Sportstensor: The future of sports prediction algorithms

• Introduction
• Why is this important?
• Miner and Validator Functionality
  • Miner
  • Validator
• Running Miners and Validators
  • Running a Miner
  • Running a Validator
• Community
• License

Introduction

Welcome to Sportstensor—the convergence of cutting-edge technology and sports data analytics. We are pioneering unprecedented innovation in sports prediction algorithms, powered by the Bittensor network.

The Sportstensor subnet is designed to incentivize the discovery of competitive advantages such as 'edge' over closing market odds, enabling top miners within the network to establish machine-driven dominance across the sports prediction landscape.

Why is this important?

• Closing odds represent the pinnacle of market efficiency, determined by thousands of advanced machine learning algorithms.
• Our subnet fosters the development of true machine intelligence by outperforming competing algorithms in a highly competitive AI-versus-AI environment.
• Even with sophisticated models, bettors struggle to be profitable, as bookmakers frequently impose strict limits on consistent winners.
• There is substantial demand for high-performing predictive AI from betting operators, financial firms, syndicates, and algorithmic traders seeking more accurate models.
• We attract top AI and machine learning talent from diverse industries, encouraging them to revolutionize sports prediction markets.
• By decentralizing the creation and improvement of predictive models, we reduce reliance on any single entity or algorithm, enhancing resilience and driving innovation in the sports prediction market.

Miner and Validator Functionality

Miner

• Receives requests from the Validator containing specific information such as team names and match details.
• Accesses historical data and current statistics relevant to the teams involved in the query from sports databases.
• Utilizes trained machine learning models to analyze the data and predict the team they think will win and the probability.
• Returns the prediction back to the Validator for confirmation and further action.

Miners must return two key pieces of information when responding to prediction requests:

1. probabilityChoice: The predicted outcome (Home team, Away team, Draw).
2. probability: The probability for that outcome, as a float between 0 and 1. e.g. 0.6 represents 60% probability.

Caution

Miners who fail to respond or provide incorrect responses will be penalized.

Validator

• Match Syncing: The validator operates in an endless loop, syncing match data every 30 minutes. This includes checking upcoming, in-progress, and completed games.
• League Commitment: Validators send requests every 15 minutes to acquire the active league a miner is committed to. Miners are required to respond with predictions for all matches in their committed league or receive a penalty. Miners who fail to commit to a league will be incrementally penalized until committing or deregistering.
• Match Prediction Requests: Prediction requests are sent out at specific time intervals (24 hours, 12 hours, 4 hours, and 10 minutes before a match). Miners are penalized for non-responses.
• Closing Edge Scoring: After match completion, the validator calculates the closing edge scores for each prediction and updates the local database to be used later in the scoring and weights logic.
• Prediction Cleanup: Non-registered miners and outdated predictions are regularly cleaned from the system to ensure only valid data is kept.

Scoring and Weights

• Incentives and scores are calculated every 20 minutes in a background thread.
• Each active league is iterated through to calculate scores for that league.
• During each league iteration, every miner is scored for their prediction accuracy.
• The max number of predictions included for a miner per league is determined by the league’s ROLLING_PREDICTION_THRESHOLD_BY_LEAGUE multiplied by 2.
  • This creates a constantly rolling forward set of predictions per miner per league, so older predictions outside these thresholds will not be scored.
• Incentive scores are calculated through a series of complex algorithms. Please see our whitepaper for more details. Also analyze vali_utils/scoring_utils.py.
• After all active leagues have been scored, league-specific scoring percentages are applied.
• Final scores are aggregated and logged for weight setting.
• Validators set the miners' weights on the chain based on these scores.

Running Miners and Validators

Running a Miner

Requirements

• Python 3.8+
• Pip
• CPU
• If running on runpod, runpod/pytorch:2.2.1-py3.10-cuda12.1.1-devel-ubuntu22.04 is a good base template.

Setup

1. To start, clone the repository and cd to it:

git clone https://github.com/sportstensor/sportstensor/
cd sportstensor

2. Install pm2 if you don't already have it: pm2.io.
3. Next, install the Sportstensor package: pip install -e .
4. Copy the example environment file for miner configuration:

cp neurons/example.miner.env neurons/miner.env

5. Update miner.env with your league commitment. Ensure that your commitment reflects the league you are participating in. This will enable the system to send you predictions for the appropriate matches.
6. Override or replace our base model found at st/models/base.py or tap into the parent controller st/sport_prediction_model.py to integrate your own model.

League Committments Return Format

When responding to league commitment requests, miners need to provide the active league they are submitting predictions for. The miner.env file will allow miners to define their league commitment without having to restart their miner.

# NFL, MLB, NBA, etc. -- check common/data.py for all available leagues
LEAGUE_COMMITMENTS=NFL

Prediction Return Format

When responding to prediction requests, miners need to provide two key pieces of information:

1. probabilityChoice: The predicted outcome (Home team, Away team, Draw).
2. probability: The probability for that outcome, as a float between 0 and 1. e.g. 0.6 represents 60% probability.

Failure to respond or incorrectly formatted responses will result in penalties as described in the scoring and incentive mechanisms.

Start Process: Testnet

pm2 start neurons/miner.py --name sportstensor-miner -- \
    --netuid 172 \
    --subtensor.network test \
    --wallet.name {wallet} \
    --wallet.hotkey {hotkey} \
    --axon.port {port} \ # Do not use ports 8091, 9933 and 9944
    --axon.external_ip {ip} \
    --logging.trace \
    --blacklist.validator_min_stake 0

Start Process: Mainnet

pm2 start neurons/miner.py --name sportstensor-miner -- \
    --netuid 41 \
    --wallet.name {wallet} \
    --wallet.hotkey {hotkey} \
    --axon.port {port} \ # Do not use ports 8091, 9933 and 9944
    --axon.external_ip {ip} \
    --logging.trace \
    --blacklist.force_validator_permit

Note

Make sure your port is open and firewall is configured to allow port access

Running a Validator

Requirements

• Python 3.8+
• Pip
• CPU
• If running on runpod, runpod/pytorch:2.2.1-py3.10-cuda12.1.1-devel-ubuntu22.04 is a good base template.

Importance of the Database

The subnet creates a sqlite database in the root directory titled SportsTensorEdge.db. This database is extremely important to the functionality of the subnet as it holds historical prediction data for all miners which are used in the scoring mechanism.

Caution

If you need to reinstall or migrate the subnet code, please be very careful and bring along the database with you!

Please reach out to the subnet team for any questions or assistance.

Weights & Biases

It is recommended to utilize W&B. Set environment variable with export WANDB_API_KEY=<your API key>. Alternatively, you can disable W&B with --wandb.off

Setup

1. To start, clone the repository and cd to it:

git clone https://github.com/sportstensor/sportstensor/
cd sportstensor

2. Install pm2 if you don't already have it: pm2.io.
3. Next, install the Sportstensor package: pip install -e .

Run auto-updating validator with PM2 (recommended)

pm2 start vali_auto_update.sh --name Sportstensor-validator -- \
    --netuid 41 \
    --wallet.name {wallet} \
    --wallet.hotkey {hotkey} \
    --axon.port {port} \
    --logging.trace

Note

You might need to adjust "python" to "python3" within the vali_auto_update.sh depending on your preferred system python.

Additionally, validators can use the flag --neuron.batch_size X to set a different batch size for sending requests to miners.

Run basic validator with PM2

pm2 start neurons/validator.py --name Sportstensor-validator -- \
    --netuid {netuid} \
    --wallet.name {wallet} \
    --wallet.hotkey {hotkey} \
    --axon.port {port} \
    --logging.trace

Environments

Network	Netuid
Mainnet	41
Testnet	172

Community

Join the vibrant Bittensor community and find our channel #פ • sporτsτensor • 41 on Discord.

License

The Sportstensor subnet is released under the MIT License.

---