Best way to define multidimensional dictionaries in python?

Question

I'm currently using the method below to define a multidimensional dictionary in python. My question is: Is this the preferred way of defining multidimensional dicts?

from collections import defaultdict

def site_struct(): 
    return defaultdict(board_struct)

def board_struct(): 
    return defaultdict(user_struct)

def user_struct(): 
    return dict(pageviews=0,username='',comments=0)

userdict = defaultdict(site_struct)

to get the following structure:

userdict['site1']['board1']['username'] = 'tommy'

I'm also using this to increment counters on the fly for a user without having to check if a key exists or is set to 0 already. E.g.:

userdict['site1']['board1']['username']['pageviews'] += 1

this code will not run the in python interpreter. this makes it hard to understand what you're trying to do. — krumpelstiltskin, May 26 '13 at 00:37

score 45 · Answer 1 · answered Mar 05 '10 at 17:01

45

Tuples are hashable. Probably I'm missing the point, but why don't you use a standard dictionary with the convention that the keys will be triples? For example:

userdict = {}
userdict[('site1', 'board1', 'username')] = 'tommy'

answered Mar 05 '10 at 17:01

Federico A. Ramponi

43,319
29
102
130

1

I recommend this approach, unless you need to pass userdict['site1'] to some function. – gurney alex Mar 05 '10 at 17:04
I like this idea. This is exactly what tuples are good for because of their immutable nature. – jathanism Mar 05 '10 at 17:04
2

This works very well, as long as you don't need to list all the board entries under 'site1'. – Shane Holloway Mar 05 '10 at 18:03
1

using this approach I won't be able to do things like.. userdict[('site1', 'board1', 'user1', 'pageviews)] += 1 – James Mar 05 '10 at 18:53
got this: KeyError: ('site1', 'board1', 'user1', 'pageviews') – James Mar 05 '10 at 19:07
@beagleguy Use `userdict = defaultdict(int)` – Roberto Bonvallet Mar 05 '10 at 19:08
3

And you don't need the parentheses: `userdict['site1', 'board1', 'username'] = 'tommy'`. – Roberto Bonvallet Mar 05 '10 at 19:08
@beagleguy: the value needs to exist before you can add one to it. Once it exists, though, you can increment it with += 1 – Bryan Oakley Mar 05 '10 at 19:58
@Bryan I'm aware the value needs to exist first but if I define the default dict as shown it sets it for me to 0. Just wasn't sure if there was another way to do it. – James Mar 05 '10 at 20:44
8

You cannot iterate over a single dimension using the tuple approach. – Sridhar Ratnakumar Apr 06 '11 at 19:27
Another problem with this approach is that the dict cannot be JSON encoded. – tleb Feb 11 '17 at 21:16

andersonvom · Answer 2 · 2012-04-01T00:36:35.753

16

You can create a multidimensional dictionary of any type like this:

from collections import defaultdict
from collections import Counter

def multi_dimensions(n, type):
  """ Creates an n-dimension dictionary where the n-th dimension is of type 'type'
  """  
  if n<=1:
    return type()
  return defaultdict(lambda:multi_dimensions(n-1, type))

>>> m = multi_dimensions(5, Counter)
>>> m['d1']['d2']['d3']['d4']
Counter()

edited Apr 01 '12 at 00:36

answered Apr 01 '12 at 00:29

andersonvom

11,249
4
33
37

When I try this, I get `TypeError: Required argument 'name' (pos 1) not found` which refers to the line `return type()`. – hepcat72 May 23 '16 at 21:47
I figured out my issue - it's just a difference between how I define "N-Dimensional" dictionaries & how it's defined above. E.g. a single datum is indexed by 0 dimensions. A simple array is indexed by 1 dim. The vals in an array of arrays is indexed by 2 dims. And dicts are just arrays indexed by strings. The vals it holds are not (to me) an additional dimension, though I see how you could think of it that way. In the example above, I would've called it a 4D dict, not 5. So my code wasn't working because I was expecting the type instead of a dict @ 1 dimension shy of what it gave me. – hepcat72 May 24 '16 at 13:38

Burhan Khalid · Answer 3 · 2013-01-26T08:50:52.443

This is highly subjective - but if your dict is getting to be more than 2-n deep, I would revert to creating a class and defining methods/properties to do what you want.

I think your end goal would be to create a "multi-column" data structure so you can store any number of attributes for a particular object (in this case, a site). Creating a site class would accomplish that and then you can stick instances wherever you would any other object (variable) - with some noted exceptions.

score 1 · Answer 4 · answered Mar 05 '10 at 18:02

1

This is a pretty subjective question from my perspective. For me, the real question would be at what point do you promote this nested data structure to objects with methods to insulate you from changes. However, I've been known to create large prototyping namespaces with the following:

from collections import defaultdict

def nesteddict(): 
  return defaultdict(nesteddict)

answered Mar 05 '10 at 18:02

Shane Holloway

6,469
4
26
36

This solution preserves the ability to iterate over dimensions, but does create a dependency for nesteddict. – GregB Aug 29 '11 at 23:58

Best way to define multidimensional dictionaries in python?

4 Answers4

Linked