Grid search parallel

name: grid-search-parallel description: Executing a grid search across multiple hyperparameters in parallel. Use this to speed up computationally intensive tasks like DBSCAN parameter sweeps.

Parallel Grid Search

This skill covers how to efficiently iterate over a hyperparameter space using parallel processing.

Grid Definition

Define the ranges for each parameter:

• min_samples: 3 to 9 (step 1)
• epsilon: 4 to 24 (step 2)
• shape_weight: 0.9 to 1.9 (step 0.1)

Parallel Execution with Joblib

Use joblib.Parallel and joblib.delayed to distribute work.

from joblib import Parallel, delayed
import itertools

def evaluate_params(params):
    min_samples, epsilon, shape_weight = params
    # ... evaluation logic ...
    return {'F1': avg_f1, 'delta': avg_delta, ...}

# Generate parameter grid
param_grid = list(itertools.product(min_samples_range, epsilon_range, shape_weight_range))

# Execute in parallel
results = Parallel(n_jobs=-1)(delayed(evaluate_params)(p) for p in param_grid)

Data Management

• Pre-load data once and pass to the worker function or use global variables (if safe in the environment).
• Group annotations by image (file_rad) beforehand to avoid repeated filtering inside the loop.