evograd-diff 0.1.0__py3-none-any.whl

evograd/__init__.py

@@ -0,0 +1,67 @@
+"""
+EvoGrad: Metaheuristics in a Differentiable Wonderland.
+
+EvoGrad is a PyTorch-based framework for differentiable evolutionary
+computation and swarm intelligence. It provides:
+
+- **Differentiable Algorithms**: GA, DE, PSO, CMA-ES with gradient flow
+- **Classical Algorithms**: GPU-accelerated versions without backpropagation
+- **Modular Operators**: Pluggable selection, crossover, mutation operators
+- **Flexible API**: pymoo-inspired interface with dependency injection
+
+Quick Start
+-----------
+>>> from evograd.core.problem import Problem
+>>> from evograd.core.minimize import minimize
+>>> from evograd.core.termination import MaxEvaluations
+>>> from evograd.algorithms import GA
+>>> from evograd.operators import TournamentSelection, SBX, PolynomialMutation
+>>>
+>>> # Define a problem
+>>> problem = Problem(
+...     objective=lambda x: (x ** 2).sum(dim=-1),  # Sphere function
+...     n_var=10,
+...     xl=-5.0,
+...     xu=5.0,
+... )
+>>>
+>>> # Create algorithm with custom operators
+>>> algorithm = GA(
+...     pop_size=100,
+...     selection=TournamentSelection(tournament_size=3),
+...     crossover=SBX(eta=15, prob=0.9),
+...     mutation=PolynomialMutation(eta=20),
+...     adaptive=True,        # Enable gradient-based hyperparameter learning
+...     differentiable=True,  # Enable gradient-based population learning
+... )
+>>>
+>>> # Run optimization
+>>> result = minimize(problem, algorithm, termination=MaxEvaluations(10000), seed=42)
+>>> print(f"Best fitness: {result.best_fitness:.6f}")
+
+Architecture
+------------
+evograd/
+├── core/           # Problem, Algorithm base, Result, Termination, minimize/maximize
+├── algorithms/     # GA, DE, PSO, CMA-ES implementations
+├── operators/      # Selection, Crossover, Mutation, Repair
+├── benchmarks/     # Classic functions, CEC2017
+└── utils/          # Device, Callbacks, Duplicates
+
+License
+-------
+MIT License - See LICENSE file for details.
+
+Authors
+-------
+Andrea Tangherloni <andrea.tangherloni@unibocconi.it>
+"""
+
+__version__ = "0.1.0"
+__author__ = "Andrea Tangherloni"
+
+__all__ = [
+    # Version info
+    "__version__",
+    "__author__",
+    ]

evograd/algorithms/__init__.py

@@ -0,0 +1,138 @@
+"""
+EvoGrad Algorithms Module.
+
+This module provides implementations of population-based optimisation
+algorithms that leverage the EvoGrad framework for both classical and
+differentiable operation.
+
+Available Algorithms:
+    - GA: Genetic Algorithm with SBX crossover and polynomial mutation
+    - DE: Differential Evolution with multiple mutation strategies
+    - SHADE: Success-History based Adaptive DE
+    - LSHADE: SHADE with Linear Population Size Reduction
+    - PSO: Particle Swarm Optimisation
+    - CMAES: Covariance Matrix Adaptation Evolution Strategy
+
+All algorithms inherit from the base Algorithm class and follow
+the dependency injection pattern for operators.
+
+Example:
+    >>> from evograd.algorithms import GA, DE, SHADE, PSO, CMAES
+    >>> from evograd.core import Problem, minimize
+    >>> 
+    >>> problem = Problem(
+    ...     objective=lambda x: (x**2).sum(dim=-1),
+    ...     n_var=30,
+    ...     xl=-100.0,
+    ...     xu=100.0,
+    ... )
+    >>> 
+    >>> # Genetic Algorithm
+    >>> ga = GA(pop_size=100, differentiable=True)
+    >>> result = minimize(problem, ga, max_evals=10000)
+    >>> 
+    >>> # Differential Evolution
+    >>> de = DE(pop_size=100, variant="DE/rand/1/bin", adaptive=True)
+    >>> result = minimize(problem, de, max_evals=10000)
+    >>> 
+    >>> # SHADE (Self-Adaptive DE)
+    >>> shade = SHADE(pop_size=100, memory_size=100)
+    >>> result = minimize(problem, shade, max_evals=10000)
+    >>> 
+    >>> # L-SHADE (SHADE with population reduction)
+    >>> lshade = LSHADE(pop_size_init=18*30, pop_size_min=4)
+    >>> result = minimize(problem, lshade, max_evals=10000)
+    >>> 
+    >>> # Particle Swarm Optimisation
+    >>> pso = PSO(pop_size=100, adaptive=True)
+    >>> result = minimize(problem, pso, max_evals=10000)
+    >>> 
+    >>> # CMA-ES
+    >>> cmaes = CMAES(pop_size=50, adaptive=True)
+    >>> result = minimize(problem, cmaes, max_evals=10000)
+"""
+
+# Genetic Algorithm
+from evograd.algorithms.ga import (
+    GA,
+    ga_default,
+    ga_steady_state,
+    ga_comma)
+
+# Differential Evolution
+from .de import (
+    DE,
+    DEVariant,
+    de_default,
+    de_rand_1_bin,
+    de_best_1_bin,
+    de_current_to_best_1_bin,
+)
+
+# SHADE and L-SHADE
+from .shade import (
+    SHADE,
+    LSHADE,
+    SHADEMemory,
+    shade_default,
+    shade_adaptive,
+    lshade_default,
+    lshade_adaptive,
+)
+
+# Particle Swarm Optimisation
+from .pso import (
+    PSO,
+    pso_default,
+    pso_constriction)
+
+# CMA-ES
+from .cmaes import (
+    CMAES,
+    cmaes_default,
+    cmaes_small,
+    cmaes_large,
+    cmaes_adaptive,
+    cmaes_ipop,
+    cmaes_bipop,
+)
+
+
+__all__ = [
+    # GAs
+    "GA",
+    "ga_default",
+    "ga_steady_state",
+    "ga_comma",
+    
+    # DE
+    "DE",
+    "DEVariant",
+    "de_default",
+    "de_rand_1_bin",
+    "de_best_1_bin",
+    "de_current_to_best_1_bin",
+    
+    # SHADE family
+    "SHADE",
+    "LSHADE",
+    "SHADEMemory",
+    "shade_default",
+    "shade_adaptive",
+    "lshade_default",
+    "lshade_adaptive",
+    
+    # PSO
+    "PSO",
+    "pso_default",
+    "pso_constriction",
+    
+    # CMA-ES
+    "CMAES",
+    "cmaes_default",
+    "cmaes_small",
+    "cmaes_large",
+    "cmaes_adaptive",
+    "cmaes_ipop",
+    "cmaes_bipop",
+]