Rate RNN

Rate RNN#

The rate package provides continuous-variable rate RNN implementations for training models on cognitive tasks.

Core Modules#

Module Overview#

model.py

Contains the main FR_RNN_dale class and backward-compatible task functions.

tasks.py

Task-based architecture with abstract base classes and concrete task implementations for cognitive tasks.

utils.py

Utility functions for GPU management, training helpers, and network configuration.

Quick Reference#

Main Classes:

  • FR_RNN_dale: Main rate RNN class with Dale’s principle

  • RNNConfig: Configuration dataclass for rate RNN parameters

  • AbstractTask: Base class for all cognitive tasks

  • TaskFactory: Factory for creating task instances

Task Classes:

  • GoNogoTask: Go-NoGo impulse control task

  • XORTask: Temporal XOR working memory task

  • ManteTask: Context-dependent sensory integration task

Key Functions:

  • set_gpu(): GPU device configuration

  • create_default_config(): Create default configuration

  • TaskFactory.create_task(): Create task instances

Supported Tasks:

  • Go-NoGo: Binary decision task with response inhibition

  • XOR: Temporal exclusive OR requiring working memory

  • Mante: Context-dependent sensory integration task

Contents