FAQ and Troubleshooting

Objectives

Resolve common issues
Understand algorithm behavior
Get help when needed

Installation Issues

ModuleNotFoundError: cec2013 or mmo

Problem: Python can’t find the CEC2013 or mmo module.

Solution: Use the activation wrapper script instead of directly activating the venv:

# Wrong (doesn't set PYTHONPATH)
source venv/bin/activate

# Correct (sets PYTHONPATH automatically)
source activate-mmo.sh

The activate-mmo.sh script is created by setup.sh and automatically configures PYTHONPATH to include both the project root and the CEC2013 benchmarks directory.

ImportError: numpy.distutils

Problem: Old numpy compatibility issue.

Solution: Upgrade numpy:

pip install --upgrade numpy

Algorithm Issues

Algorithm finds few/no solutions

Possible causes:

Budget too low
Function has very narrow basins
Population size too small

Solutions:

Increase budget (try 2-5x)
Decrease CMA-ES initial sigma
Increase GA population size

Solutions are duplicates

Problem: Multiple solutions at the same location.

Cause: Clustering threshold too small or CMA-ES converging to same point from multiple seeds.

Solution: Increase solution uniqueness threshold in post-processing:

from scipy.spatial.distance import pdist, squareform

def remove_duplicates(solutions, threshold=1e-4):
    """Remove duplicate solutions."""
    if len(solutions) <= 1:
        return solutions

    distances = squareform(pdist(solutions[:, :-1]))
    keep = [0]
    for i in range(1, len(solutions)):
        if np.min(distances[i, keep]) > threshold:
            keep.append(i)
    return solutions[keep]

Algorithm is slow

Possible causes:

Function evaluation is expensive (not CEC2013 benchmarks)
Too many CMA-ES iterations
Large population size

Solutions:

For expensive custom functions: reduce evaluation cost or budget
Reduce CMA-ES max iterations
Use a surrogate model for expensive functions

VM Issues

Connection refused

Cause: VM not running or firewall blocking.

Solutions:

Check VM status in NAIC portal
Verify your IP is whitelisted
Try: ping <VM_IP>

Permission denied (publickey)

Cause: SSH key issue.

Solutions:

# Fix key permissions
chmod 600 /path/to/your-key.pem

# Verify key path is correct
ls -la /path/to/your-key.pem

Host key verification failed

Cause: VM IP changed (common with NAIC Orchestrator).

Solution:

ssh-keygen -R <old_VM_IP>

Jupyter not accessible

Causes:

SSH tunnel not running
Jupyter not started
Wrong port

Solutions:

Start tunnel: ssh -N -L 8888:localhost:8888 -i key.pem ubuntu@<VM_IP>
Start Jupyter: jupyter lab --no-browser --ip=127.0.0.1 --port=8888
Use alternative port if 8888 is busy

Performance Questions

What budget should I use?

Guidelines based on CEC2013:

Dimension	Optima	Recommended Budget
1-2D	< 10	50,000
2D	10-50	200,000
2-10D	> 50	400,000
10-20D	any	400,000+

Getting Help

Resources

NAIC Support: support@naic.no
VM Workflows Guide: https://training.pages.sigma2.no/tutorials/naic-cloud-vm-workflows/
NAIC Portal: https://orchestrator.naic.no

Reporting Issues

When reporting issues, include:

Error message (full traceback)
Python version: python --version
Package versions: pip list
Steps to reproduce

Algorithm Reference

For algorithm details, see the research paper:

Johannsen et al. (2022). A scalable, hybrid genetic algorithm for continuous multimodal optimization in moderate dimensions. Nordic Machine Intelligence.

Keypoints

Most import errors are solved by using source activate-mmo.sh (not source venv/bin/activate)
Increase budget for difficult functions with many optima
VM connection issues often stem from IP changes or key permissions
Contact support@naic.no for NAIC-specific issues