What's the fastest way to split dictionary keys into a string-type tuples and append another string to last items in the tuples?

user_610569 0
Viewed 336

Given a dictionary of string key and integer values, what's the fastest way to

  1. split each key into a string-type key tuple
  2. then append a special substring </w> to the last item in the tuple

Given:

counter = {'The': 6149,
     'Project': 205,
     'Gutenberg': 78,
     'EBook': 5,
     'of': 39169,
     'Adventures': 2,
     'Sherlock': 95,
     'Holmes': 198,
     'by': 6384,
     'Sir': 30,
     'Arthur': 18,
     'Conan': 3,
     'Doyle': 2,}

The goal is to achieve:

counter = {('T', 'h', 'e</w>'): 6149,
 ('P', 'r', 'o', 'j', 'e', 'c', 't</w>'): 205,
 ('G', 'u', 't', 'e', 'n', 'b', 'e', 'r', 'g</w>'): 78,
 ('E', 'B', 'o', 'o', 'k</w>'): 5,
 ('o', 'f</w>'): 39169,
 ('A', 'd', 'v', 'e', 'n', 't', 'u', 'r', 'e', 's</w>'): 2,
 ('S', 'h', 'e', 'r', 'l', 'o', 'c', 'k</w>'): 95,
 ('H', 'o', 'l', 'm', 'e', 's</w>'): 198,
 ('b', 'y</w>'): 6384,
 ('S', 'i', 'r</w>'): 30,
 ('A', 'r', 't', 'h', 'u', 'r</w>'): 18,
 ('C', 'o', 'n', 'a', 'n</w>'): 3,
 ('D', 'o', 'y', 'l', 'e</w>'): 2,}

One way to do it is to

  • iterate through the counter and
  • converting all but the last character to the tuple
  • add to the tuple and create an outer tuple
  • and assign the tuple key to the count

I've tried

{(tuple(k[:-1])+(k[-1]+'</w>',) ,v) for k,v in counter.items()}

In more verbose form:

new_counter = {}
for k, v in counter.items():
    left = tuple(k[:-1])
    right = tuple(k[-1]+'w',)
    new_k = (left + right,)
    new_counter[new_k] = v

Is there a better way to do this?

Regarding the adding tuple and casting it to an outer tuple. Why is this allowed? Isn't tuple supposed to be immutable?

tuples dictionary python string
6 Answers

I would propose a slightly modified version of your solution. Instead of using tuple constructor you can use tuple unpacking:

>>> {(*a[:-1],f'a[-1]</w>',):b for a,b in counter.items()}

The benefit of using tuple unpacking is you will get better performance as compared to tuple constructor. I will shed some more light on this by using timeit. I will be using randomly generated dict. Each key in the dict will have 2 randomly chosen characters from lower case alphabets and each value will be an integer in range 0-100. For all these benchmarks I am using Python 3.7.0

Benchmark with 100 elements in dict

$ python -m timeit -s "import random" -s "import string" -s "counter = {''.join(random.sample(string.ascii_lowercase,2)): random.randint(0,100) for _ in range(100)}" "{(*a[:-1],f'a[-1]</w>',):b for a,b in counter.items()}
$ 10000 loops, best of 5: 36.6 usec per loop

$ python -m timeit -s "import random" -s "import string" -s "counter = {''.join(random.sample(string.ascii_lowercase,2)): random.randint(0,100) for _ in range(100)}" "{tuple(key[:-1])+(key[-1]+'</w>',):value for key,value in counter.items()}"
$ 5000 loops, best of 5: 59.7 usec per loop

Benchmark with 1000 elements in dict

$ python -m timeit -s "import random" -s "import string" -s "counter = {''.join(random.sample(string.ascii_lowercase,2)): random.randint(0,100) for _ in range(1000)}" "{(*a[:-1],f'a[-1]</w>',):b for a,b in counter.items()}"
$ 1000 loops, best of 5: 192 usec per loop

$ python -m timeit -s "import random" -s "import string" -s "counter = {''.join(random.sample(string.ascii_lowercase,2)): random.randint(0,100) for _ in range(1000)}" "{tuple(key[:-1])+(key[-1]+'</w>',):value for key,value in counter.items()}"
$ 1000 loops, best of 5: 321 usec per loop

Benchmark with dict posted in question

$ python -m timeit -s "import random" -s "import string" -s "counter = counter = {'The': 6149, 'Project': 205, 'Gutenberg': 78, 'EBook': 5, 'of': 39169, 'Adventures': 2, 'Sherlock': 95, 'Holmes': 198, 'by': 6384, 'Sir': 30, 'Arthur': 18, 'Conan': 3,'Doyle': 2}" "{(*a[:-1],f'a[-1]</w>',):b for a,b in counter.items()}"
$ 50000 loops, best of 5: 7.28 usec per loop

$ python -m timeit -s "import random" -s "import string" -s "counter = counter = {'The': 6149, 'Project': 205, 'Gutenberg': 78, 'EBook': 5, 'of': 39169, 'Adventures': 2, 'Sherlock': 95, 'Holmes': 198, 'by': 6384, 'Sir': 30, 'Arthur': 18, 'Conan': 3,'Doyle': 2}" "{tuple(key[:-1])+(key[-1]+'</w>',):value for key,value in counter.items()}"
$ 20000 loops, best of 5: 11 usec per loop
user_57411720

You are close making a little changes to your code using tuple. You cannot modify the elements of a tuple, but you can replace one tuple with another::

{tuple(key[:-1])+(key[-1]+'</w>',):value for key,value in counter.items()}

{('T', 'h', 'e</w>'): 6149,
 ('P', 'r', 'o', 'j', 'e', 'c', 't</w>'): 205,
 ('G', 'u', 't', 'e', 'n', 'b', 'e', 'r', 'g</w>'): 78,
 ('E', 'B', 'o', 'o', 'k</w>'): 5,
 ('o', 'f</w>'): 39169,
 ('A', 'd', 'v', 'e', 'n', 't', 'u', 'r', 'e', 's</w>'): 2,
 ('S', 'h', 'e', 'r', 'l', 'o', 'c', 'k</w>'): 95,
 ('H', 'o', 'l', 'm', 'e', 's</w>'): 198,
 ('b', 'y</w>'): 6384,
 ('S', 'i', 'r</w>'): 30,
 ('A', 'r', 't', 'h', 'u', 'r</w>'): 18,
 ('C', 'o', 'n', 'a', 'n</w>'): 3,
 ('D', 'o', 'y', 'l', 'e</w>'): 2}
user_83208610

Or use str.split, and do str.join and '</w>' adding beforehand:

>>> counter = {'The': 6149,
     'Project': 205,
     'Gutenberg': 78,
     'EBook': 5,
     'of': 39169,
     'Adventures': 2,
     'Sherlock': 95,
     'Holmes': 198,
     'by': 6384,
     'Sir': 30,
     'Arthur': 18,
     'Conan': 3,
     'Doyle': 2,}
>>> {tuple((' '.join(k)+'</w>').split()):v for k,v in counter.items()}
{('T', 'h', 'e</w>'): 6149, ('P', 'r', 'o', 'j', 'e', 'c', 't</w>'): 205, ('G', 'u', 't', 'e', 'n', 'b', 'e', 'r', 'g</w>'): 78, ('E', 'B', 'o', 'o', 'k</w>'): 5, ('o', 'f</w>'): 39169, ('A', 'd', 'v', 'e', 'n', 't', 'u', 'r', 'e', 's</w>'): 2, ('S', 'h', 'e', 'r', 'l', 'o', 'c', 'k</w>'): 95, ('H', 'o', 'l', 'm', 'e', 's</w>'): 198, ('b', 'y</w>'): 6384, ('S', 'i', 'r</w>'): 30, ('A', 'r', 't', 'h', 'u', 'r</w>'): 18, ('C', 'o', 'n', 'a', 'n</w>'): 3, ('D', 'o', 'y', 'l', 'e</w>'): 2}
>>>

Timings:

import timeit
print('bro-grammer:',timeit.timeit(lambda: [{(*a[:-1],f'a[-1]</w>',):b for a,b in counter.items()} for i in range(1000)],number=10))
print('Sandeep Kadapa:',timeit.timeit(lambda: [{tuple(key[:-1])+(key[-1]+'</w>',):value for key,value in counter.items()} for i in range(1000)],number=10))
print('U9-Forward:',timeit.timeit(lambda: [{tuple((' '.join(k)+'</w>').split()):v for k,v in counter.items()} for i in range(1000)],number=10))

Output:

bro-grammer: 0.1293355557653911
Sandeep Kadapa: 0.20885866344797197
U9-Forward: 0.3026948357193003
user_87083640

I would go for something like this:

def f(string):
    l = list(string)
    l[-1] = l[-1] + '</w>'
    return tuple(l)
dict((f(k), v) for k, v in counter.items())

output:

{('A', 'd', 'v', 'e', 'n', 't', 'u', 'r', 'e', 's</w>'): 2,
 ('A', 'r', 't', 'h', 'u', 'r</w>'): 18,
 ('C', 'o', 'n', 'a', 'n</w>'): 3,
 ('D', 'o', 'y', 'l', 'e</w>'): 2,
 ('E', 'B', 'o', 'o', 'k</w>'): 5,
 ('G', 'u', 't', 'e', 'n', 'b', 'e', 'r', 'g</w>'): 78,
 ('H', 'o', 'l', 'm', 'e', 's</w>'): 198,
 ('P', 'r', 'o', 'j', 'e', 'c', 't</w>'): 205,
 ('S', 'h', 'e', 'r', 'l', 'o', 'c', 'k</w>'): 95,
 ('S', 'i', 'r</w>'): 30,
 ('T', 'h', 'e</w>'): 6149,
 ('b', 'y</w>'): 6384,
 ('o', 'f</w>'): 39169}
user_39393180

With Python 3, you can use starred expression in tuples.

You can try:

>>> {(*key[:-1], key[-1] + '</w>'): value for key, value in counter.items()}
user_15139330

You can also remove .items() from your iteration and go like this:

{tuple(i[:-1]) + (i[-1]+'</w>',):counter[i] for i in counter}

This is a little faster.

 timeit.timeit(lambda: {tuple(i[:-1]) + (i[-1]+'w',):counter[i] for i in counter}, number=10)
 0.000192291005179286
user_37447470
Related