config.yaml.template

# ============================================================
# Plexe Configuration Template (AUTO-GENERATED)
# ============================================================
#
# This file shows all available configuration options.
# Uncomment and modify values to override defaults.
# All fields are optional - only specify what you want to change.
#
# Generated from: plexe/config.py (Config Pydantic model)
# To regenerate: python scripts/generate_config_template.py > config.yaml.template

# ------------------------------------------------------------
# Search Settings
# ------------------------------------------------------------

# Maximum hypothesis rounds in model search (each creates ~3 variants)
# Type: integer
# Default: 10
# max_search_iterations: 10

# Maximum variants to execute concurrently (controls resource usage)
# Type: integer
# Default: 3
# max_parallel_variants: 3

# ------------------------------------------------------------
# Training Settings
# ------------------------------------------------------------

# Timeout for training runs (seconds)
# Type: integer
# Default: 1800
# training_timeout: 1800

# Default epochs for neural network training (Keras, PyTorch)
# Type: integer
# Default: 10
# nn_default_epochs: 10

# Maximum epochs for neural network training (Keras, PyTorch)
# Type: integer
# Default: 50
# nn_max_epochs: 50

# Default batch size for neural network training (Keras, PyTorch)
# Type: integer
# Default: 32
# nn_default_batch_size: 32

# ------------------------------------------------------------
# LLM Settings (per agent role)
# ------------------------------------------------------------

# LLM for statistical profiling agent
# Type: string
# Default: "anthropic/claude-sonnet-4-5-20250929"
# statistical_analysis_llm: "anthropic/claude-sonnet-4-5-20250929"

# LLM for ML task analysis agent
# Type: string
# Default: "anthropic/claude-sonnet-4-5-20250929"
# ml_task_analysis_llm: "anthropic/claude-sonnet-4-5-20250929"

# LLM for metric selection agent
# Type: string
# Default: "anthropic/claude-sonnet-4-5-20250929"
# metric_selection_llm: "anthropic/claude-sonnet-4-5-20250929"

# LLM for dataset splitting agent
# Type: string
# Default: "anthropic/claude-sonnet-4-5-20250929"
# dataset_splitting_llm: "anthropic/claude-sonnet-4-5-20250929"

# LLM for baseline builder agent
# Type: string
# Default: "anthropic/claude-sonnet-4-5-20250929"
# baseline_builder_llm: "anthropic/claude-sonnet-4-5-20250929"

# LLM for feature engineering agent
# Type: string
# Default: "anthropic/claude-sonnet-4-5-20250929"
# feature_processor_llm: "anthropic/claude-sonnet-4-5-20250929"

# LLM for model definition agent
# Type: string
# Default: "anthropic/claude-sonnet-4-5-20250929"
# model_definer_llm: "anthropic/claude-sonnet-4-5-20250929"

# LLM for model evaluation agent
# Type: string
# Default: "anthropic/claude-sonnet-4-5-20250929"
# evaluation_llm: "anthropic/claude-sonnet-4-5-20250929"

# LLM for hypothesiser agent
# Type: string
# Default: "anthropic/claude-sonnet-4-5-20250929"
# hypothesiser_llm: "anthropic/claude-sonnet-4-5-20250929"

# LLM for planner agent
# Type: string
# Default: "anthropic/claude-sonnet-4-5-20250929"
# planner_llm: "anthropic/claude-sonnet-4-5-20250929"

# LLM for insight extractor agent
# Type: string
# Default: "anthropic/claude-sonnet-4-5-20250929"
# insight_extractor_llm: "anthropic/claude-sonnet-4-5-20250929"

# ------------------------------------------------------------
# Logging & Agent Settings
# ------------------------------------------------------------

# Python logging level (DEBUG, INFO, WARNING, ERROR)
# Type: string
# Default: "INFO"
# log_level: "INFO"

# Smolagents verbosity level (0=silent, 1=normal, 2=verbose)
# Type: integer
# Default: 0
# agent_verbosity_level: 0

# ------------------------------------------------------------
# OpenTelemetry Tracing
# ------------------------------------------------------------

# Enable OpenTelemetry tracing
# Type: boolean
# Default: false
# enable_otel: false

# OTLP endpoint URL (e.g., https://cloud.langfuse.com/api/public/otel)
# Type: string | null
# Default: null
# otel_endpoint: null

# Authentication headers for OTLP endpoint
# Type: object
# Default: null
# otel_headers: null

# ------------------------------------------------------------
# Evaluation Settings
# ------------------------------------------------------------

# Minimum improvement over baseline (1.1 = 10% better)
# Type: number
# Default: 1.1
# performance_threshold: 1.1

# ------------------------------------------------------------
# Sampling Settings
# ------------------------------------------------------------

# Training sample size for fast search iterations
# Type: integer
# Default: 30000
# train_sample_size: 30000

# Validation sample size for fast search iterations
# Type: integer
# Default: 10000
# val_sample_size: 10000

# ------------------------------------------------------------
# Code Generation
# ------------------------------------------------------------

# Standard library modules allowed for agent code generation
# Type: array
# Default: null
# allowed_base_imports: null

# ------------------------------------------------------------
# Model Constraints
# ------------------------------------------------------------

# Restrict to specific model types (null = all allowed)
# Type: array | null
# Default: null
# allowed_model_types: null

# ------------------------------------------------------------
# Spark Execution
# ------------------------------------------------------------

# Spark backend: 'local' (PySpark) or 'databricks'
# Type: string
# Default: "local"
# spark_mode: "local"

# Number of Spark worker threads (local mode only)
# Type: integer
# Default: 8
# spark_local_cores: 8

# Spark driver memory (local mode only)
# Type: string
# Default: "8g"
# spark_driver_memory: "8g"

# ------------------------------------------------------------
# Databricks Settings
# ------------------------------------------------------------

# Use Databricks serverless compute
# Type: boolean
# Default: false
# databricks_use_serverless: false

# Databricks cluster ID (if specified)
# Type: string | null
# Default: null
# databricks_cluster_id: null

# Databricks workspace URL
# Type: string | null
# Default: null
# databricks_host: null

# Databricks access token
# Type: string | null
# Default: null
# databricks_token: null

# Databricks config profile name
# Type: string | null
# Default: null
# databricks_profile: null

# ------------------------------------------------------------
# Dataset Format
# ------------------------------------------------------------

# CSV delimiter character
# Type: string
# Default: ","
# csv_delimiter: ","

# Whether CSV files have header row
# Type: boolean
# Default: true
# csv_header: true

# ------------------------------------------------------------
# Runtime Overrides
# ------------------------------------------------------------

# User identifier (set at runtime)
# Type: string | null
# Default: null
# user_id: null

# Experiment identifier (set at runtime)
# Type: string | null
# Default: null
# experiment_id: null

# ------------------------------------------------------------
# LiteLLM Global Settings
# ------------------------------------------------------------

# Enable SSL certificate verification for LiteLLM
# Type: boolean
# Default: true
# litellm_ssl_verify: true

# Drop unsupported parameters instead of raising errors
# Type: boolean
# Default: false
# litellm_drop_params: false

# ------------------------------------------------------------
# LiteLLM Routing
# ------------------------------------------------------------

# Per-model LiteLLM routing configuration
# Type: RoutingConfig (object)
# Default: null
#
# routing_config:
#   default:
#     api_base: null
#     headers: {}
#   providers:
#     my-proxy:
#       api_base: "https://proxy.example.com/v1"
#       headers:
#         authorization: "${PROXY_TOKEN}"
#   models:
#     anthropic/claude-sonnet-4-5-20250929: my-proxy

# ============================================================
# Example: Minimal Config for Quick Testing
# ============================================================
#
# Uncomment these for quick local testing:
#
# agent_verbosity_level: 0
# max_search_iterations: 5
# train_sample_size: 10000
# val_sample_size: 3000