# Read list of dictionaries with nested dictionaries

I have a file as content a list of dictionaries (Around 75000). For instance, this is an example of first line I got when reading the file (value for v):

{
"id": 1,
"name": "Explosives",
"category_id": 1,
"average_price": 294,
"is_rare": 0,
"max_sell_price": 1774,
"min_sell_price": 18,
"sell_price_upper_average": 924,
"is_non_marketable": 0,
"ed_id": 128049204,
"category": {
"id": 1,
"name": "Chemicals"
}
}


My actual working code is :

for v in d:
commodities_reference = []
for k, g in v.items():
if isinstance(g, dict):
dict1 = g
my_value1 = dict1.get("id")
my_value2 = dict1.get("name")

for s, i in v.items():
if not isinstance(i, dict):
commodities_reference.append(i)
commodities_reference.append(my_value1)
commodities_reference.append(my_value2)


Output wanted = All the values in same list in the same order for doing a SQL INSERT Statement afterwards (Meaning values from the nested dict must be also at the end.)

[1, 'Explosives', 1, 294, 0, 755, 1774, 99, 18, 176, 924, 0, 128049204, 1, 'Chemicals']


From a performance perspective, with SQLITE3/python 3.7, it is a catastrophe. I am looking for some advices in order to make it more efficient. I am thinking about using executemany statement but it seems it takes tuple instead of list.

• I'm not sure I fully understand the requirement (same order as what?) - could you show the actual expected output from the given input? That would make it clearer. – Toby Speight Nov 14 '18 at 14:12
• the data example I gave is the content of v, not d. In my case, d has 333 dictionaries in it. – Deareim Nov 14 '18 at 14:21
• Is the problem the code you posted or sqlite insert? Seems more likely to be the insert – juvian Nov 14 '18 at 20:21
• I solve the issue of the insert by using executemany instead with tuple. But it doesn't solve the code issue as it seems to me it could more optimized. – Deareim Nov 14 '18 at 20:43

Currently you iterate over each dictionary twice. You can do it in one pass:

for v in d:
commodities_reference = []
for k, g in v.items():
if isinstance(g, dict):
commodities_reference.append(g["id"])
commodities_reference.append(g["name"])
else:
commodities_reference.append(g)


Note that this appends the values when it encounters it. This means that in Python < 3.7 (cPython < 3.6) it is not guaranteed that the dictionary is actually the last item to be looked at, since dictionaries were not to guaranteed to be in insertion order.

You could even make this a generator and slightly more general:

def get_values_recursive(x):
for value in x.values():
if isinstance(value, dict):
yield from get_values_recursive(value)
else:
yield value

for v in d:
commodities_reference = list(get_values_recursive(v))
# do something with it...
print(commodities_reference)


When using the given example, this is the result:

>>> list(get_values_recursive(v))
[1, 'Explosives', 1, 294, 0, 755, 1774, 99, 18, 176, 924, 0, 128049204, 1, 'Chemicals']


When putting your code into a function, this generator is almost twice as fast with the given v:

In [13]: %timeit op(v)
5.32 µs ± 43.3 ns per loop (mean ± std. dev. of 7 runs, 100000 loops each)

In [14]: %timeit list(get_values_recursive(v))
3.64 µs ± 10.8 ns per loop (mean ± std. dev. of 7 runs, 100000 loops each)


Note that both take on the order of micro seconds, so unless you need to process more than 100000 items per second your bottleneck is probably in those SQL statements and how you execute them.

• So I have tested your "general' function and I see a lot of performance improvment, so thank you again. Just missing the fact that I need the data without dictionnaries inside like in my example. That is why I had a get on dictionnary. – Deareim Nov 16 '18 at 10:54
• @Deareim: I don't understand. When I call the function with the given example there are no dictionaries left in the output (see edited answer). – Graipher Nov 16 '18 at 12:16