profiling different flavors of dictionary implementations

dictprofile.py

import sys
from gffutils.helpers import DefaultOrderedDict
from collections import defaultdict
from collections import OrderedDict
import random

kind = sys.argv[1]

nlines = 200000
nkeys = 5
vals_per_key = 3


seen = set()
order = []


def dictupdate(d, k, v, seen, order):
    """
    Append value `v` to list at key `k` in dict `d`.
    """
    val = d.setdefault(k, [])
    val.append(v)
    return seen, order


def defaultupdate(d, k, v, seen, order):
    """
    Append value `v` to list at key `k` in dict `d`.
    """
    d[k].append(v)
    return seen, order

def naiveorderupdate(d, k, v, seen, order):
    defaultupdate(d, k, v, seen, order)
    if k not in seen:
        seen.update([k])
        order.append(k)
    return seen, order

lookup = {
    'dict': (lambda: dict(), dictupdate),
    'defaultordered': (lambda: DefaultOrderedDict(list), defaultupdate),
    'default': (lambda: defaultdict(list), defaultupdate),
    'ordered': (lambda: OrderedDict(), dictupdate),
    'naiveordered': (lambda: defaultdict(list), naiveorderupdate),
}

func, update = lookup[kind]

for line in range(nlines):
    d = func()
    seen = set()
    order = []
    for i in range(nkeys):
        for j in range(vals_per_key):
            seen, order = update(d, i, j, seen, order)

# check at least the last created dict to make sure it's correct
assert dict(d) == {0: [0, 1, 2], 1: [0, 1, 2], 2: [0, 1, 2], 3: [0, 1, 2], 4: [0, 1, 2]}