Add RLSSMConfig class for reinforcement learning models

[cpaniaguam]

This PR introduces the RLSSMConfig class, a specialized configuration class for Reinforcement Learning Sequential Sampling Models (RLSSM) that inherits from the newly created BaseModelConfig abstract base class (#860).

Key Features

RLSSMConfig Class

• Inherits from BaseModelConfig: Leverages the common configuration interface while adding RLSSM-specific functionality
• RLSSM-specific fields:
  • params_default: List of default parameter values (matches list_params order)
  • decision_process: Reference to the decision model (e.g., "LAN")
  • learning_process: Dictionary of learning functions for computed parameters
  • n_params, n_extra_fields: Validation counts for parameter consistency
  • decision_model, lan_model: Additional metadata for RLSSM models

Methods

• from_rlssm_dict(): Factory method to create configs from existing RLSSM dictionary structures
• validate(): Comprehensive validation ensuring parameter counts, bounds, and required fields are consistent
• get_defaults(): Returns default values and bounds for parameters (returns float instead of ParamSpec)
• to_config(): Converts RLSSMConfig to standard Config for backward compatibility
• data property: Alias for response field to support both naming conventions

Key Differences from Standard Config

• Uses params_default (list) instead of default_priors (dict) to match RLSSM parameter ordering
• Supports hierarchical structure with separate decision and learning processes
• Returns numeric defaults rather than prior distributions
• Validates consistency between parameter lists and counts

Testing

• Added 25 comprehensive tests in test_rlssm_config.py
• 100% coverage of RLSSMConfig class
• Tests cover creation, validation, defaults, conversion, and data alias functionality

Usage Example

from hssm.config import RLSSMConfig

# Create from dictionary (matching rlssm_model_config_list structure)
config = RLSSMConfig.from_rlssm_dict("rlwm", {
    "name": "rlwm",
    "description": "RLWM model with collapsing bound decision process",
    "n_params": 4,
    "n_extra_fields": 3,
    "list_params": ["alpha", "beta", "v", "a"],
    "extra_fields": ["participant_id", "trial_id", "feedback"],
    "params_default": [0.5, 0.3, 1.0, 1.5],
    "bounds": {
        "alpha": (0.0, 1.0),
        "beta": (0.0, 1.0),
        "v": (-3.0, 3.0),
        "a": (0.3, 2.5),
    },
    "decision_model": "LAN",
    "LAN": "angle",
    "data": ["rt", "response"],
    "choices": [0, 1],
})

# Validate configuration
config.validate()

# Convert to standard Config for HSSM compatibility
standard_config = config.to_config()

Related

• Depends on PR Create base config class #860 (BaseModelConfig implementation)
• Part of issue RLSSM model config #805: RLSSM model configuration refactoring

[Copilot]

Pull request overview

This PR introduces the RLSSMConfig class, a specialized configuration class for Reinforcement Learning Sequential Sampling Models (RLSSM) that inherits from the BaseModelConfig abstract base class. The implementation provides a flexible and well-tested configuration system tailored to the unique needs of RLSSM models.

Key changes:

• Adds RLSSMConfig class with RLSSM-specific fields like params_default (list format), decision_process, learning_process, and validation counts
• Implements factory method from_rlssm_dict() to create configs from existing RLSSM dictionary structures
• Provides to_config() method for backward compatibility with standard Config class
• Includes comprehensive test suite with 25 tests achieving 100% coverage

Reviewed changes

Copilot reviewed 2 out of 2 changed files in this pull request and generated 6 comments.

File	Description
src/hssm/config.py	Adds the RLSSMConfig class (lines 244-460) with validation, conversion, and property methods for RLSSM model configuration
tests/test_rlssm_config.py	Comprehensive test suite covering creation, validation, defaults retrieval, conversion to Config, and data alias functionality

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[AlexanderFengler]

I think this is on the right track, but it's not yet being tested with integration right? Left a few comments.

[Copilot]

Pull request overview

Copilot reviewed 2 out of 2 changed files in this pull request and generated 5 comments.

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[AlexanderFengler]

Few very small comments. Thanks @cpaniaguam

[AlexanderFengler]

as per comment above, I think it's semantically a bit confusing to have the RLSSM default show up here.

Also sidenote that I just notice now.

We are doing a redefinition of reponse --> ['rt', 'response'] here, which is a bit confusing.

Maybe renaming this argument to observed_data_vars makes more sense globally. (This can be either ["response"] or ["rts", "response"] or even something entirely different later).

[cpaniaguam]

Thanks for this @AlexanderFengler. I think @krishnbera was using the name data for this. However, in HSSM model configs response was being used already. In the interest of being more consistent I sort of defaulted to

response = ["rt", "response"]

for this config class. We can use observed_data_vars as you suggest. Again, we can make changes to this later as we actually start to use it to build models. Thoughts?

[krishnbera]

Looks good to me!

Revised