Trainable dendrite pool

[steffencruz]

In the neuron source code there is a backward call to the dendrite pool

        # Pass rewards backward for potential PPO.
        if train_network:
            self.dendrite_pool.backward( 
                forward_calls = forward_calls,
                rewards = rewards,
                timeout = timeout
            )
            bittensor.logging.trace( 'Applied backward to network.' )

What is this doing and how can we include this? @joeylegere

Possible solution

Make use of fitness values in DummyDendritePool #68 and adjust uid-specific fitness based on rewards in the backward call

[steffencruz]

@joeylegere Interesting question is: "Does backprop into the network have a beneficial effect on the network or just the validator?"

[steffencruz]

This requires some thought and we're not convinced that it's valuable to us.