Open steffencruz opened 1 year ago
@joeylegere has a slight preference for pickling.
Next step for this would be to modify neuron
class so that it pickles the event history intermittently and then we load that pickle file into mirror neuron
as an experiment.
We can test this by running a validator on testnet. I will need support on this @ifrit98 .
We can test this by running a validator on testnet. I will need support on this @ifrit98 .
Yep. Can help out with that. Vune caught an issue we have been having attempting to use the testnet but the metagraph being empty and threw up a PR. Not sure if it's merged yet, but the fix is simple in case we run into errors.
thanks
There is interest to port a minimal version of this repo into bittensor so that we can track validators and the network in general.
It seems like the simplest way to do this is to save the forward events either by pickling to history queue or by writing the pre existing event logs to disk. There is also interest in reusing the data visualizations and analysis capabilities of this repo, and so we should focus on making the
analysis
stage of experiments in this repo stable.This will provide a direct path to creating value out of this repo and so should be carefully considered.
Estimated size of pickled object
~500 kb for an experiment with 1000 events, so expected pickled file size would be around 1MB/2000 events (or forward passes).