Installation

Core Installation

Clone the repository

git clone <repository-url>
cd mle_expert_coding

Install core dependencies

pip install -r requirements.txt

Configure API keys

Create .env in project root:

# Required
OPENAI_API_KEY=your-openai-api-key
GOOGLE_API_KEY=your-google-api-key

# Optional
ANTHROPIC_API_KEY=your-anthropic-api-key

Benchmark Installation

MLE-Bench
ALE-Bench

MLE-Bench provides Kaggle competition problems.Prerequisites:

Git LFS (sudo apt-get install git-lfs or brew install git-lfs)

Installation:

# Clone and install MLE-Bench
git clone https://github.com/openai/mle-bench.git
cd mle-bench
git lfs install
git lfs fetch --all
git lfs pull
pip install -e .
cd ..

# Install MLE-specific dependencies
pip install -r benchmarks/mle/requirements.txt

Verify:

PYTHONPATH=. python -m benchmarks.mle.runner --list

ALE-Bench provides AtCoder algorithmic optimization problems.Prerequisites:

Docker
libcairo2-dev (sudo apt-get install -y libcairo2-dev)

Installation:

# Clone and install ALE-Bench
git clone https://github.com/SakanaAI/ALE-Bench.git
cd ALE-Bench
pip install .
pip install ".[eval]"

# Build Docker container for evaluation
bash ./scripts/docker_build_202301.sh $(id -u) $(id -g)
cd ..

Verify:

PYTHONPATH=. python -m benchmarks.ale.runner --list

Optional: Knowledge Search

The knowledge search provides ML domain expertise using semantic search (Weaviate) and graph structure (Neo4j).

KG Graph Search (Recommended)
KG LLM Navigation (Legacy)

Uses Weaviate for embeddings + Neo4j for graph connections.

Start Weaviate

docker run -d --name weaviate \
    -p 8081:8080 -p 50051:50051 \
    -e AUTHENTICATION_ANONYMOUS_ACCESS_ENABLED=true \
    -e PERSISTENCE_DATA_PATH='/var/lib/weaviate' \
    semitechnologies/weaviate:latest

Start Neo4j

docker run -d --name neo4j \
    --restart unless-stopped \
    -p 7474:7474 -p 7687:7687 \
    -e NEO4J_AUTH=neo4j/password123 \
    neo4j:latest

Configure connection

Add to .env:

# Required for embeddings
OPENAI_API_KEY=your-openai-api-key

# Neo4j (graph structure)
NEO4J_URI=bolt://localhost:7687
NEO4J_USER=neo4j
NEO4J_PASSWORD=password123

# Weaviate (embeddings)
WEAVIATE_URL=http://localhost:8081

Index wiki pages

PYTHONPATH=. python -c "
from src.knowledge.search import KnowledgeSearchFactory, KGIndexInput
search = KnowledgeSearchFactory.create('kg_graph_search', enabled=True)
search.index(KGIndexInput(
    wiki_dir='data/wikis',
    persist_path='data/indexes/wikis.json',
))
print('Wiki pages indexed')
"

Uses Neo4j with LLM-guided graph traversal.

Start Neo4j

    docker run -d --name neo4j \
    --restart unless-stopped \
    -p 7474:7474 -p 7687:7687 \
    -e NEO4J_AUTH=neo4j/password \
    neo4j:latest

Configure connection

Add to .env:

NEO4J_URI=bolt://localhost:7687
NEO4J_USER=neo4j
NEO4J_PASSWORD=password

Load knowledge data

PYTHONPATH=. python -c "
from src.knowledge.search.kg_llm_navigation_search import KGLLMNavigationSearch
import json
search = KGLLMNavigationSearch()
with open('benchmarks/mle/data/kg_data.json') as f:
    search.index(json.load(f))
print('Knowledge graph loaded')
"

Environment Variables

Variable	Required	Default	Description
`OPENAI_API_KEY`	Yes	-	OpenAI API key (also for embeddings)
`GOOGLE_API_KEY`	Yes	-	Google API key for Gemini
`ANTHROPIC_API_KEY`	No	-	Anthropic API key for Claude
`CUDA_DEVICE`	No	`0`	GPU device for ML training
`NEO4J_URI`	No	`bolt://localhost:7687`	Neo4j connection URI
`NEO4J_USER`	No	`neo4j`	Neo4j username
`NEO4J_PASSWORD`	No	`password`	Neo4j password
`WEAVIATE_URL`	No	`http://localhost:8081`	Weaviate server URL

Verify Installation

# Check core installation
python -c "from src.execution.orchestrator import OrchestratorAgent; print('Core OK')"

# Check MLE-Bench
python -c "import mlebench; print('MLE-Bench OK')"

# Check ALE-Bench
python -c "import ale_bench; print('ALE-Bench OK')"

# Check Neo4j
python -c "from neo4j import GraphDatabase; print('Neo4j driver OK')"

# Check Weaviate
python -c "import weaviate; print('Weaviate client OK')"

# Check Knowledge Search
python -c "from src.knowledge.search import KGGraphSearch; print('Knowledge Search OK')"

Getting Started

Core Concepts

Benchmarks

Components

Deployment

Core Installation

Benchmark Installation

Optional: Knowledge Search

Environment Variables

Verify Installation

Getting Started

Core Concepts

Benchmarks

Components

Deployment

​Core Installation

​Benchmark Installation

​Optional: Knowledge Search

​Environment Variables

​Verify Installation

Core Installation

Benchmark Installation

Optional: Knowledge Search

Environment Variables

Verify Installation