### Basic Installation

Source: https://github.com/tianyi-stack/madevolve/blob/main/README.md

Installs the MadEvolve package with basic dependencies.

```bash
pip install -e .
```

--------------------------------

### Installation with Optional Dependencies

Source: https://github.com/tianyi-stack/madevolve/blob/main/README.md

Installs MadEvolve with specific optional features like Google Gemini support, Bayesian optimization, parsing, or similarity metrics. The '[full]' option installs all features, and '[dev]' installs development dependencies. '[all]' installs everything.

```bash
# Google Gemini support
pip install -e ".[google]"

# Bayesian optimization for inner-loop tuning
pip install -e ".[optimizer]"

# Syntax tree parsing
pip install -e ".[parsing]"

# Structural similarity metrics
pip install -e ".[similarity]"

# Full installation (all features)
pip install -e ".[full]"

# Development dependencies
pip install -e ".[dev]"

# Everything
pip install -e ".[all]"
```

--------------------------------

### Configuration File Example

Source: https://github.com/tianyi-stack/madevolve/blob/main/README.md

An example YAML configuration file for MadEvolve, detailing settings for models, population, evolution parameters, and execution.

```yaml
# config.yaml

# Model configuration
models:
  primary: gpt-4o              # Primary model for evolution
  fallback: claude-sonnet-4-20250514  # Fallback model on failure
  selection_strategy: ucb      # Options: ucb, thompson, random, fixed

# Population settings
population:
  size: 50                     # Total population size
  elite_size: 10               # Number of elite solutions to preserve
  island_count: 4              # Number of parallel islands
  migration_interval: 10       # Generations between migrations
  migration_rate: 0.1          # Fraction of population to migrate

# Evolution parameters
evolution:
  generations: 100             # Maximum generations
  mutation_rate: 0.8           # Probability of mutation
  crossover_rate: 0.2          # Probability of crossover
  strategy: hybrid             # Options: differential, holistic, hybrid

# Execution settings
executor:
  backend: native              # Options: native, slurm
  timeout: 300                 # Evaluation timeout (seconds)
  max_retries: 3               # Retries on failure
  parallel_workers: 4          # Concurrent evaluations
```

--------------------------------

### YAML Configuration Example

Source: https://github.com/tianyi-stack/madevolve/blob/main/README.md

Example of storage and inner-loop optimization settings in YAML format.

```yaml
storage:
  checkpoint_interval: 10      # Save checkpoint every N generations
  cache_enabled: true          # Enable LLM response caching
  database_path: ./cache.db    # SQLite cache location

# Inner-loop optimization (optional)
inner_loop:
  enabled: false
  method: grid_search          # Options: grid_search, bayesian, autodiff
  budget: 100                  # Evaluation budget per candidate
```

--------------------------------

### JSON Configuration Example

Source: https://github.com/tianyi-stack/madevolve/blob/main/README.md

Example of configuration settings in JSON format, including models, population, and evolution parameters.

```json
{
  "models": {
    "primary": "gpt-4o",
    "fallback": "claude-sonnet-4-20250514"
  },
  "population": {
    "size": 50,
    "elite_size": 10
  },
  "evolution": {
    "generations": 100
  }
}
```

--------------------------------

### EvolutionOrchestrator Initialization and Run

Source: https://github.com/tianyi-stack/madevolve/blob/main/README.md

Example of initializing and running the main EvolutionOrchestrator class.

```python
orchestrator = EvolutionOrchestrator(
    config: EvolutionConfig,
    results_dir: str = "./results",
    checkpoint_path: Optional[str] = None,
)
orchestrator.run()
```

--------------------------------

### Python Custom Adapter Example

Source: https://github.com/tianyi-stack/madevolve/blob/main/README.md

Example of creating custom adapters for domain-specific analysis in Python, including defining metrics and registering adapters.

```python
from madevolve.analyzer import (
    ScenarioAdapter,
    MetricsAdapter,
    PromptAdapter,
    ReportTemplateAdapter,
    BaseMetrics,
    register_adapter,
)
from dataclasses import dataclass
from typing import Any, Dict

# 1. Define your metrics structure
@dataclass
class MyMetrics(BaseMetrics):
    accuracy: float = 0.0
    loss: float = 0.0

```

--------------------------------

### Custom Scenario Adapter Implementation

Source: https://github.com/tianyi-stack/madevolve/blob/main/README.md

Example of implementing a ScenarioAdapter that bundles custom metrics, prompt, and template adapters, and how to register it.

```python
# 4. Bundle into a ScenarioAdapter
class MyScenarioAdapter(ScenarioAdapter):
    def __init__(self):
        self._metrics = MyMetricsAdapter()
        self._prompts = MyPromptAdapter()  # Your implementation
        self._templates = MyTemplateAdapter()  # Your implementation

    @property
    def metrics_adapter(self) -> MetricsAdapter:
        return self._metrics

    @property
    def prompt_adapter(self) -> PromptAdapter:
        return self._prompts

    @property
    def template_adapter(self) -> ReportTemplateAdapter:
        return self._templates

# 5. Register and use
register_adapter('my_scenario', MyScenarioAdapter)
```

--------------------------------

### Custom Metrics Adapter Implementation

Source: https://github.com/tianyi-stack/madevolve/blob/main/README.md

Example of how to implement a custom metrics adapter by inheriting from MetricsAdapter and defining methods for scenario name, description, parsing metrics, and generating comparison tables and summaries.

```python
class MyMetricsAdapter(MetricsAdapter):
    @property
    def scenario_name(self) -> str:
        return "my_scenario"

    @property
    def scenario_description(self) -> str:
        return "My custom evolution scenario"

    def parse_metrics(self, metrics_data: Dict[str, Any]) -> MyMetrics:
        return MyMetrics(
            combined_score=metrics_data.get("score", 0.0),
            accuracy=metrics_data.get("accuracy", 0.0),
            loss=metrics_data.get("loss", 0.0),
        )

    def get_metrics_comparison_table(self, baseline: BaseMetrics, best: BaseMetrics) -> str:
        return f"| Metric | Baseline | Best |\n|--------|----------|------|\n| Score | {baseline.combined_score:.4f} | {best.combined_score:.4f} |"

    def get_metrics_summary(self, metrics: BaseMetrics) -> str:
        return f"Score: {metrics.combined_score:.4f}"

    def get_key_metrics_for_llm(self, metrics: BaseMetrics) -> str:
        return f"Combined Score: {metrics.combined_score}"
```

--------------------------------

### Python API - Basic Usage

Source: https://github.com/tianyi-stack/madevolve/blob/main/README.md

Demonstrates how to configure and run the EvolutionOrchestrator using the Python API. It sets up population, model, and executor configurations, then initiates the evolution process.

```python
from madevolve import (
    EvolutionOrchestrator,
    EvolutionConfig,
    PopulationConfig,
    ModelConfig,
    ExecutorConfig,
)

# Configure the evolution
config = EvolutionConfig(
    population=PopulationConfig(
        size=50,
        elite_size=10,
        island_count=4,
    ),
    model=ModelConfig(
        primary="gpt-4o",
        fallback="claude-sonnet-4-20250514",
    ),
    executor=ExecutorConfig(
        backend="native",
        timeout=300,
    ),
    generations=100,
    mutation_rate=0.8,
    crossover_rate=0.2,
)

# Create and run the orchestrator
orchestrator = EvolutionOrchestrator(
    config=config,
    results_dir="./results",
)
orchestrator.run()
```

--------------------------------

### Using Registered Adapters

Source: https://github.com/tianyi-stack/madevolve/blob/main/README.md

Python code demonstrating how to list available adapters, retrieve a specific adapter, and use it with DataExtractor and ReportGenerator to extract data and generate reports.

```python
from madevolve.analyzer import (
    get_adapter,
    list_adapters,
    DataExtractor,
    ReportGenerator,
)

# List available adapters
print(list_adapters())

# Get adapter and generate report
adapter = get_adapter('my_scenario')
extractor = DataExtractor(adapter)
data = extractor.extract_evolution_data("/path/to/results")

generator = ReportGenerator(adapter)
report = generator.generate_full_report(data, output_path="report.md")
```

--------------------------------

### Command Line Interface - Run Evolution

Source: https://github.com/tianyi-stack/madevolve/blob/main/README.md

Runs the evolution process using a specified configuration file and output directory. The '--resume' flag allows resuming from a checkpoint, and '--quiet' suppresses the banner.

```bash
# Run evolution with a configuration file
madevolve run -c config.yaml -o ./results

# Resume from a checkpoint
madevolve run -c config.yaml -o ./results --resume checkpoint.pkl

# Run without banner
madevolve run -c config.yaml -o ./results --quiet

# Show version
madevolve version
```

--------------------------------

### Environment Variables for API Keys

Source: https://github.com/tianyi-stack/madevolve/blob/main/README.md

Sets environment variables for LLM provider API keys. Replace 'sk-...' with actual keys.

```bash
export OPENAI_API_KEY="sk-..."
export ANTHROPIC_API_KEY="sk-ant-..."
export GOOGLE_API_KEY="..."
export DEEPSEEK_API_KEY="..."
```

--------------------------------

### ExecutorConfig Dataclass

Source: https://github.com/tianyi-stack/madevolve/blob/main/README.md

Definition of the dataclass for execution backend settings.

```python
@dataclass
class ExecutorConfig:
    backend: str = "native"  # "native" or "slurm"
    timeout: int = 300
    max_retries: int = 3
    parallel_workers: int = 4
```

--------------------------------

### EvolutionConfig Dataclass

Source: https://github.com/tianyi-stack/madevolve/blob/main/README.md

Definition of the top-level configuration dataclass for evolution experiments.

```python
@dataclass
class EvolutionConfig:
    population: PopulationConfig
    model: ModelConfig
    executor: ExecutorConfig
    storage: StorageConfig
    generations: int = 100
    mutation_rate: float = 0.8
    crossover_rate: float = 0.2
```

--------------------------------

### ModelConfig Dataclass

Source: https://github.com/tianyi-stack/madevolve/blob/main/README.md

Definition of the dataclass for LLM provider settings.

```python
@dataclass
class ModelConfig:
    primary: str = "gpt-4o"
    fallback: str = "claude-sonnet-4-20250514"
    selection_strategy: str = "ucb"
    max_tokens: int = 4096
    temperature: float = 0.7
```

--------------------------------

### PopulationConfig Dataclass

Source: https://github.com/tianyi-stack/madevolve/blob/main/README.md

Definition of the dataclass for population and island model settings.

```python
@dataclass
class PopulationConfig:
    size: int = 50
    elite_size: int = 10
    island_count: int = 4
    migration_interval: int = 10
    migration_rate: float = 0.1
```

=== COMPLETE CONTENT === This response contains all available snippets from this library. No additional content exists. Do not make further requests.