CODEWITHSUNDEEP
python
Omroth
Omroth

Reputation: 1119

Speeding up deepcopying many small objects in python

I have an object class:

class CellDynamicState():
    def __init__(self):
        self.ids_here = []
        self.name = "xyz"
        self.cell_static_occupied = False

And I create a list of 10,000 of them:

dynamic_cell_grid = [[CellDynamicState() for y in range(100)] for x in range(100)]

Then later I need to copy this array:

copy.deepcopy(dynamic_cell_grid)

This is slow, half a second or more.

I'm looking for suggestions for a better approach to this. If the sub-fields were all numeric, or even of known length, I would consider using numpy arrays - and I may go in that direction (using a max length for ids_here, and a dictionary for any dynamic members I might wish to assign to a cell.) But are there any other interesting ideas?

EDIT

For interest, this is how I am looking at converting to using np arrays:

class DynamicState():
    def __init__(self):
        self.n_ids_here_array = np.zeros([x_side, y_side], dtype=np.int)
        self.ids_here_array = np.zeros([x_side, y_side, consts.max_ids_at_cell], dtype=np.int)

        self.static_occupied = np.zeros([x_side, y_side], dtype=np.bool)

    def remove_id_here(self, ss, cx, cy, id):
        found = False
        for i in range(self.n_ids_here_array[cx,cy]):
            if self.ids_here_array[cx,cy,i] == id:
               self.ids_here_array[cx,cy,i] = self.ids_here_array[cx,cy,self.n_ids_here_array[cx,cy]-1]
               found = True
        
        assert found, "didn't find id"

        self.n_ids_here_array[cx,cy] -= 1

    def add_id_here(self, ss, cx, cy, id):
        assert self.n_ids_here_array[cx,cy] < consts.max_ids_at_cell - 1, "too many ids at cell"

        self.ids_here_array[cx,cy,self.n_ids_here_array[cx,cy]] = id

        self.n_ids_here_array[cx,cy] += 1

Upvotes: 0

Views: 45

Answers (1)

gerda die gandalfziege
gerda die gandalfziege

Reputation: 762

this code runs in 0,006.775900010325131 ms on my machine

import numpy as np

array = np.full((100, 100), CellDynamicState())

A nother option is this:

array = np.zeros(shape=(100, 100), dtype=CellDynamicState)

which runs in 0.004143754999859084 ms

Upvotes: 2

Related Questions