reset

Resets the neural network to its initial untrained state.

This method performs a complete engine reset, clearing all learned parameters and training history while maintaining the same architectural configuration.

Reset Operations:

1. Training History Cleanup: Clears all stored training examples

2. Counter Reset: Resets training steps and interaction counts to zero

3. Model Reinitialization: Creates new Xavier-initialized weight matrices

4. State Validation: Ensures clean initialization state

Use Cases:

• Fresh training start after poor convergence

• A/B testing with different training datasets

• Debug scenarios requiring clean state

• Model retraining from scratch

Preserved Configuration:

• Neural network architecture (4×8×3)

• Learning rate and random seed

• Coroutine scope and initialization state

• Engine configuration parameters

Performance Considerations:

• Thread Safety: Protected by model mutex for atomic reset

• Memory Cleanup: Immediate garbage collection of old training data

• Initialization Cost: Brief overhead for new weight generation

State After Reset:

• All weights and biases freshly initialized with Xavier distribution

• Training counters zeroed (steps=0, interactions=0)

• Empty training history and confidence tracking

• Engine remains in initialized state, ready for new training

See also