Tasks

Tasks#

The tasks module provides the task-based architecture for spiking RNNs, enabling evaluation of spiking neural networks on cognitive tasks.

Core Classes#

Factory Classes#

Overview#

The spiking tasks module provides specialized task implementations for evaluating spiking neural networks. These tasks extend the rate-based task framework with spiking-specific evaluation capabilities.

Key Features:

  • Spiking-Specific Interface: Designed for spiking neural network evaluation

  • Performance Metrics: Multi-trial evaluation with detailed performance analysis

  • Visualization Support: Built-in plotting and visualization capabilities

  • Extensible Registry: Dynamic task registration for custom implementations

  • Sample Trial Types: Configurable trial types for visualization and analysis

Task Evaluation Workflow:

  1. Task Creation: Use SpikingTaskFactory.create_task() or instantiate directly

  2. Single Trial: Call evaluate_trial() for individual trial assessment

  3. Multi-Trial: Use evaluate_performance() for comprehensive evaluation

  4. Visualization: Generate plots with create_visualization()

Available Tasks:

  • Go-NoGo: Impulse control evaluation for spiking networks

  • XOR: Working memory assessment with temporal logic

  • Mante: Context-dependent decision making evaluation

Example Usage#

from spiking.tasks import SpikingTaskFactory
from spiking.eval_tasks import evaluate_task

# Create a spiking task
task = SpikingTaskFactory.create_task('go_nogo')

# Generate stimuli for specific trial types
go_stimulus, go_label = task.generate_stimulus('go')
nogo_stimulus, nogo_label = task.generate_stimulus('nogo')

# Evaluate with a trained spiking network
performance = task.evaluate_performance(spiking_rnn, n_trials=100)
print(f"Accuracy: {performance['overall_accuracy']:.2f}")

# High-level evaluation interface
performance = evaluate_task(
    task_name='go_nogo',
    model_dir='models/go-nogo/',
    n_trials=100
)

Custom Spiking Task Creation#

from spiking.tasks import AbstractSpikingTask, SpikingTaskFactory

class MyCustomSpikingTask(AbstractSpikingTask):
    def get_default_settings(self):
        return {'T': 200, 'custom_param': 1.0}

    def get_sample_trial_types(self):
        return ['type_a', 'type_b']  # For visualization

    def generate_stimulus(self, trial_type=None):
        # Generate stimulus logic
        return stimulus, label

    def evaluate_performance(self, spiking_rnn, n_trials=100):
        # Multi-trial performance evaluation
        return {'accuracy': 0.85, 'n_trials': n_trials}

# Register with factory
SpikingTaskFactory.register_task('my_custom', MyCustomSpikingTask)

# Now works with eval_tasks.py
# python -m spiking.eval_tasks --task my_custom --model_dir models/custom/

Integration with eval_tasks.py#

The tasks module is fully integrated with the evaluation system:

  • Dynamic Task Discovery: eval_tasks.py automatically supports all registered tasks

  • Generic Visualization: Uses get_sample_trial_types() for plot generation

  • CLI Support: Command-line interface adapts to new tasks automatically

  • Error Handling: Robust error handling for custom task implementations

Contents