Is a Python dictionary an example of a hash table?

Question

One of the basic data structures in Python is the dictionary, which allows one to record "keys" for looking up "values" of any type. Is this implemented internally as a hash table? If not, what is it?

Phillida

If you're interested in the technical details, one article in Beautiful Code deals with the internals of Python's dict implementation. – Aubyn

Here is a talk by Brandon Craig Rhodes discussing how python dictionary works, youtube.com/watch?v=C4Kc8xzcA68. – Elinorelinore

I looked for a diagram representing a dict for a while now, which decipt the implementation in memory and CPython. Thanks for referencing the book! – Shred

Answer 1

Yes, it is a hash mapping or hash table. You can read a description of python's dict implementation, as written by Tim Peters, here.

That's why you can't use something 'not hashable' as a dict key, like a list:

>>> a = {}
>>> b = ['some', 'list']
>>> hash(b)
Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
TypeError: list objects are unhashable
>>> a[b] = 'some'
Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
TypeError: list objects are unhashable

You can read more about hash tables or check how it has been implemented in python and why it is implemented that way.

Dodecagon

The Tim Peters link seams to be broken, is there a clean link out there? – Hals

@MattAlcock: I've updated the link. Sometimes (usually due to someone wanting their email address removed somewhere) the python list archives are rebuilt and the ids of emails change, thus breaking these links. The pydotorg admins generally try to avoid that these days. – Scientism

But using .keys() can retrieve a list of keys. A real hash table wouldn't store keys, just hashes to save space. – Mosby

More complete description of python dict implementation here: laurentluce.com/posts/python-dictionary-implementation – Tabatha

@noɥʇʎԀʎzɐɹƆ - the key itself is not stored, only a reference to it and the hash. – Dodecagon

Answer 2

There must be more to a Python dictionary than a table lookup on hash(). By brute experimentation I found this hash collision:

>>> hash(1.1)
2040142438
>>> hash(4504.1)
2040142438

Yet it doesn't break the dictionary:

>>> d = { 1.1: 'a', 4504.1: 'b' }
>>> d[1.1]
'a'
>>> d[4504.1]
'b'

Sanity check:

>>> for k,v in d.items(): print(hash(k))
2040142438
2040142438

Possibly there's another lookup level beyond hash() that avoids collisions between dictionary keys. Or maybe dict() uses a different hash.

(By the way, this in Python 2.7.10. Same story in Python 3.4.3 and 3.5.0 with a collision at hash(1.1) == hash(214748749.8).)

(I haven't found any collisions in Python 3.9.6. Since the hashes are bigger -- hash(1.1) == 230584300921369601 -- I estimate it would take my desktop a thousand years to find one. So I'll get back to you on this.)

Eckert

So collisions are unavoidable. Set S may contain an infinitely large number of items, and you want it to hash to a number a computer can store. Every usable implementation of a hash table resolves collisions, with two of the most frequent methods being a) open addressing and b) chaining. Just because it doesn't utilize a perfect hash doesn't mean it's not a hash table. – Canna

Collisions will happen in general, because there are infinite possible hashable values and finite hash codes. Even a hash table would have to handle collision somehow. – Rawley

@YanfengLiu I believe those are exactly the same points TurnipEntropy made. – Eckert

In Python 3.7, it looks like there are 2E20 minus 1 possible hash values, in fact. From -1E20 minus 1 to (+)1E20 minus 1. Try hash('I wandered lonely as a cloud, that drifts on high o\'er vales and hills, when all at once, I saw a crowd, a host of golden daffodils.') This gives a 19-digit decimal - -4037225020714749784 if you're geeky enough to care. Continue in your own words, kids, and the hash is still a 19-digit number. I assume there is a limit on length of string you can hash in Python, but safe to say many more possible strings than possible values. And hash(False) = 0 by the way. – Inactivate

The reason why it doesn't break the dictionary is because under the hood the duplicate values are implemented using a linked list, and they are stored alongside a pointer back to the key they were generated from. – Reveal

There are many. hash(0) = 0; hash(2 ** 61 - 1) = 0; hash(-2 ** 61 + 1) = 0 – Consols

Answer 3

Yes. Internally it is implemented as open hashing based on a primitive polynomial over Z/2 (source).

Lush

Answer 4

To expand upon nosklo's explanation:

a = {}
b = ['some', 'list']
a[b] = 'some' # this won't work
a[tuple(b)] = 'some' # this will, same as a['some', 'list']

Rikki