AttributeError: Can't get attribute 'Individual' on <module 'deap.creator' about enas-pdm HOT 4 CLOSED

Thanks for your answers and sharing. Your patient and detailed reply is very helpful to me. Thank you again sincerely!

from enas-pdm.

Hi, 943fansi. Thanks for reaching out.
To find the reason, I've tried to run the code, but i could not get the same error. So, please understand that i cannot provide you an exact answer to address the problem.

Based on the discussion in the forum you shared, i guess that the error likely comes from the code related to the multiprocessing via pathos. I used a pathos multiprocessing in my code, and currently the variable 'jobs' is set to '2' for that.
So, i suggest you to set this to '1' to deactivate the multiprocessing or directly remove the pathos related codes in 'evolutionary_algorithm.py'.

If the answer would not be helpful and the problem persists, please let me know.

from enas-pdm.

Hi, 943fansi. Thanks for reaching out. To find the reason, I've tried to run the code, but i could not get the same error. So, please understand that i cannot provide you an exact answer to address the problem.

Based on the discussion in the forum you shared, i guess that the error likely comes from the code related to the multiprocessing via pathos. I used a pathos multiprocessing in my code, and currently the variable 'jobs' is set to '2' for that. So, i suggest you to set this to '1' to deactivate the multiprocessing or directly remove the pathos related codes in 'evolutionary_algorithm.py'.

If the answer would not be helpful and the problem persists, please let me know.

Thank you for your advice.Your suggestion is very useful, it does work as you guess when 'jobs' is set to '1'.

Another problem is something about gpu on training.
【Could not create cudnn handle: CUDNN_STATUS_INTERNAL_ERROR】
Could you tell me the Equipment you used for this code?
I notice that you set 5014 in the code,but my GPU is RTX2060 has only 6GB.I guess the error occurs because of Insufficient memory.So I change 5014 to 3072,Will this change affect the experimental results？

gpus = tf.config.experimental.list_physical_devices('GPU')
if gpus:
try:

            # Currently, memory growth needs to be the same across GPUs
            for gpu in gpus:
                # tf.config.experimental.set_memory_growth(gpu, True)
                tf.config.experimental.set_virtual_device_configuration(gpu, [
                    tf.config.experimental.VirtualDeviceConfiguration(memory_limit=5014)])
            logical_gpus = tf.config.experimental.list_logical_devices('GPU')
            # print(len(gpus), "Physical GPUs,", len(logical_gpus), "Logical GPUs")
        except RuntimeError as e:
            # Memory growth must be set before GPUs have been initialized
            print(e)

from enas-pdm.

I'm happy to hear that you could handle the issue with that.
The GPU i used is NVIDIA TITAN Xp which has 12GB. I didn't manually set memory_limit to '5014', and do not know how does it affect to our computation.
But, if you can run the code with the limit of 3072 and without changing any tunable parameters, then you can obtain an optimized network as an output ( the discovered and selected architecture can be slightly different depending on the computational resources you used because the fitness of each individual can be different too).

You can somewhat mitigate the memory issue with changing the batch size, but it may affect the training time and the final results.

One last thing want to share is that i used a TF determinism which enables to have reproducible results from GPU computation. But, in certain case, it can cause ineffective memory use and slow down the training. So, i recommend you to remove "os.environ['TF_DETERMINISTIC_OPS'] = '1'" any of the files in the repository unless you have to get an strictly reproducible results.

from enas-pdm.