test_individuals.py

""" Code based on lenia tutorials and flow lenia writen in Pytorch by Etienne Guichard
    This will create a full screen image and a save button, you need to tab out to the save button
    Parameters are not saved raw, a set of smaller parameters is saved and used to reconstruct the state of the CA """
import numpy as np
import torch

import utils
from utils import *
import configs
import uuid

refresh = False
####Setup. Here is what you change####
device = "cuda:0"  # <-- set to cpu if no gpu
#cv2.namedWindow("Lenia", cv2.WND_PROP_FULLSCREEN)
#cv2.setWindowProperty("Lenia", cv2.WND_PROP_FULLSCREEN, cv2.WINDOW_FULLSCREEN)
starting_area = 50  # <-- size of random starting area
full_noise = False# <-- if starting area covers full screen or not, overides size
load_shape_from_save = False
full_screen_sim_x, full_screen_sim_y = int(150), int(150)  # <-dims, the bigger the slower
config_type = None # <-- None for random, file name for a parameter set saved on file
has_food = False  # <-- only works on flow lenia, set to true for food
s_uuid = str(uuid.uuid4())
is_flow = True# <-- set to true if using flow lenia, allows for some extra parameters
config = configs.load_saved_file(config_type)  # <-- config.load_saved_files for loading from a file, config.Name for loading know lenia creatrures, check configs for names
ca_type = Lenia_Diff_MassConserve
# <-- Lenia_Flow for flow lenia, Lenia_classic for classic lenia, Lenia_Diff for differentiable Lenia
save_path = "./base/"  # <-- change for prefered path to save parameters
shape_path = "shapes/"
loasf_shape_path = "shapes/biphase.npy"
frame = 0
save_frames = False
saved_frames_path = "saved_frames/img_"

mode = "soft"  # <-- hard for hard clip, soft for soft clip
saved_shaped = None
""" Variables for random parameter generation when not loading form file
    These variables give the range to or choice for the parameter constructor to select from
    If a range is not wanted set these to a single value """

has_food_range = [3]  # <-- number of possible channels channels if food is available
no_food_range = [1]  # <-- number of possible channels if no food is available
dt_range = [5]  # <-- posible range of for integration factor dt = 1/choice(dt_range)
kernel_range = [25,31]  # <-- range for kernel size
n_range = [3,15]  # <-- range for number of kenels
pk_choice = ["Sparse","Dense"]  # <- choice for wether sparse or dense kernels

############

###Set up Lenia. Do not change anything here #####

nca, x, saver = utils.construct_ca(s_uuid, save_path, full_screen_sim_x, full_screen_sim_y, mode, config, ca_type,
                                   has_food=has_food, is_flow=is_flow, starting_area=starting_area,
                                   full_noise=full_noise, has_food_range=has_food_range, no_food_range=no_food_range,
                                   dt_range=dt_range, kernel_range=kernel_range,n_range=n_range,pk_choice=pk_choice )

dict = saver.save_to_dict()
################### Save Button ##############################
app = App(dict, save_path, s_uuid, nca.__class__.__name__)  # need to save the c0 and c1 config

if load_shape_from_save:
    x_load = np.load(loasf_shape_path)
    x = torch.tensor(x_load, device=device)
def event(event,x,y,flags,param):
    global refresh
    if event == cv2.EVENT_MBUTTONDOWN:
        refresh = True
#### Simulation ####
while True:
    if refresh:
        refresh = False
        x =  get_starting_pattern(None,starting_area, x.shape[2], x.shape[0], x.shape[1], has_food = has_food,full=full_noise, wavelength=dict["R"])
        if load_shape_from_save:
            x_load = np.load(loasf_shape_path)
            x = torch.tensor(x_load, device=device)
    x = nca(x)
    print(x.mean())
    grid = x.clone().cpu().permute((1, 0, 2)).detach().float().numpy()

    if save_frames and (frame < 500):
        gray = cv2.cvtColor(grid, cv2.COLOR_BGR2GRAY)
        print(gray)
        # Threshold to make a mask
        mask = cv2.threshold(gray, 0.1, 255, cv2.THRESH_BINARY)[1]
        print(mask.shape)
        result = np.dstack((grid, mask))
        cv2.imwrite(saved_frames_path + str(frame)+ ".png", result*255)
        frame+=1
    key = render(grid, has_food, waitkey=1, multiplier=3)
    if key == 82:
        saved_shaped = x.clone().cpu().detach().float().numpy()
        np.save(shape_path + "biphase", saved_shaped)
    cv2.setMouseCallback('Lenia', event)