heves
How to handle very big enumerables in comprehensions?
For a school assignment, I need to esentially brute-force check every possible solution for a puzzle. This is acceptable, as the task is meant to teach us about prefiltering and optimizing. I have done both of those as best as I could, and now my code will run out of memory for longer puzzles where the number of possible solutions can range from a couple hundred-thousand to a few millions.
# From a list of sublists, generate all possible permutations, using one element from each sublist
def perms_from_lists([]), do: []
def perms_from_lists(list) do
Enum.reduce(list, fn sublist, acc ->
for a <- acc, b <- sublist do
[b | List.wrap(a)]
end
end)
|> Enum.map(&Enum.reverse/1)
end
def solutions(puzzle) do
# This would generate a lot of solutions, possibly (for longer puzzles) in the millions,
# depending on how effective is my prefiltering on the given puzzle
Enum.filter((for x <- perms_from_lists(transform_and_prefilter(puzzle)),
do: if sol_okay?(puzzle, x), do: x), fn x -> x != nil end)
end
I’m guessing the problem may be that my program wants to store every value given by perms_from_lists() in the memory, but how do I prevent this?
Or maybe perms_from_lists() performs poorly for such large operations?
I have heard about Streams but haven’t used them, could this be possibly solved by them?
Marked As Solved
Eiji
Here goes a simplified example:
defmodule Example do
# in case where empty list is passed as an input
def sample([]), do: []
# in case first element is an empty list
def sample([[] | tail]), do: sample(tail)
# we are using list 2 times
# the first would be iterated
# the second is a copy for `sol_okay?/2` function
def sample(list) do
case sample(list, [], list) do
[] -> {:error, :not_found}
result -> result
end
end
# in case we already have result skip everything else
def sample(_list, {:ok, _result} = data, _list_copy), do: data
# data here is finished element of output list
# due to prepending the data is reversed
# which is fixed by `:lists.reverse/1`
def sample([], data, list_copy) do
solution = :lists.reverse(data)
if sol_okay?(list_copy, solution), do: {:ok, solution}, else: []
end
# when heads i.e. elements of first list ends
# all we need to do is to return an empty list
def sample([[] | _tail], _data, _list_copy), do: []
# collecting data and recursion part
def sample([[sub_head | head] | tail], data, list_copy) do
# nested recursion
new_data = sample(tail, [sub_head | data], list_copy)
if new_data == [] do
# solution not found, we are searching using tail recursion
sample([head | tail], data, list_copy)
else
# solution is already found and there is no need to return it
new_data
end
end
def sol_okay?(_puzzle, [:b, 2]), do: true
def sol_okay?(_puzzle, _solution), do: false
end
Why simplified? Well … first of all copying a full list for every processed combination is not the best idea. Perhaps you would like to store them somewhere, for example in :ets table. I just give you a simplified code which could be refactored by changing minimal amount of code and without any problem.
Also sol_okay?/2 function is naive as we have no idea what you want there and again for simplicity I made a simplest possible example i.e. pattern matching. However this function is still really useful. If you debug solution variable in 2nd clause you would find that the code is properly skipping other combinations when valid combination was found.
Also Liked
derek-zhou
I believe that for new user of Elixir (or any functional language for that matter), it is better not to learn list comprehension. Just use recursion (both body and tail), There is nothing you can do with list comprehension but not recursion.
al2o3cr
This seems like a good application for streams; they trade additional runtime complexity for memory usage. They can also be useful to skip unnecessary calculations by stopping the stream early, but that’s not applicable since you need every solution.
Here’s a version of your code above that uses Enum.filter all-at-once:
def solutions_enum(puzzle) do
puzzle
|> perms_from_lists()
# in here: all the permutations are in memory
|> Enum.filter(&sol_okay?(puzzle, &1))
end
and the same code but with streams:
def solutions_enum(puzzle) do
puzzle
|> stream_perms_from_lists()
# in here: only one permutation is in memory at a time
|> Stream.filter(&sol_okay?(puzzle, &1))
# this accumulates all the results
|> Enum.to_list()
end
The new stream_perms_from_lists/1 will involve a bunch of functions from Stream. For some inspiration, here’s a similar-but-different problem (generating permutations) solved with streams:
cmo
Hi,
I’d suggest benchmarking different methods until you find one that works. You can use Benchee for that.
You might like to avoid putting the entire function on one long line. If you give things names and keep each line relatively short it is easier to read.
Did you by chance read the documentation for Stream? At the top or explains how to swap from Enum to Stream.
Source of this code: How to generate permutations from different sets of elements? - #12 by Eiji
Popular in Questions
Other popular topics
Categories:
Sub Categories:
Forums
Popular Tags
- #ecto
- #liveview
- #troubleshooting
- #learning-elixir
- #deployment
- #library
- #erlang
- #testing
- #genserver
- #mix
- #absinthe
- #remote-other
- #otp
- #plug
- #how-to-question
- #macros
- #postgres
- #channels
- #elixirconf
- #exunit
- #discussion
- #code-sync
- #javascript
- #podcasts
- #onsite
- #dialyzer
- #docker
- #authentication
- #umbrella
- #full-time-contract
- #podcasts-by-brainlid
- #ecto-query
- #elixir-ls
- #phoenix_html
- #iex
- #blog-post
- #graphql
- #genstage
- #ai
- #websockets
- #supervisor
- #advent-of-code
- #elixirconf-us
- #distillery
- #processes
- #forms
- #api
- #metaprogramming
- #security
- #performance









