Tips for golfing in Python

Question

What general tips do you have for golfing in Python? I'm looking for ideas which can be applied to code-golf problems and which are also at least somewhat specific to Python (e.g. "remove comments" is not an answer).

Please post one tip per answer.

Answer 1

Use a=b=c=0 instead of a,b,c=0,0,0. (Note that this uses the same instance for each variable, so don't do this with objects like lists if you intend to mutate them independently)

Use a,b,c='123' instead of a,b,c='1','2','3'.

Answer 2

Conditionals can be lengthy. In some cases, you can replace a simple conditional with (a,b)[condition]. If condition is true, then b is returned.

Compare

if a<b:return a
else:return b

To this

return(b,a)[a<b]

Answer 3

A great thing I did once is:

if 3 > a > 1 < b < 5: foo()

instead of:

if a > 1 and b > 1 and 3 > a and 5 > b: foo()

Python’s comparison operators rock.

---

Using that everything is comparable in Python 2, you can also avoid the and operator this way. For example, if a, b, c and d are integers,

if a<b and c>d:foo()

can be shortened by one character to:

if a<b<[]>c>d:foo()

This uses that every list is larger than any integer.

If c and d are lists, this gets even better:

if a<b<c>d:foo()

Answer 4

If you're using a built-in function repeatedly, it might be more space-efficient to give it a new name, if using different arguments:

r=range
for x in r(10):
 for y in r(100):print x,y

Answer 5

Use string substitution and exec to deal with long keywords like lambda that are repeated often in your code.

a=lambda b:lambda c:lambda d:lambda e:lambda f:0   # 48 bytes  (plain)
exec"a=`b:`c:`d:`e:`f:0".replace('`','lambda ')    # 47 bytes  (replace)
exec"a=%sb:%sc:%sd:%se:%sf:0"%(('lambda ',)*5)     # 46 bytes  (%)

The target string is very often 'lambda ', which is 7 bytes long. Suppose your code snippet contains n occurences of 'lambda ', and is s bytes long. Then:

• The plain option is s bytes long.
• The replace option is s - 6n + 29 bytes long.
• The % option is s - 5n + 22 + len(str(n)) bytes long.

From a plot of bytes saved over plain for these three options, we can see that:

• For n < 5 lambdas, you're better off not doing anything fancy at all.
• For n = 5, writing exec"..."%(('lambda ',)*5) saves 2 bytes, and is your best option.
• For n > 5, writing exec"...".replace('`','lambda ') is your best option.

For other cases, you can index the table below:

          1  2  3  4  5  6  7  8  9 10 11 12 13 14 15 16 17 18 19 (occurences)
       +---------------------------------------------------------
     3 |  -  -  -  -  -  -  -  -  -  -  -  -  -  -  r  r  r  r  r  
     4 |  -  -  -  -  -  -  -  -  -  r  r  r  r  r  r  r  r  r  r  
     5 |  -  -  -  -  -  -  -  r  r  r  r  r  r  r  r  r  r  r  r  
     6 |  -  -  -  -  -  r  r  r  r  r  r  r  r  r  r  r  r  r  r  
     7 |  -  -  -  -  %  r  r  r  r  r  r  r  r  r  r  r  r  r  r  
     8 |  -  -  -  %  %  r  r  r  r  r  r  r  r  r  r  r  r  r  r  
     9 |  -  -  -  %  %  r  r  r  r  r  r  r  r  r  r  r  r  r  r  
    10 |  -  -  %  %  %  r  r  r  r  r  r  r  r  r  r  r  r  r  r  
    11 |  -  -  %  %  %  r  r  r  r  r  r  r  r  r  r  r  r  r  r  
    12 |  -  -  %  %  %  r  r  r  r  r  r  r  r  r  r  r  r  r  r   r = replace
    13 |  -  -  %  %  %  r  r  r  r  r  r  r  r  r  r  r  r  r  r   % = string %
    14 |  -  %  %  %  %  r  r  r  r  r  r  r  r  r  r  r  r  r  r   - = do nothing
    15 |  -  %  %  %  %  r  r  r  r  r  r  r  r  r  r  r  r  r  r  
  (length)

For example, if the string lambda x,y: (length 11) occurs 3 times in your code, you're better off writing exec"..."%(('lambda x,y:',)*3).

Answer 6

Sometimes your Python code requires you to have 2 levels of indentation. The obvious thing to do is use one and two spaces for each indentation level.

However, Python 2 considers the tab and space characters to be different indenting levels.

This means the first indentation level can be one space and the second can be one tab character.

For example:

if 1:
 if 1:
	pass

Answer 7

Store lookup tables as magic numbers

Say you want to hardcode a Boolean lookup table, like which of the first twelve English numbers contain an n.

0: False
1: True
2: False
3: False
4: False
5: False
6: False
7: True
8: False
9: True
10:True
11:True
12:False

Then, you can implement this lookup table concisely as:

3714>>i&1

with the resulting 0 or 1 being equal to False to True.

The idea is that the magic number stores the table as a bitstring bin(3714) = 0b111010000010, with the n-th digit (from the end) corresponding the the nth table entry. We access the nth entry by bitshifting the number n spaces to the right and taking the last digit by &1.

This storage method is very efficient. Compare to the alternatives

n in[1,7,9,10,11]
'0111010000010'[n]>'0'

You can have your lookup table store multibit entries that can be extracted like

 340954054>>4*n&15

to extract the relevant four-bit block.

Answer 8

Use extended slicing to select one string from many

>>> for x in 0,1,2:print"fbboaaorz"[x::3]
... 
foo
bar
baz

vs

>>> for x in 0,1,2:print["foo","bar","baz"][x]
... 
foo
bar
baz

In this Boolean two-string case, one can also write

b*"string"or"other_string"

for

["other_string","string"][b]

Unlike interleaving, this works for strings of any length, but can have operator precedence issues if b is instead an expression.

Answer 9

Collapse two numerical loops into one

Say you're iterating over the cells of an m*n grid. Instead of two nested for loops, one for the rows and one for the columns, it's usually shorter to write a single loop to iterate over the m*n cells of the grid. You can extract the row and column of the cell inside the loop.

Original code:

for i in range(m):
 for j in range(n):
  do_stuff(i,j)

Golfed code:

for k in range(m*n):
 do_stuff(k/n,k%n)

In effect, you're iterating over the Cartesian product of the two ranges, encoding the pair (i,j) as x=i*n+j. You've save a costly range call and a level of indentation inside the loop. The order of iteration is unchanged.

Use // instead of / in Python 3. If you refer to i and j many times, it may be shorter to assign their values i=k/n, j=k%n inside the loop.

Answer 10

Use `n` to convert an integer to a string instead of using str(n):

>>> n=123
>>> `n`
'123'

Note: Only works in Python 2.

Answer 11

For integer n, you can write

• n+1 as -~n
• n-1 as ~-n

because the bit flip ~x equals -1-x. This uses the same number of characters, but can indirectly cut spaces or parens for operator precedence.

Compare:

while n-1:  #Same as while n!=1 
while~-n:

c/(n-1)
c/~-n

or f(n)+1
or-~f(n) 

(n-1)/10+(n-1)%10
~-n/10+~-n%10

The operators ~ and unary - are higher precedence than *, /, %, unlike binary +.

Answer 12

Unless the following token starts with e or E. You can remove the space following a number.

For instance:

if i==4 and j==4:
    pass

Becomes:

if i==4and j==4:
    pass

Using this in complicated one line statements can save quite a few characters.

EDIT: as @marcog pointed out, 4or a will work, but not a or4 as this gets confused with a variable name.

Answer 13

A nice way to convert an iterable to list on Python 3:

imagine you have some iterable, like

i = (1,2,3,4)
i = range(4)
i = (x**2 for x in range(5))

But you need a list:

x=list(i)  #the default way
*x,=i      #using starred assignment -> 4 char fewer

It's very useful to make a list of chars out of a string

s=['a','b','c','d','e']
s=list('abcde')
*s,='abcde'

Answer 14

Extended iterable unpacking ("Starred assignment", Python 3 only)

The best way to explain this is via an example:

>>> a,*b,c=range(5)
>>> a
0
>>> b
[1, 2, 3]
>>> c
4

We've already seen a use for this — turning an iterable into a list in Python 3:

a=list(range(10))
*a,=range(10)

Here are a few more uses.

Getting the last element from a list

a=L[-1]
*_,a=L

In some situations, this can also be used for getting the first element to save on parens:

a=(L+[1])[0]
a,*_=L+[1]

Assigning an empty list and other variables

a=1;b=2;c=[]
a,b,*c=1,2

Removing the first or last element of a non-empty list

_,*L=L
*L,_=L

These are shorter than the alternatives L=L[1:] and L.pop(). The result can also be saved to a different list.

Tips courtesy of @grc

Answer 15

Instead of range(x), you can use the * operator on a list of anything, if you don't actually need to use the value of i:

for i in[1]*8:pass

as opposed to

for i in range(8):pass

If you need to do this more than twice, you could assign any iterable to a variable, and multiply that variable by the range you want:

r=1,
for i in r*8:pass
for i in r*1000:pass

Note: this is often longer than exec"pass;"*8, so this trick should only be used when that isn't an option.

Answer 16

You can use the good old alien smiley face to reverse sequences:

[1, 2, 3, 4][::-1] # => [4, 3, 2, 1]

Answer 17

Use ~ to index from the back of a list

If L is a list, use L[~i] to get the i'th element from the back.

This is the i'th element of the reverse of L. The bit complement ~i equals -i-1, and so fixes the off-by-one error from L[-i].

Answer 18

For ages it bothered me that I couldn't think of a short way to get the entire alphabet. If you use range enough that R=range is worth having in your program, then

[chr(i+97)for i in R(26)]

is shorter than the naive

'abcdefghijklmnopqrstuvwxyz'

, but otherwise it's longer by a single character. It haunted me that the clever one that required some knowledge of ascii values ended up being more verbose than just typing all the letters.

Until I saw this answer for My Daughter's Alphabet. I can't follow the edit history well enough to figure out if this genius was the work of the OP or if it was a suggestion by a commenter, but this is (I believe) the shortest way to create an iterable of the 26 letters in the Roman alphabet.

map(chr,range(97,123))

If case doesn't matter, you can strip off another character by using uppercase:

map(chr,range(65,91))

I use map way too much, I don't know how this never occurred to me.

Answer 19

set literals in Python2.7

You can write sets like this S={1,2,3} This also means you can check for membership using {e}&S instead of e in S which saves one character.

Answer 20

Choosing one of two numbers based on a condition

You already know to use the list selection [x,y][b] with a Boolean b for the ternary expression y if b else x. The variables x, y, and b can also be expressions, though note that both x and y are evaluated even when not selected.

Here's some potential optimizations when x and y are numbers.

• [0,y][b] -> y*b
• [1,y][b] -> y**b
• [x,1][b] -> b or x
• [x,x+1][b] -> x+b
• [x,x-1][b] -> x-b
• [1,-1][b] -> 1|-b
• [x,~x][b] -> x^-b
• [x,y][b] -> x+z*b (or y-z*b), where z=y-x.

You can also switch x and y if you can rewrite b to be its negation instead.

Answer 21

When you have two boolean values, a and b, if you want to find out if both a and b are true, use * instead of and:

if a and b: #7 chars

vs

if a*b: #3 chars

if either value is false, it will evaluate as 0 in that statement, and an integer value is only true if it is nonzero.

Answer 22

Although python doesn't have very golfable switch statements (the match/case syntax since Python 3.10), you can emulate them with dictionaries. For example, if you wanted to golf this:

match a:
 case 1: runCodeOne()
 case 2: runCodeTwo()
 case 3: runCodeThree()
 case _: runDefault()

You could use if statements, or you could use one of these:

# Code as strings
exec{1:"runCodeOne()",2:"runCodeTwo()",3:"runCodeThree()"}[a]

# Functions with non-consecutive indices
{1:runCodeOne,4:runCodeFour,30:runCodeThirty)[a]()

# Functions with consecutive indices
# (you can't add a default value)
(0,runCodeOne,runCodeTwo,runCodeThree)[a]()

# Code as strings (and default code)
exec{ <same as before> }.get(a,"runDefault()")

# Run functions (and default function)
{ <same as before> }­.get(a,runDefault)()

This can be used to condense return statements:

# Before
def getValue(k):
 if k=='blah':return 1
 if k=='foo':return 2
 if k=='bar':return 3
 return 4

# After
# getValue = ...
lambda k:{'blah':1,'foo':2,'bar':3}.get(k,4)

Answer 23

PEP448 – Additional Unpacking Generalizations

With the release of Python 3.5, manipulation of lists, tuples, sets and dicts just got golfier.

Turning an iterable into a set/list

Compare the pairs:

set(T)
{*T}

list(T)
[*T]

tuple(T)
(*T,)

Much shorter! Note, however, that if you just want to convert something to a list and assign it to a variable, normal extended iterable unpacking is shorter:

L=[*T]
*L,=T

A similar syntax works for tuples:

T=*L,

which is like extended iterable unpacking, but with the asterisk and comma on the other side.

Joining lists/tuples

Unpacking is slightly shorter than concatenation if you need to append a list/tuple to both sides:

[1]+T+[2]
[1,*T,2]

(1,)+T+(2,)
(1,*T,2)

Printing the contents of multiple lists

This isn't limited to print, but it's definitely where most of the mileage will come from. PEP448 now allows for multiple unpacking, like so:

>>> T = (1, 2, 3)
>>> L = [4, 5, 6]
>>> print(*T,*L)
1 2 3 4 5 6

Updating multiple dictionary items

This probably won't happen very often, but the syntax can be used to save on updating dictionaries if you're updating at least three items:

d[0]=1;d[1]=3;d[2]=5
d={**d,0:1,1:3,2:5}

This basically negates any need for dict.update.

Answer 24

Ceil and Floor

If you ever want to get the rounded-up result for a division, much like you'd do with // for floor, you could use math.ceil(3/2) for 15 or the much shorter -(-3//2) for 8 bytes.

math.floor(n)   : 13 bytes+12 for import
n//1            : 4  bytes

math.ceil(n)    : 12 bytes+12 for import
-(-n//1)        : 8  bytes

Answer 25

loops up to 4 items may be better to supply a tuple instead of using range

for x in 0,1,2:

vs

for x in range(3):

Answer 26

Use += instead of append and extend

A.append(B)  

can be shortened to:

A+=B,

B, here creates a one-element tuple which can be used to extend A just like [B] in A+=[B].

---

A.extend(B)

can be shortened to:

A+=B

Answer 27

Change import * to import*

---

If you haven't heard, import* saves chars!

from math import*

is only 1 character longer than import math as m and you get to remove all instances of m.

Even one time use is a saver!

Answer 28

A one line function can be done with lambda:

def c(a):
  if a < 3: return a+10
  else: return a-5

can be converted to (note missing space 3and and 10or)

c=lambda a:a<3and a+10or a-5

Answer 29

Exploit Python 2 string representations

Python 2 lets you convert an object x to its string representation `x` at a cost of only 2 chars. Use this for tasks that are easier done on the object's string than the object itself.

Join characters

Given a list of characters l=['a','b','c'], one can produce ''.join(l) as `l`[2::5], which saves a byte.

The reason is that `l` is "['a', 'b', 'c']" (with spaces), so one can extract the letters with a list slice, starting that the second zero-indexed character a, and taking every fifth character from there. This doesn't work to join multi-character strings or escape characters represented like '\n'.

Concatenate digits

Similarly, given a non-empty list of digits like l=[0,3,5], one can concatenate them into a string '035' as `l`[1::3].

This saves doing something like map(str,l). Note that they must be single digits, and can't have floats like 1.0 mixed in. Also, this fails on the empty list, producing ].

Check for negatives

Now, for a non-string task. Suppose you have a list l of real numbers and want to test if it contains any negative numbers, producing a Boolean.

You can do

'-'in`l`

which checks for a negative sign in the string rep. This shorter than either of

any(x<0for x in l)
min(l+[0])<0   

For the second, min(l)<0 would fail on the empty list, so you have to hedge.

Answer 30

Length tradeoff reference

I've think it would be useful to have a reference for the character count differences for some common alternative ways of doing things, so that I can know when to use which. I'll use _ to indicate an expression or piece of code.

Assign to a variable: +4

x=_;x
_

So, this breaks even if you

• Use _ a second time: _ has length 5
• Use _ a third time: _ has length 3

Assign variables separately: 0

x,y=a,b
x=a;y=b

• -2 when a equals b for x=y=a

Expand lambda to function def: +7

lambda x:_
def f(x):return _

• -2 for named functions
• -1 if _ can touch on the left
• -1 in Python 2 if can print rather than return
• +1 for starred input *x

Generically, if you're def to save an expression to a variable used twice, this breaks even when the expression is length 12.

lambda x:g(123456789012,123456789012)
def f(x):s=123456789012;return g(s,s)

STDIN rather than function: +1

def f(x):_;print s
x=input();_;print s

• -1 for line of code needed in _ if not single-line
• +4 if raw_input needed in Python 2
• -4 if input variable used only once
• +1 if function must return rather than print in Python 2

Use exec rather than looping over range(n): +0

for i in range(n):_
i=0;exec"_;i+=1;"*n

• +2 for Python 3 exec()
• -4 if shifted range range(c,c+n) for single-char c
• -5 when going backwards from n to 1 via range(n,0,-1)
• -9 if index variable never used

Apply map manually in a loop: +0

for x in l:y=f(x);_
for y in map(f,l):_

Apply map manually in a list comprehension: +8

map(f,l)
[f(x)for x in l]

• -12 when f must be written in the map as the lambda expression lambda x:f(x), causing overall 4 char loss.

Apply filter manually in a list comprehension: +11

filter(f,l)
[x for x in l if f(x)]

• -1 if f(x) expression can touch on the left
• -12 when f must be written in the filter as the lambda expression lambda x:f(x), causing overall 1 char loss.

Import versus import single-use: +4*

import _;_.f
from _ import*;f

• Breaks even when _ has length 5
• import _ as x;x.f is always worse except for multiple imports
• __import__('_').f is also worse

Thanks to @Sp3000 for lots of suggestions and fixes.