Federal Data Subnet

Inspiration

Federal Data Subnet was inspired by the vast amount of open data provided by the Federal Reserve in the United States. We believe it would be valuable if there was a way to create intelligence using the data to predict economic outcomes, and provide it in an open way that fosters competition and open access.

What it does

The Federal Data Subnet is a network of Miners and Validators built on top of the bittensor blockchain. Validators use the Federal Reserve API to pull up to date information on various data sets and ask Miners to predict future values of those datasets.

Miner predictions are stored on a database and once the real data comes in, validators score Miners on how well their predictions are (MSE), while taking into account the difficulty of predicitions (predictions of bond rates 1 month from now are harder to predict than tomorrows rates)

Validators then send the miners trust/rank scores to the main subtensor blockchain to tell the chain how to distribute rewards on the subnet to incentivize ever increasing intelligent models.

How we built it

We created a script to pull various bond data from the Federal Reserve API and create a dataframe indicating what predictions validators are looking for from the Miner.

We then created a database where predictions, miner tasks, and validator request could all interact on one network. If fully implemented, this processes would be on a block chain.

Challenges we ran into

We were unable to create a functioning subnet because of the high degree of difficulty and time constraints.

Accomplishments that we're proud of

Despite not having a functional subnet, we are happy we were able to talk through our issues with the implementation and gain a clearer perspective of how to move forward! We also discussed how we would be able to add other data sets in the future to make the network more robust

What we learned

We learned that there a many aspects to consider when creating a new network.

What's next for Federal Data Subnet

We would like to continue exploring this idea, and hopefully transfer our ideas into a running testnet subnet. We would also experiment with categorizing various datasets into groups to create expert models in different areas.