Configuration

This guide covers configuring DeepFix for your environment, including server configuration, client configuration, and MLflow integration.

Server Configuration

The DeepFix server requires configuration for the LLM backend used by the analysis agents.

Environment Variables

Create a .env file in the deepfix-server directory with the following variables:

# LLM Configuration
DEEPFIX_LLM_API_KEY=sk-...                    # API key for your LLM provider
DEEPFIX_LLM_BASE_URL=https://api.your-llm.com/v1  # Base URL of the LLM API
DEEPFIX_LLM_MODEL_NAME=gpt-4o                # Model name (e.g., gpt-4o, llama-3.1)
DEEPFIX_LLM_TEMPERATURE=0.4                  # Temperature (0.0-2.0)
DEEPFIX_LLM_MAX_TOKENS=6000                  # Maximum tokens per request
DEEPFIX_LLM_CACHE=true                       # Enable caching ("true"/"false")
DEEPFIX_LLM_TRACK_USAGE=true                 # Track token usage ("true"/"false")

# Server Configuration
LIT_SERVER_API_KEY=ACCESS-TOKEN-TO-BE-DEFINED  # Optional: API key for server

LLM Provider Configuration

OpenAI / Compatible API

DEEPFIX_LLM_API_KEY=sk-...
DEEPFIX_LLM_BASE_URL=https://api.openai.com/v1
DEEPFIX_LLM_MODEL_NAME=gpt-4o
DEEPFIX_LLM_TEMPERATURE=0.4
DEEPFIX_LLM_MAX_TOKENS=6000

Anthropic Claude

DEEPFIX_LLM_API_KEY=sk-ant-...
DEEPFIX_LLM_BASE_URL=https://api.anthropic.com/v1
DEEPFIX_LLM_MODEL_NAME=claude-3-opus-20240229
DEEPFIX_LLM_TEMPERATURE=0.7
DEEPFIX_LLM_MAX_TOKENS=8000

Local LLM (Ollama, etc.)

DEEPFIX_LLM_API_KEY=not-needed
DEEPFIX_LLM_BASE_URL=http://localhost:11434/v1
DEEPFIX_LLM_MODEL_NAME=llama3
DEEPFIX_LLM_TEMPERATURE=0.7
DEEPFIX_LLM_MAX_TOKENS=4000

Server Launch Configuration

Launch the server with configuration:

# Using .env file
uv run deepfix-server launch -e deepfix-server/.env -port 8844 -host 127.0.0.1

# Using environment variables directly
export DEEPFIX_LLM_API_KEY=sk-...
export DEEPFIX_LLM_BASE_URL=https://api.openai.com/v1
uv run deepfix-server launch -port 8844 -host 0.0.0.0

Server Configuration Options

• -host: Server host (default: 127.0.0.1)
• -port: Server port (default: 8844)
• -e: Path to .env file
• --version: Show server version

Client Configuration

Configure the DeepFix SDK client for your needs.

Basic Client Configuration

from deepfix_sdk.client import DeepFixClient

# Minimal configuration
client = DeepFixClient(api_url="http://localhost:8844")

# With custom timeout
client = DeepFixClient(
    api_url="http://localhost:8844",
    timeout=120  # Request timeout in seconds
)

MLflow Configuration

Configure MLflow integration for experiment tracking:

from deepfix_sdk.client import DeepFixClient
from deepfix_sdk.config import MLflowConfig

# Create MLflow configuration
mlflow_config = MLflowConfig(
    tracking_uri="http://localhost:5000",      # MLflow tracking server URI
    experiment_name="my-ml-experiment",        # Experiment name
    run_name="baseline-run"                    # Run name
)

# Initialize client with MLflow
client = DeepFixClient(
    api_url="http://localhost:8844",
    mlflow_config=mlflow_config
)

Artifact Configuration

Configure which artifacts to load:

from deepfix_sdk.config import ArtifactConfig

artifact_config = ArtifactConfig(
    load_dataset_metadata=True,    # Load dataset metadata
    load_checks=True,               # Load deepchecks reports
    load_model_checkpoint=False,    # Load model checkpoints
    load_training=False             # Load training artifacts
)

# Use with pipelines (see API reference)

Environment Variables for Client

The client can also use environment variables:

# Optional: Set API key if needed
export DEEPFIX_API_KEY=your-api-key

# MLflow configuration
export MLFLOW_TRACKING_URI=http://localhost:5000
export MLFLOW_EXPERIMENT_NAME=my-experiment

MLflow Configuration

DeepFix integrates with MLflow for experiment tracking and artifact storage.

Setting Up MLflow Server

Option 1: Local MLflow Server

# Start MLflow tracking server
mlflow server --backend-store-uri sqlite:///mlflow.db --default-artifact-root ./mlruns --host 0.0.0.0 --port 5000

Option 2: Docker MLflow Server

# docker-compose.yml
services:
  mlflow:
    image: ghcr.io/mlflow/mlflow:latest
    ports:
      - "5000:5000"
    command: >
      mlflow server
      --backend-store-uri sqlite:///mlflow.db
      --default-artifact-root /mlruns
      --host 0.0.0.0
      --port 5000
    volumes:
      - ./mlruns:/mlruns

Option 3: Remote MLflow Server

Use a hosted MLflow server (e.g., Databricks, MLflow on AWS):

mlflow_config = MLflowConfig(
    tracking_uri="https://your-mlflow-server.com",
    experiment_name="my-experiment"
)

Configuring MLflow in Code

from deepfix_sdk.client import DeepFixClient
from deepfix_sdk.config import MLflowConfig

# Configure MLflow
mlflow_config = MLflowConfig(
    tracking_uri="http://localhost:5000",
    experiment_name="deepfix-analysis",
    run_name="dataset-diagnosis-1"
)

client = DeepFixClient(
    api_url="http://localhost:8844",
    mlflow_config=mlflow_config
)

# All operations will be tracked in MLflow
client.ingest(...)
result = client.diagnose_dataset(...)

MLflow Artifact Storage

DeepFix stores artifacts in MLflow:

• Dataset Metadata: Dataset statistics and properties
• Deepchecks Reports: Data quality check results
• Model Checkpoints: Model state and configuration
• Training Artifacts: Training metrics and logs

Access artifacts via MLflow UI or Python API:

import mlflow

# Access MLflow run
run = mlflow.get_run(run_id="...")

# List artifacts
artifacts = mlflow.list_artifacts(run_id="...")

# Download artifacts
mlflow.artifacts.download_artifacts(run_id="...", artifact_path="dataset_metadata")

Advanced Configuration

Custom Timeout Settings

# Increase timeout for large datasets
client = DeepFixClient(
    api_url="http://localhost:8844",
    timeout=300  # 5 minutes
)

Batch Size Configuration

# Adjust batch size based on available memory
client.ingest(
    dataset_name="my-dataset",
    train_data=train_dataset,
    batch_size=16  # Increase for better throughput, decrease for memory constraints
)

Server URL Configuration

# Local development
client = DeepFixClient(api_url="http://localhost:8844")

# Remote server
client = DeepFixClient(api_url="http://deepfix.example.com:8844")

# With authentication (if configured)
client = DeepFixClient(
    api_url="https://deepfix.example.com:8844",
    timeout=120
)

Configuration Best Practices

Security

1. Never commit .env files: Add .env to .gitignore
2. Use environment variables in production: Set via deployment system
3. Rotate API keys regularly: Update DEEPFIX_LLM_API_KEY periodically
4. Use secure connections: Prefer HTTPS for remote servers

Performance

1. Optimize batch sizes: Balance memory usage and throughput
2. Configure appropriate timeouts: Based on dataset size and network latency
3. Enable caching: Set DEEPFIX_LLM_CACHE=true for repeated analyses
4. Use local MLflow: For faster artifact access during development

Development

1. Use separate configs: Different .env files for dev/staging/prod
2. Version control config templates: Keep env.example updated
3. Document configuration: Note any custom settings
4. Test configurations: Verify settings before production deployment

Troubleshooting

Configuration Issues

Problem: Server fails to start

# Solution: Verify all required environment variables are set
cat deepfix-server/.env

# Check for missing variables
DEEPFIX_LLM_API_KEY=...
DEEPFIX_LLM_BASE_URL=...
DEEPFIX_LLM_MODEL_NAME=...

Problem: LLM connection errors

# Solution: Verify API key and base URL
echo $DEEPFIX_LLM_API_KEY
echo $DEEPFIX_LLM_BASE_URL

# Test LLM API connectivity
curl -H "Authorization: Bearer $DEEPFIX_LLM_API_KEY" $DEEPFIX_LLM_BASE_URL/models

Problem: MLflow connection errors

# Solution: Verify MLflow server is running
import mlflow
mlflow.set_tracking_uri("http://localhost:5000")
mlflow.list_experiments()  # Should not raise error

Configuration Validation

# Validate client configuration
client = DeepFixClient(api_url="http://localhost:8844")

# Test connection
try:
    # Attempt a simple operation
    result = client.diagnose_dataset("test-dataset")
except Exception as e:
    print(f"Configuration error: {e}")

Next Steps

• Quickstart Guide - Get started with configured DeepFix
• MLflow Integration Guide - Advanced MLflow setup
• Deployment Guide - Production configuration
• API Reference - Configuration API details