Website · Installation · Dashboard · Twitter · Bittensor
Table of contents
What is Bittensor?
Bittensor is a mining network, similar to Bitcoin, that includes built-in incentives designed to encourage computers to provide access to machine learning models in an efficient and censorship-resistant manner. Bittensor is comprised of Subnets, Miners, and Validators.
Explain Like I'm Five
Bittensor is an API that connects machine learning models and incentivizes correctness through the power of the blockchain.
Subnets are decentralized networks of machines that collaborate to train and serve machine learning models.
Miners run machine learning models. They send signals to the Validators.
Validators recieve trade signals from Miners. Validators ensure trades are valid, store them, and track portfolio returns.
This repository contains the code for the Proprietary Trading Network (PTN) developed by Taoshi.
PTN receives signals from quant and deep learning machine learning trading systems to deliver the world's most complete trading signals across a variety of asset classes.
🛠️ Open Source Strategy Building Techniques (In Our Taoshi Community)
🫰 Signals From a Variety of Asset Classes - Forex, Indices, Crypto
📈 Higher Payouts
📉 Lower Registration Fees
💪 Superior Cryptocurrency Infrastructure
🔎 Trading Visualization Dashboard
PTN is the most challenging & competitive network in the world. Our miners need to provide futures based signals ( long/short) that are highly efficient and effective across various markets to compete (forex, crypto, indices). The top miners are those that provide the most returns, while never exceeding certain drawdown limits.
- Miners can submit LONG, SHORT, or FLAT signal for Forex, Crypto, and Indices trade pairs into the network.
- Orders outside of market hours are ignored.
- Miners can only open 1 position per trade pair/symbol at a time.
- Miners must have a minimum of 10 closed positions within the past 30 days to be evaluated for the scoring system. If they have between 1 and 10 positions closed, they will be provided an incredibly small amount of TAO each round for the first month as they get onboarded to help avoid eliminations.
- Positions are uni-directional. Meaning, if a position starts LONG (the first order it receives is LONG), it can't flip SHORT. If you try and have it flip SHORT (using more leverage SHORT than exists LONG) it will close out the position. You'll then need to open a second position which is SHORT with the difference.
- Position leverage is bound per trade_pair. If an order would cause the position's leverage to exceed the boundary, the position leverage will be clamped.
- You can take profit on an open position using LONG and SHORT. Say you have an open LONG position with .75x leverage and you want to reduce it to a .5x leverage position to start taking profit on it. You would send in a SHORT signal of size .25x leverage to reduce the size of the position. LONG and SHORT signals can be thought of working in opposite directions in this way.
- You can explicitly close out a position by sending in a FLAT signal.
- Miners are eliminated if their portfolio return falls below certain thresholds or if they they are detected as plagiarising other miners. (more info in the "Eliminations" section).
- There is a fee per trade pair position. The fee scales with leverage. e.x a 10x leveraged position will have a 10x higher fee.
- There is a minimum registration fee of 5 TAO on the mainnet subnet.
- There is an immunity period of 9 days to help miners submit orders to become competitive with existing miners.
- Based on portfolio metrics such as omega score and total portfolio return, weights/incentive get set to reward the best miners. This is based on both open and closed positions.
With this system only the world's best traders & deep learning / quant based trading systems can compete.
In the Proprietary Trading Network, the performance of each miner's portfolio is constantly monitored to ensure competitiveness and integrity within the network. Eliminations are a crucial part of maintaining the network's quality, and they occur under two circumstances MDD, and Plagiarism.
- Daily MDD Limit: If, on a daily close (UTC), a miner's portfolio experiences more than a 5% drawdown when compared to its realized all-time high , the miner will be eliminated.
- Anytime MDD Limit: Similarly, if at any point, the miner's portfolio undergoes a 10% drawdown, the miner will be eliminated.
- Open Position Treatment: Open positions across different trade pairs are considered together for MDD calculations. For example, if multiple open positions across different pairs collectively result in a return that doesn't breach the MDD threshold, the miner remains active. Conversely, a single poor-performing open position can result in elimination.
When monitoring a miner's portfolio, validators calculate "drawdown". This drawdown value is calculated by comparing the instantaneous value of the portfolio to its all-time high realized return. All portfolios start at a value of 1x or 100%. Thus, after the first closed trade, if the portfolio goes up by 10%, the drawdown is taken as (1.1 - 1.1) / 1.1 = 0%. The portfolio has no drawdown because it has been increasingly in value only.
Another example - if the first position, open or closed, results in a portfolio drop of 6%, the miner has a drawdown of (.94 - 1) / 1 = 6%
Another example - if the first two closed positions result in a portfolio drop of 6% each, the miner has a drawdown of (.94 * .94 - 1) / 1 = 11.64%
Another example - If the first closed position results in a portfolio gain of 50% and the second position, open or closed, results in a portfolio drop of 6%, the miner has a drawdown of (1.41 - 1.5) / 1.5 = 6%
Miners who repeatedly copy another miner's trades will be eliminated. Our system analyzes the uniqueness of each submitted order. If an order is found to be a copy (plagiarized), it triggers the miner's elimination.
After elimination, miners are not immediately deregistered from the network. They will undergo a waiting period, determined by registration timelines and the network's immunity policy, before official deregistration. Upon official deregistration, the miner forfeits registration fees paid.
Take a look at the current competition Dashboard
Please see our Validator Installation guide.
Please see our Miner Installation guide.
We recommend joining our community hub via Discord to get assistance in building a trading strategy. We have partnerships with both Glassnode and LunarCrush who provide valuable data to be able to create an effective strategy. Analysis and information on how to build a deep learning ML based strategy will continue to be discussed in an open manner by team Taoshi to help guide miners to compete.
For instructions on how to contribute to Taoshi, see CONTRIBUTING.md and Taoshi's code of conduct.
Refer to the License page for information about Taoshi's licensing.
Bittensor's source code in this repository is licensed under the MIT License.