Open kongzii opened 2 months ago
Moving https://github.com/gnosis/prediction-market-agent-tooling/issues/157 into PMA, to keep in mind that some more intelligent bet sizing is desirable.
If any universal functions/classes/methods come out of implementing this, we can promote that to the PMAT repo, of course.
Currently blocked by https://github.com/gnosis/prediction-market-agent-tooling/issues/161
Another suggestion :
-> If we are defining bet_sizes in a mathematical way (e.g. based on difference between agent_p_yes
and market_p_yes
), then it's OK to have bet_amount : float
-> If we let the LLM decide on the bet amount, probably better to have it pick from a set of fixed sizes (e.g. `pick from [0.01, 0.1, 1., 10] xDAI instead of allowing for a continous range (see https://arize.com/blog-course/numeric-evals-for-llm-as-a-judge/)
We do have these betting strategies implemented in python to choose from https://github.com/gnosis/prediction-market-agent-tooling/tree/main/prediction_market_agent_tooling/tools/betting_strategies
Gabriel's original text:
Having worked a bit more in the PMA (https://github.com/gnosis/prediction-market-agent) repo, I believe we should think about having some infrastructure for the agent being able to determine the size of each bet. Currently this is the (standard) implementation for this:
One option that comes to mind is let the agent determine the amount, based on a number of factors, e.g. xDAI balance, confidence on prediction, time left for market to close, etc. Those are all parameters we can already provide the agent.