pygad.load() tries to run the model

[krkaufma]

Hello and thank you for building PyGAD! It is an awesome module.

I am attempting to load a model checkpoint simply to export the plots.

The issue I am running into is the model appears to try and start running when I use pygad.load() on a model checkpoint (I set the number of generations arbitrarily high). Similarly, I found it odd that I had to have the callback and fitness function defined to be able to load the model.

The code:

import pygad
from main import callback_generation
from fitness import fitness_function

path = '../results/2023_01_31_15_05/model/genetic_2023_01_31_15_05_1744.pkl'

model = pygad.load(path)

model.plot_fitness()
model.plot_new_solution_rate()
model.plot_genes(plot_type='scatter')
model.plot_genes(graph_type='boxplot')

When I run pygad.load, the loaded genetic algorithm immediately creates a population and tries to run. Because I moved this model to another machine, this ultimately fails due to a missing file.

The error message (slightly redacted): 
FileNotFoundError: [Errno 2] No such file or directory: '../results/2023_02_06_14_42/...

It would be ideal if a user could load any model checkpoint to investigate it and the solutions without needing the callback and fitness functions or the model immediately attempting to run().

[ahmedfgad]

Hi @krkaufma,

Thank you.

The pygad.load() function does run the algorithm. It is a very simple function that just loads the pickled object and return it.

def load(filename):
    try:
        with open(filename + ".pkl", 'rb') as file:
            ga_in = pickle.load(file)
    ...
    return ga_in

Regarding defining the fitness and callback functions again, we will consider finding an alternative.

[krkaufma]

Thanks for the reply @ahmedfgad.

Is there another method to load the pickled model and query data from it without running it? I did not find such a method in the docs.

What happens if I save the model after it reaches a stop criteria? Will it remember that it has already reached completion and not continue to run?

Thank you!

[ahmedfgad]

No there is no other method to load the picked object.

I just tested saving the model after and before completion and it does not happen to continue running after calling the load() function.

[ahmedfgad]

@krkaufma,

Just to let you know that a new library called cloudpickle (https://github.com/cloudpipe/cloudpickle) will be used instead of pickle. This supports pickling the fitness function and all callback functions so that you do not have to redefine them.