(s),s))
class Solution:
def minChanges(self, s: str) -> int:
return sum(map(ne,*[iter(s)]*2))
Counters (collections.Counter()
) can be updated, similar to dict.update()
, it's much faster than a sum of counters.
E.g. c[i]+=1
is equivalent to c.update([i])
, c[i]-=1
is c.update({i:-1})
.
To delete a key you can use the .pop
method (same as del
), it's shorter than popitem()
.
You can easily remove zero and negative values from a counter (it's the official way, see documentation):
c = Counter({1:1,2:0,3:-1}); print(c:=+c) #{1: 1}, same as c += Counter()
Since python 3.7, as a dict subclass, Counter inherited the capability to remember insertion order.
class Solution:
def reductionOperations(self, n: List[int]) -> int:
return sum(i*v for i,(_,v)in enumerate(sorted(Counter(n).items())))
class Solution:
def reductionOperations(self, n: List[int]) -> int:
return sum(i*v for i,v in enumerate(Counter(sorted(n)).values()))
Since Python 3.10 you can use total()
to compute sum of the counts.
class Solution:
def minSteps(self, s: str, t: str) -> int:
return sum((Counter(s)-Counter(t)).values())
class Solution:
def minSteps(self, s: str, t: str) -> int:
return(Counter(s)-Counter(t)).total()
The controversial walrus operator (:=
) added in Python 3.8 (PEP-572
that resulted in Guido's resign),
can be used to define or update a variable or a function (mostly used for recursive functions).
You can define and call a recursive function in a single line with Y-combinator, e.g.:
return (lambda y,x:y(y,x))(lambda f,x:1 if x==0 else x*f(f,x-1),5)
But the walrus operator syntax is much more concise:
return (f:=lambda x:1 if x==0 else x*f(x-1))(5)
Many oneliners would be impossible to do without it (or rather, very hard, with nested lambdas).
Sometimes you don't even need extra brackets, e.g. in map(f:=x,y)
or next(g,f:=x)
so it may be shorter than operators separated by semicolons.
class Solution:
def numOfMinutes(self, n: int, h: int, m: List[int], t: List[int]) -> int:
return max(map(f:=cache(lambda i:~i and t[i]+f(m[i])),m))
class Solution(object):
def guessNumber(self, n: int) -> int:
l,r = 1, n
while l <= r:
m = (l + r) // 2
res = guess(m)
if res == 0:
return m
elif res > 0:
l = m + 1
else:
r = m - 1
return 0
class Solution:
def guessNumber(self, n: int) -> int:
return (f:=lambda l,h:h if l+1==h else f(m,h) if guess(m:=(l+h)//2)>0 else f(l,m))(0,n)
class Solution:
def reverse(self, x: int) -> int:
r, x = 0, abs(x)
while x:
r = r*10 + x%10
x //= 10
return ((x>0)-(x<0))*min(2**31, r)
class Solution:
def reverse(self, x: int) -> int:
return ((x>0)-(x<0))*min(2**31,(f:=lambda r,x:f(r*10 + x%10, x//10) if x else r)(0,abs(x)))
class Solution:
def topKFrequent(self, words: List[str], k: int) -> List[str]:
return nsmallest(k,(f:=Counter(words)).keys(),lambda x:(-f[x],x))
You can't use walrus operator for structures, however, you can use __setattr__
for dictionaries or __setitem__
for lists if you need an assignment
(functions return None
). To set a key for the list or for the dictionary, you can also use setattr
or setitem
functions from the operator
module,
e.g. c[x]=1
is the same as setitem(c,x,1)
.
class Solution:
def addOneRow(self, root: TreeNode, v: int, d: int, isLeft: bool = True) -> TreeNode:
if d == 1:
return TreeNode(v, root if isLeft else None, root if not isLeft else None)
if not root:
return None
root.left = self.addOneRow(root.left, v, d - 1, True)
root.right = self.addOneRow(root.right, v, d - 1, False)
return root
class Solution:
def addOneRow(self, root: TreeNode, v: int, d: int, isLeft: bool = True) -> TreeNode:
return TreeNode(v, root if isLeft else None, root if not isLeft else None) if d==1 else \
setattr(root,'left', self.addOneRow(root.left, v, d - 1, True)) or \
setattr(root,'right', self.addOneRow(root.right, v, d - 1, False)) or root if root else None
class Solution(object):
def deleteMiddle(self, head):
def f(a, b):
if not b:
return a.next
a.next = f(a.next, b.next.next) if b.next else f(a.next, b.next)
return a
return f(head, head.next)
class Solution(object):
def deleteMiddle(self, head):
return (f:=lambda a,b:setattr(a,'next',f(a.next, b.next.next) if b.next
else f(a.next, b.next)) or a if b else a.next)(head, head.next)
Note that setitem
also supports slices:
# TLE, too slow
class Solution:
def countPrimes(self, n):
g=range(2,n);return len(reduce(lambda r,x:r-set(range(x**2,n,x))if x in r else r,g,set(g)))
class Solution:
def countPrimes(self, n):
a = [0,0]+[1]*(n-2)
for i in range(2,int(n**0.5)+1):
if a[i]:
a[i*i:n:i] = [0]*len(a[i*i:n:i])
return sum(a)
class Solution:
def countPrimes(self, n):
return sum(reduce(lambda a,i:a[i] and setitem(a,slice(i*i,n,i),[0]*len(a[i*i:n:i])) or a,
range(2,int(n**0.5)+1), [0,0]+[1]*(n-2)))
You can also calculate primes like this:
class Solution:
def primeSubOperation(self, a: List[int]) -> bool:
m,p=1,[0]+[i for i in range(2,999)if all(i%j for j in range(2,i))];\
return all(m<(m:=x-p[bisect_right(p,x-m)-1]+1)for x in a)
Note slices can extend the list implicitly, e.g.:
a = [0,1,2]
a[3:4] = [3] # the result is [0,1,2,3]
Be careful though, slicing doesn't extend list beyond the slice size:
a = [0,1]
a[3:4] = [3,4] # the result is [0,1,3,4], NOT [0,1,?,3,4] (!)
Examples:
class Solution:
def longestObstacleCourseAtEachPosition(self, o: List[int]) -> List[int]:
d = []
for e in o:
i = bisect_right(d,e)
if i==len(d):
d.append(0)
d[i] = e
yield i+1
class Solution:
def longestObstacleCourseAtEachPosition(self, o: List[int]) -> List[int]:
d = []
for e in o:
i = bisect_right(d,e)
d[i:i+1] = [e]
yield i+1
class Solution:
def longestObstacleCourseAtEachPosition(self, o: List[int]) -> List[int]:
d=[];return[setitem(d,slice(i:=bisect_right(d,e),i+1),[e])or i+1for e in o]
Sometimes exec
is shorter than setitem
.
ParkingSystem=type('',(),{'__init__':lambda s,a,b,c:setattr(s,'p',[0,a,b,c]),'addCar':lambda s,t:\
setitem(s.p,t,s.p[t]-1)or s.p[t]>=0})
ParkingSystem=type('',(),{'__init__':lambda s,a,b,c:setattr(s,'p',[0,a,b,c]),'addCar':lambda s,t:\
exec('s.p[t]-=1')or s.p[t]>=0})
You can write a class or a subclass implementation in one line.
MyHashSet=type('',(set,),{'remove':set.discard,'contains':set.__contains__})
MyStack=type('',(list,),{'push':list.append,'top':lambda s:s[-1],'empty':lambda s:not s})
Counter subclassing fails since Python 3.7, type() doesn't support MRO entry resolution; use types.new_class()
.
When you try to use types.new_class()
it says TypeError: .__init_subclass__() takes no keyword arguments')
.
This can be avoided by creating the class first, then adding the methods to it separately, e.g.:
с=Counter;с.insert=lambda s,x:s.update({x})or s[x]<2;с.remove=lambda s,x:s.pop(x,0);
с.getRandom=lambda s:choice([*s]);RandomizedSet=с
Sometimes (not always) you can skip __init__
and use static attributes.
UndergroundSystem=type('',(),{'h':{},'m':{},'checkIn':lambda s,i,v,t:setitem(s.m,i,(v,t)),
'checkOut':lambda s,i,d,w:(v:=s.m[i][0])and setitem(s.h,(v,d),[*map(sum,zip(s.h.pop((v,d),
(0,0)),(w-s.m[i][1],1)))]),'getAverageTime':lambda s,v,d:truediv(*s.h[v,d])})
Binary search can be replaced by the built-in bisect
methods.
Custom binary search can use either an item getter object or a key function (since Python 3.10).
class Solution:
def guessNumber(self, n: int) -> int:
l,r = 1, n
while l <= r:
m = (l + r) // 2
res = guess(m)
if res == 0:
return m
elif res > 0:
l = m + 1
else:
r = m - 1
return 0
class Solution:
def guessNumber(self, n: int) -> int:
return bisect_left(type('',(),{'__getitem__':lambda _,i: -guess(i)})(), 0, 1, n)
class Solution:
def guessNumber(self, n: int) -> int:
return bisect_left(range(n), 0, key=lambda num: -guess(num))
Note that built-in methods don't support negative left margin, so you have to subtract it from the result:
class Solution:
def kthSmallestProduct(self, a: List[int], b: List[int], k: int) -> int:
f=lambda x:sum(bisect_right(b,x//y)if y>0 else len(b)-bisect_left(b,ceil(x/y))if y<0 else
(x>=0)*len(b)for y in a)
l,r = -10**10-1, 10**10+1
while l < r:
m = (l + r)//2
if f(m) >= k:
r = m
else:
l = m + 1
return l
class Solution:
def kthSmallestProduct(self, a: List[int], b: List[int], k: int) -> int:
f=lambda x:sum(bisect_right(b,x//y)if y>0 else len(b)-bisect_left(b,ceil(x/y))if y<0 else
(x>=0)*len(b)for y in a)
return bisect_left(range(2*(r:=10**10)),k,key=lambda i:f(i-r))-r
Bisect implementation for reference:
While loops are not very oneliner-friendly. You can use count()
generator with next()
.
Note that next
default parameter gets initialized first so you can use it for the startup code (but can't use to calculate result).
class Solution:
def twoSum(self, nums: List[int], target: int) -> List[int]:
seen = {}
for i,x in enumerate(nums):
if target-x in seen:
return seen[target-x], i
seen[x] = i
return False
class Solution:
def twoSum(self, n: List[int], t: int) -> List[int]:
return next(((m[t-x],i)for i,x in enumerate(n)if t-x in m or setitem(m,x,i)),m:={})
class Solution:
def breakPalindrome(self, s: str) -> str:
for i in range(len(s) // 2):
if s[i] != 'a':
return s[:i] + 'a' + s[i + 1:]
return s[:-1] + 'b' if s[:-1] else ''
class Solution:
def breakPalindrome(self, s: str) -> str:
return next((s[:i]+'a'+s[i+1:]for i in range(len(s)//2)if s[i]!='a'),s[:-1]and s[:-1]+'b')
class Solution:
def isPossible(self, target: List[int]) -> bool:
s = sum(target)
q = [-a for a in target]
heapify(q)
while True:
x = -heappop(q)
if x==1:
return True
if s==x:
return False
d = 1 + (x-1) % (s-x)
if x==d:
return False
s = s - x + d
heappush(q, -d)
class Solution:
def isPossible(self, target: List[int]) -> bool:
return (s:=sum(target),q:=[-a for a in target],heapify(q)) and next((x==1 for _ in count()
if (x:=-heappop(q))==1 or s==x or (d:=1+(x-1)%(s-x))==x or not (s:=s-x+d,heappush(q,-d))),1)
You can also use takewhile()
, it's also a generator, so you need to expand it (e.g. with repeat(0)
).
class Solution:
def maxSlidingWindow(self, nums: List[int], k: int) -> List[int]:
r, d = [], deque()
for i, n in enumerate(nums):
while d and n>=nums[d[-1]]:
d.pop()
d.append(i)
if d[0] == i-k:
d.popleft()
r.append(nums[d[0]])
return r[k-1:]
class Solution:
def maxSlidingWindow(self, nums: List[int], k: int) -> List[int]:
return (d:=deque()) or reduce(lambda r,p:(
any(takewhile(lambda _:d and p[1]>=nums[d[-1]] and d.pop(), repeat(0))),
d.append(p[0]), d[0]==p[0]-k and d.popleft(), r.append(nums[d[0]])) and r,
enumerate(nums), [])[k-1:]
You could also try any()
or all()
as a while loop instead of next()
, it may be shorter.
You can assure that expression never returns None
, using []
([None]
evaluates to True
).
class Solution:
def lastStoneWeight(self, stones: List[int]) -> int:
stones.sort()
while len(stones) > 1:
insort(stones,stones.pop() - stones.pop())
return stones[0]
class Solution:
def lastStoneWeight(self, s: List[int]) -> int:
return next((s[0] for _ in count() if not s[1:] or insort(s,s.pop()-s.pop())),s.sort())
class Solution:
def lastStoneWeight(self, s: List[int]) -> int:
return (s.sort(),all(s[1:] and [insort(s,s.pop()-s.pop())] for _ in count()),s[0])[2]
You can also evalulate multiline code with exec
. Unlike eval
, is not limited to a single string.
class Solution:
def minimumOneBitOperations(self, n: int) -> int:
return next((r for _ in count()if not(n and(r:=r^n,n:=n//2))),r:=0)
class Solution:
def minimumOneBitOperations(self, n: int) -> int:
r=[0];exec('while n:\n r[0]^=n\n n//=2');return r[0]
class Solution:
def minimumOneBitOperations(self, n: int) -> int:
return(f:=lambda n:n and n^f(n//2))(n)
To swap values you can use either exec
(inline version of a,b=b,a
) or a temporary variable (t:=a,a:=b,b:=t
).
Note that eval
accepts only a single expression, and returns the value of the given expression,
whereas exec
ignores the return value from its code, and always returns None
, its use has no effect
on the compiled bytecode of the function where it is used. It does however affect existing variables.
Example:
class Solution:
def sortColors(self, nums: List[int]) -> None:
def fn(t,b):
red, white, blue = t
return (swap:=lambda a,x,y:exec('a[x],a[y]=a[y],a[x]'),(swap(nums,red,white),
(red+1,white+1,blue))[1] if nums[white]==0 else ((red,white+1,blue) if nums[white]==1
else (swap(nums,white,blue),(red,white,blue-1))[1]))[1]
reduce(fn, nums, [0,0,len(nums)-1])
class Solution:
def sortColors(self, nums: List[int]) -> None:
(s:=lambda a,x,y:(t:=a[x],setitem(a,x,a[y]),setitem(a,y,t),a)[3],
f:=lambda a,i,j,k:(f(s(a,i,j),i+1,j+1,k) if a[j]==0 else f(a,i,j+1,k) if a[j]==1
else f(s(a,j,k),i,j,k-1)) if i<=j<=k else None)[1](nums,0,0,len(nums)-1)
Also you can try a swap function here (but it's pretty long, I don't use it):
swap = lambda a,x,y:(lambda f=a.__setitem__:(f(x,(a[x],a[y])),f(y,a[x][0]),f(x,a[x][1])))()
You can use map
for a lot of things, for example to traverse through adjacent cells.
class Solution:
def maxAreaOfIsland(self, grid: List[List[int]]) -> int:
def dfs(i,j):
if 0<=i<len(grid) and 0<=j<len(grid[0]) and grid[i][j]:
grid[i][j] = 0
return 1 + sum(map(dfs,(i+1,i,i-1,i),(j,j+1,j,j-1)))
return 0
return max(dfs(i,j) for i in range(len(grid)) for j in range(len(grid[0])))
class Solution:
def maxAreaOfIsland(self, g: List[List[int]]) -> int:
return max((f:=lambda i,j:setitem(g[i],j,0) or 1 + sum(map(f,(i+1,i,i-1,i),(j,j+1,j,j-1)))
if 0<=i<len(g) and 0<=j<len(g[0]) and g[i][j] else 0)(i,j)
for i in range(len(g)) for j in range(len(g[0])))
Though it's shorter to use complex numbers for 2d maps (introduced by Stephan Pochmann):
class Solution:
def maxAreaOfIsland(self, grid):
grid = {i + j*1j: val for i, row in enumerate(grid) for j, val in enumerate(row)}
def area(z):
return grid.pop(z, 0) and 1 + sum(area(z + 1j**k) for k in range(4))
return max(map(area, set(grid)))
class Solution:
def maxAreaOfIsland(self, grid):
return max(map(a:=lambda z: g.pop(z, 0) and 1 + sum(a(z + 1j**k) for k in range(4)),
set(g:= {i + j*1j: val for i, row in enumerate(grid) for j, val in enumerate(row)})))
Complex numbers in general are very useful as 2d coordinates:
class Solution:
def isPathCrossing(self, p: str) -> bool:
z=0;return len(p)>=len({0,*{z:=z+1j**'NESW'.find(c)for c in p}})
You can convert lists or tuples to True
with !=0
instead of bool()
(3 chars shorter).
class Solution:
def numIslands(self, grid: List[List[str]]) -> int:
grid = {i + j*1j:int(val) for i,row in enumerate(grid) for j,val in enumerate(row)}
def f(z):
return grid.pop(z,0) and bool([f(z + 1j**k) for k in range(4)])
return sum(map(f, set(grid)))
class Solution:
def numIslands(self, grid: List[List[str]]) -> int:
return sum(map(f:=lambda z:g.pop(z,0) and [f(z + 1j**k) for k in range(4)]!=0,
set(g:={i + j*1j:int(x) for i,row in enumerate(grid) for j,x in enumerate(row)})))
class Solution:
def closedIsland(self, grid: List[List[str]]) -> int:
g = {i+j*1j:1-x for i,r in enumerate(grid) for j,x in enumerate(r)}
f = lambda z:g.pop(z,0) and [f(z+1j**k) for k in range(4)]!=0
sum(f(z) for z in set(g) if not(0<z.real<len(grid)-1 and 0<z.imag<len(grid[0])-1))
return sum(map(f,set(g)))
class Solution:
def closedIsland(self, grid: List[List[str]]) -> int:
return (g:={i+j*1j:1-x for i,r in enumerate(grid) for j,x in enumerate(r)},
f:=lambda z:g.pop(z,0) and [f(z+1j**k) for k in range(4)]!=0,[f(z) for z in set(g)
if not(0<z.real<len(grid)-1 and 0<z.imag<len(grid[0])-1)]) and sum(map(f,set(g)))
class Solution:
def uniquePathsIII(self, grid: List[List[int]]) -> int:
def f(z,r):
if x:=g.pop(z,0):
if x==3 and not g:
r = r + 1
for k in range(4):
r = f(z + 1j**k, r)
g.update({z:x})
return r
g = {i + j*1j:x+1 for i, row in enumerate(grid) for j,x in enumerate(row) if x!=-1}
return f(next(z for z,x in g.items() if x==2),0)
class Solution:
def uniquePathsIII(self, grid: List[List[int]]) -> int:
return (g:={i + j*1j:x+1 for i, row in enumerate(grid) for j,x in enumerate(row)
if x!=-1}) and (f:=lambda z,r:[(x:=g.pop(z,0)) and (x==3 and not g and (r:=r+1),
[r:=f(z + 1j**k,r) for k in range(4)],g.update({z:x}))] and r)
(next(z for z,x in g.items() if x==2), 0)
Unicode find (NOT Union Find) is the greatest trick of all time to solve graph problems. The idea is to use string replace in a Unicode space. Introduced by Stephan Pochmann.
class Solution:
def findRedundantConnection(self, edges: List[List[int]]) -> List[int]:
t = ''.join(map(chr, range(1001)))
for u,v in edges:
if t[u]==t[v]:
return [u,v]
t = t.replace(t[u],t[v])
class Solution:
def findRedundantConnection(self, e: List[List[int]]) -> List[int]:
t=''.join(map(chr,range(1001)));
return next((u,v)for u,v in e if t[u]==t[v]or not(t:=t.replace(t[u],t[v])))
Another example:
class Solution:
def swimInWater(self, g: List[List[int]]) -> int:
n = len(g)
t,r = ''.join(map(chr,range(n*n))),range(n)
for w,i,j in sorted((g[i][j],i,j)for i,j in product(r,r)):
for x,y in ((i+1,j),(i-1,j),(i,j+1),(i,j-1)):
if n>y>=0<=x<n and g[x][y]<=w:
t = t.replace(t[i*n+j],t[x*n+y])
if t[0]==t[-1]:
return w
return 0
class Solution:
def swimInWater(self, g: List[List[int]]) -> int:
n=len(g);t,r=''.join(map(chr,range(n*n))),range(n);return next((w for w,i,j in
sorted((g[i][j],i,j)for i,j in product(r,r))if[t:=t.replace(t[i*n+j],t[x*n+y]) for x,y
in((i+1,j),(i-1,j),(i,j+1),(i,j-1)) if n>y>=0<=x<n and g[x][y]<=w]and t[0]==t[-1]),0)
Another example (Q4 at https://leetcode.com/contest/weekly-contest-392):
class Solution:
def minimumCost(self, n: int, edges: List[List[int]], query: List[List[int]]) -> List[int]:
t,c = ''.join(map(chr,range(n))),{}
for u,v,w in edges:
t = t.replace(t[u],t[v])
for u,v,w in edges:
c[t[u]] = c.get(t[u],w)&w
return [0 if u==v else c[t[u]] if t[u]==t[v] else -1 for u,v in query]
class Solution:
def minimumCost(self, n: int, e: List[List[int]], q: List[List[int]]) -> List[int]:
t,c=''.join(map(chr,range(n))),{};all(t:=t.replace(t[u],t[v])for u,v,_ in e);
[setitem(c,t[u],c.get(t[u],w)&w)for u,v,w in e];
return[u!=v and t[u]!=t[v]and-1or c[t[u]]for u,v in q]
Cache decorator, @lru_cache
or @cache
(since Python 3.9) may be used as an inline function cache(lambda ...)
.
class Solution:
def maxProfit(self, k: int, prices: List[int]) -> int:
@cache
def dfs(i, k, sell):
return 0 if k==0 or i==len(prices) \
else max(dfs(i+1, k-1, 0) + prices[i], dfs(i+1, k, 1)) if sell \
else max(dfs(i+1, k, 1)-prices[i], dfs(i+1, k, sell))
return dfs(0, k, 0)
class Solution:
def maxProfit(self, k: int, prices: List[int]) -> int:
return (f:=cache(lambda i,k,s:0 if k==0 or i==len(prices)
else max(f(i+1,k-s,1-s)+prices[i]*(2*s-1),f(i+1,k,s))))(0,k,0)
class Solution:
def coinChange(self, coins: List[int], amount: int) -> int:
@cache
def f(n):
return min([1 + f(n-c) for c in coins]) if n>0 else 0 if n==0 else inf
x = f(amount)
return x if x!=inf else -1
class Solution:
def coinChange(self, coins: List[int], amount: int) -> int:
return (lambda x:x if x!=inf else -1)((f:=cache(lambda n:
min([1+f(n-c) for c in coins]) if n>0 else 0 if n==0 else inf))(amount))
It is sometimes necessary to reset cache with cache_clear
between tests to avoid Memory Limit Exceeded error.
class Solution:
def lengthOfLongestSubsequence(self, a: List[int], t: int) -> int:
return(a.sort(),r:=(f:=cache(lambda i,b:b and -inf if b<0 or i<0 else
max(1+f(i-1,b-a[i]),f(i-1,b))))(len(a)-1,t),f.cache_clear())and(-1,r)[r>0]
You can also specify maxsize option as f=lru_cache(maxsize)(lambda ...)
in case of memory issues:
class Solution:
def shortestCommonSupersequence(self, a: str, b: str) -> str:
return(f:=lru_cache(9**5)(lambda i,j:a[i:]and b[j:]and(a[i]==b[j]and a[i]+f(i+1,j+1)
or min(a[i]+f(i+1,j),b[j]+f(i,j+1),key=len))or a[i:]or b[j:]))(0,0)
Use it to flatten a loop.
class Solution:
def lengthOfLongestSubstring(self, s):
start, res, h = 0, 0, {}
for i, c in enumerate(s):
start = max(start, h.get(c,0))
res = max(res, i - start + 1)
h[c] = i + 1
return res
class Solution:
def lengthOfLongestSubstring(self, s):
def fn(a,b):
start, res, h = a
i, c = b
start = max(start, h.get(c,0))
res = max(res, i - start + 1)
h[c] = i + 1
return start,res,h
return reduce(fn,enumerate(s),[0,0,{}])[1]
class Solution:
def lengthOfLongestSubstring(self, s):
return reduce(lambda a,b:(s:=max(a[0],a[2].get(b[1],0)),max(a[1],b[0]-s+1),
{**a[2],b[1]:b[0]+1}),enumerate(s),(0,0,{}))[1]
class Solution:
def lengthOfLongestSubstring(self, s):
return reduce(lambda a,b:(lambda t,r,h,i,c:(s:=max(t,h.get(c,0)),max(r,i-s+1),
{**h,c:i+1}))(*a,*b),enumerate(s),(0,0,{}))[1]
Another example:
class Solution:
def longestValidParentheses(self, s: str) -> int:
def fn(a,b):
r, s = a
i, p = b
return (max(r,i-s[-2][0]), s[:-1]) if p==')' and s[-1][1]=='(' else (r, s+[(i,p)])
return reduce(fn, enumerate(s), (0,[(-1, ')')]))[0]
class Solution:
def longestValidParentheses(self, s: str) -> int:
return reduce(lambda a,b:(max(a[0],b[0]-a[1][-2][0]),a[1][:-1]) if b[1]==')'
and a[1][-1][1]=='(' else (a[0],a[1]+[b]),enumerate(s),(0,[(-1,')')]))[0]
The product function (and all of the itertools) is sometimes handy.
class Solution:
def nthUglyNumber(self, n: int) -> int:
return sorted(2**a*3**b*5**c for a in range(32)for b in range(20)for c in range(14))[n-1]
class Solution:
def nthUglyNumber(self, n: int) -> int:
return sorted(2**a*3**b*5**c for a,b,c in product(*map(range,(32,20,14))))[n-1]
class Solution:
def nthUglyNumber(self, n: int) -> int:
return sorted(2**a*3**b*5**c for a,b,c in product(*[range(32)]*3))[n-1]
Nobody will stop you from using semicolons, but you'd still have to convert while and for loops.
Example:
class Solution:
def swapNodes(self, h: Optional[ListNode], k: int) -> Optional[ListNode]:
q = h
i = 1
d = {}
while q:
d[i] = q
q = q.next
i += 1
d[k].val, d[i-k].val = d[i-k].val, d[k].val
return h
class Solution:
def swapNodes(self, h: Optional[ListNode], k: int) -> Optional[ListNode]:
q=h;i=1;d={};all(q and(setitem(d,i,q),q:=q.next,i:=i+1) for _
in count());d[k].val,d[i-k].val=d[i-k].val,d[k].val;return h
class Solution:
def swapNodes(self, h: Optional[ListNode], k: int) -> Optional[ListNode]:
l=[h]+[h:=h.next for _ in[1]*10**5if h];a,b=l[k-1],l[~k];a.val,b.val=b.val,a.val;return l[0]
Many leetcode problems use Fibonacci sequence that can be calculated using a variety of different methods:
- https://r-knott.surrey.ac.uk/Fibonacci/fibFormula.html
- https://en.wikipedia.org/wiki/Generating_function
- https://mathworld.wolfram.com/TribonacciNumber.html
Examples:
class Solution:
def fib(self, n: int) -> int:
a,b = 0,1
for _ in range(n):
a,b = b,a+b
return a
class Solution:
def fib(self, n: int) -> int:
r=5**.5;return round(((1+r)/2)**n/r)
class Solution:
def fib(self, n: int) -> int:
return pow(x:=2<<n,n+1,x*x+~x)%x
class Solution:
def climbStairs(self, n: int) -> int:
a=b=1
for _ in range(n):
a,b = b,a+b
return a
class Solution:
def climbStairs(self, n):
return pow(x:=2<<n,n+2,x*x+~x)%x
class Solution:
def tribonacci(self, n):
a,b,c = 1,0,0
for _ in range(n):
a,b,c = b,c,a+b+c
return c
class Solution:
def tribonacci(self, n: int) -> int:
return round((599510/325947)**n*39065/116186)
class Solution:
def tribonacci(self, n: int) -> int:
return pow(x:=2<<n,n+2,~-x*x*x+~x)%x
Many problems can be solved with a single regex:
class Solution:
def sortVowels(self, s: str) -> str:
return re.sub(t:='(?i)[aeiou]',lambda m,v=sorted(findall(t,s)):heappop(v),s)
class Solution:
def isValid(self, w: str) -> bool:
return match('^(?=.*[aeiou])(?=.*[^0-9aeiou])[a-z0-9]{3,}$',w,I)
class Solution:
def makeGood(self, s: str) -> str:
[s:=re.sub(r'(.)(?!\1)(?i:\1)','',s)for _ in s];return s
class Solution:
def isCircularSentence(self, s: str) -> bool:
return not re.search('(.) (?!\\1)',s+' '+s)
Note that key=itemgetter(n)
is the same length as key=lambda x:x[n]
but a little bit clearer to read.
The performance of itemgetter is also better (up to 2x, because of the creation of the lambda).
Sometimes you can skip key=itemgetter(0)
in comparison operations by converting an argument
to a tuple (15 characters shorter).
class Solution:
def jobScheduling(self, s: List[int], e: List[int], p: List[int]) -> int:
a=sorted(zip(s,e,p));return(f:=cache(lambda i:i-len(a)and max(f(
bisect_left(a,a[i][1],key=itemgetter(0)))+a[i][2],f(i+1))))(0)
class Solution:
def jobScheduling(self, s: List[int], e: List[int], p: List[int]) -> int:
a=sorted(zip(s,e,p));return(f:=cache(lambda i:i-len(a)and max(f(
bisect_left(a,(a[i][1],)))+a[i][2],f(i+1))))(0)
You could also use map(list.pop, v)
instead of [x[-1] for x in v]
to collect the last elements of the list.
class Solution:
def findDiagonalOrder(self, n: List[List[int]]) -> List[int]:
return map(list.pop,sorted([i+j,j,t]for i,r in enumerate(n)for j,t in enumerate(r)))
Using zip
to get elements from the list of tuples is usually shorter, but not always:
class Solution:
def findSmallestSetOfVertices(self, n: int, edges: List[List[int]]) -> List[int]:
return {*range(n)}-{*[*zip(*edges)][1]}
class Solution:
def findSmallestSetOfVertices(self, n: int, edges: List[List[int]]) -> List[int]:
return {*range(n)}-{j for _,j in edges}
You can can use a!=b!=c
in a single boolean condition, similar to 0<=i<n
and m>j>=0<=i<n
.
class Solution:
def expressiveWords(self, s: str, words: List[str]) -> int:
def f(v,w,j=0):
for i in range(len(v)):
if j<len(w) and v[i]==w[j]:
j += 1
elif v[i-1:i+2] != v[i]*3 != v[i-2:i+1]:
return False
return j==len(w)
return sum(f(s,w) for w in words)
class Solution:
def expressiveWords(self, s: str, words: List[str]) -> int:
return sum((f:=lambda v,w,j=0:next((0 for i in range(len(v)) if not(j<len(w) and v[i]==w[j]
and (j:=j+1))and v[i-1:i+2]!=v[i]*3!=v[i-2:i+1]),1) and j==len(w))(s,w) for w in words)
Python handles multi-argument comparisons in the same order as an and
operator, so you can use a shorter form:
class Solution:
def isPowerOfFour(self, n: int) -> bool:
return n>0 and log(n,4)%1==0
class Solution:
def isPowerOfFour(self, n: int) -> bool:
return n>0==log(n,4)%1
~
reverts every bit. Therefore, ~x
means -x-1
. You can use it as reversed index, i.e. for i=0
, a[~i]
means a[-1]
, etc. or just replace -x-1
with ~x
.
For integer n, you can write n+1
as -~n
, n-1
as ~-n
. This uses the same number of characters, but can indirectly cut spaces or parens for operator precedence.
class Solution:
def generateMatrix(self, n: int) -> List[List[int]]:
r=range(n);return[[4*(n-(a:=min(min(i,n-i-1),min(j,n-j-1))))
*a+(i+j-2*a+1,4*(n-2*a-1)-(i+j-2*a)+1)[i>j] for j in r] for i in r]
class Solution:
def generateMatrix(self, n: int) -> List[List[int]]:
r=range(n);return[[4*(n-(a:=min(i,j,~i+n,~j+n)))
*a+(i+j-2*a+1,4*n-6*a-i-j-3)[i>j]for j in r]for i in r]
You can replace 0 if x==y else z
with x-y and z
, it's a little bit counterintuitive, but shorter.
Condition x if c else y
can be written as c and x or y
, it's shorter but depends on x (x should not be 0).
class Solution:
def snakesAndLadders(self, board: List[List[int]]) -> int:
n,v,q = len(board),{1:0},[1]
def f(i):
x = (i - 1)%n
y = (i - 1)//n
c = board[~y][~x if y%2 else x]
return c if c>0 else i
for i in q:
for j in range(i+1, i+7):
k = f(j)
if k==n*n:
return v[i]+1
if k not in v:
v[k] = v[i]+1
q.append(k)
return -1
class Solution:
def snakesAndLadders(self, board: List[List[int]]) -> int:
return (n:=len(board),v:={1:0},q:=[1]) and next((v[i]+1 for i in q for j in range(i+1,i+7)
if (k:=(x:=(j-1)%n,y:=(j-1)//n) and ((c:=board[~y][y%2 and ~x or x])>0 and c or j))==n*n
or (k not in v and (v.update({k:v[i]+1}) or q.append(k)))),-1)
You can check if any of the numbers is negative as x|y<0
or if both numbers are non-zero as x|y
.
class Solution:
def minPathSum(self, grid: List[List[int]]) -> int:
return (f:=cache(lambda i,j:i|j<0 and inf or grid[i][j]+(i|j and min(f(i,j-1),f(i-1,j)))))
(len(grid)-1,len(grid[0])-1)
You can use bitwise &
,|
instead of and
,or
where possible. You can use x&1
instead of x==1
, if 0<=x<=2.
class Solution:
def isScramble(self, s1: str, s2: str) -> bool:
return (f:=cache(lambda a,b:a==b or any((f(a[:i],b[:i]) and f(a[i:],b[i:]))
or (f(a[i:],b[:-i]) and f(a[:i],b[-i:])) for i in range(1,len(a)))))(s1,s2)
class Solution:
def isScramble(self, s1: str, s2: str) -> bool:
return (f:=cache(lambda a,b:a==b or any((f(a[:i],b[:i])&f(a[i:],b[i:]))
|(f(a[i:],b[:-i])&f(a[:i],b[-i:])) for i in range(1,len(a)))))(s1,s2)
You can use booleans as indexes in lists, even nested: (a,(b,c)[u==w])[x==y]
, or you can multiply by a boolean.
class Solution:
def removeStars(self, s: str) -> str:
return reduce(lambda r,c:(r[:-1],r+c)[c>'*'],s)
class Solution:
def simplifyPath(self, path: str) -> str:
return'/'+'/'.join(reduce(lambda r,p:(r+[p]*('.'!=p!=''),r[:-1])[p=='..'],path.split('/'),[]))
Python 3 lacks cmp
(3-way compare) and sign function (copysign(bool(x),x)
is too long), but you can use (x>0)-(x<0)
for sign(x)
and (a>b)-(a<b)
for cmp(a,b)
. Note you can use -1,0,1
indexes for Python lists natively.
class Solution:
def stoneGameIII(self, v: List[int]) -> str:
f=cache(lambda i:i<len(v)and max(sum(v[i:i+k])-f(i+k)for k in(1,2,3)));x=f(0);
return('Tie','Alice','Bob')[(x>0)-(x<0)]
# return(('Tie','Bob')[x<0],'Alice')[x>0] # or like this (1 char shorter)
You can replace cmp
written as lambda x:(x>0)-(x<0)
with 0..__le__
or .0.__le__
(11 characters shorter).
class Solution:
def rearrangeArray(self, n: List[int]) -> List[int]:
n.sort(key=lambda x:(x>0)-(x<0));return chain(*zip(n[len(n)//2:],n))
class Solution:
def rearrangeArray(self, n: List[int]) -> List[int]:
n.sort(key=0..__le__);return chain(*zip(n[len(n)//2:],n))
You can replace x>0
predicate with 0..__lt___
function and replace x!=0
with operator.truth
or just bool
:
class Solution:
def mergeNodes(self, h: Optional[ListNode]) -> Optional[ListNode]:
return h.deserialize(str([sum(v)for k,v in groupby(eval(h.serialize(h)),bool)if k]))
You can save a few characters using asterisk operator *
.
One *
means "expand this as a list", two **
means "expand this as a dictionary".
class Solution:
def checkStraightLine(self, p):
(a,b),(c,d)=p[:2];return all((x-a)*(d-b)==(c-a)*(y-b)for x,y in p)
class Solution:
def checkStraightLine(self, p):
(a,b),(c,d),*_=p;return all((x-a)*(d-b)==(c-a)*(y-b)for x,y in p)
class Solution:
def findMaxAverage(self, n: List[int], k: int) -> float:
s=[0]+[*accumulate(n)];return max(map(sub,s[k:],s))/k
class Solution:
def findMaxAverage(self, n: List[int], k: int) -> float:
s=[0,*accumulate(n)];return max(map(sub,s[k:],s))/k
Quite a few things become shorter with statistics.mode
(most common value of discrete or nominal data).
class Solution:
def findErrorNums(self, nums: List[int]) -> List[int]:
t=sum({*nums});return sum(nums)-t,comb(len(nums)+1,2)-t
class Solution:
def findErrorNums(self, nums: List[int]) -> List[int]:
return mode(nums),comb(len(nums)+1,2)-sum({*nums})
class Solution:
def majorityElement(self, nums: List[int]) -> int:
return sorted(nums)[len(nums)//2]
class Solution:
def majorityElement(self, nums: List[int]) -> int:
return mode(nums)
# https://youtu.be/pKO9UjSeLew (Joma Tech: If Programming Was An Anime)
class Solution:
def findDuplicate(self, nums: List[int]) -> int:
tortoise = hare = nums[0]
while True:
tortoise = nums[tortoise]
hare = nums[nums[hare]]
if tortoise == hare:
break
tortoise = nums[0]
while tortoise != hare:
tortoise = nums[tortoise]
hare = nums[hare]
return hare
class Solution:
def findDuplicate(self, nums: List[int]) -> int:
return mode(nums)
You can use s.encode()
instead of map(ord,s)
It's the same length but doesn't need generation evaluation.
class Solution:
def scoreOfString(self, s: str) -> int:
return sum(abs(x-y)for x,y in pairwise(map(ord,s)))
class Solution:
def scoreOfString(self, s: str) -> int:
return sum(map(abs,map(sub,s:=s.encode(),s[1:])))
Applying a function to an iterable with starmap
and pairwise
may be done with map
(12 chars shorter):
class Solution:
def findArray(self, p: List[int]) -> List[int]:
return starmap(xor,pairwise([0]+p))
class Solution:
def findArray(self, p: List[int]) -> List[int]:
return map(xor,p,[0]+p)
Very often you can replace zip
with map
, it evaluates iterables the same way:
class Solution:
def minMovesToSeat(self, s: List[int], t: List[int]) -> int:
return sum(abs(a-b)for a,b in zip(*map(sorted,(s,t))))
class Solution:
def minMovesToSeat(self, s: List[int], t: List[int]) -> int:
return sum(map(abs,map(sub,*map(sorted,(s,t)))))
You can use numpy.convolve
for sliding windows, it's usually shorter than reduce or list comprehension:
class Solution:
def maxSatisfied(self, c: List[int], g: List[int], m: int) -> int:
t,a=0,[*map(mul,c,g)];[t:=(t,w:=sum(a[i:i+m]))[w>t]for i in range(len(c)-m+1)];return t\
+sum(c)-sum(a)
class Solution:
def maxSatisfied(self, c: List[int], g: List[int], m: int) -> int:
return max(__import__('numpy').convolve(a:=[*map(mul,c,g)],[1]*m))+sum(c)-sum(a)
You can use count()
and map
to replace an enumerate
list comprehension (a few characters shorter):
class Solution:
def maximumImportance(self, n: int, r: List[List[int]]) -> int:
return sum(v*(n-i)for i,(_,v)in enumerate(Counter(chain(*r)).most_common()))
class Solution:
def maximumImportance(self, n: int, r: List[List[int]]) -> int:
return-sum(map(mul,count(-n),sorted(Counter(chain(*r)).values())[::-1]))
You can replace ceil(x/k)
with -(-x//k)
(1 character shorter):
class Solution:
def maxKelements(self, a: List[int], k: int) -> int:
a.sort();return sum((x:=a.pop(),insort(a,ceil(x/3)))[0]for _ in range(k))
class Solution:
def maxKelements(self, a: List[int], k: int) -> int:
a.sort();return sum((x:=a.pop(),insort(a,-(-x//3)))[0]for _ in range(k))
- An expression like
x&(x-1)==0
is useful to check if unsignedx
is power of 2 or 0 (Kernighan, rightmost bit). - Unless the following token starts with e or E, you can remove the space following a number. E.g.
i==4 and j==4
becomesi==4and j==4
. - There's a nice way to convert an iterable to list using star operator, e.g.
x=[*g]
equals*x,=g
(1 char shorter). You can also use this syntax to unpack iterables, e.g.a,*b,c=range(5)
meansa=1;b=[2,3,4];c=5
. - Instead of range(x), you can use the * operator on a list of anything, e.g.
[1]*8
can replacerange(8)
(unless you really need the counter value). - Conditions like
if i<len(r)
may be replaced withif r[i:]
, it's 3 characters shorter. - You can replace
set(n)
with{*n}
(2 characters shorter). - You can convert bool with
~~()
instead ofint()
(as in js) or prepend with a single+
(5 characters shorter). - You can subtract 1 or replace
not
operator with bitwise negation~-
to save on space (1-5 characters shorter). - You can check for set membership with
{x}&s
instead ofx in s
(1 character shorter). - Very often
x==0
can be replaced withx<1
(1 character shorter). - Generator expansion
[*g]
can use a traling comma*g,
in the initialization section (1 character shorter). - A condition like
h>i>=0<=j<w
can be written ash>i>-1<j<w
(1 character shorter). - You can replace
q and q[-1]==c
withq[-1:]==[c]
(3 characters shorter).
- They're not characters at all, they're just brilliant repositories of fantastic, killer one-liners (Stephen Fry interview)
- Don't Piss down my back and tell me it's raining: Notebook for Sarcastic, Witty, and Sharp Tongued One Liners
- Python One-Liners: Write Concise, Eloquent Python Like a Professional
- Python One-Liners - Concise Python Code
- Tips for Golfing in Python (Stack Overflow)
- One-line coder makes me depressed (Reddit)