Proposal: add a weight-based trading environment with composite reward shaping

[anarlavrenov]

Hi guys,

I've built a custom Gymnasium-compatible crypto trading environment that I think could be a useful addition to FinRL's environment collection, and I'd love to hear whether it fits the project's direction.

Core idea. In short, the environment handles rebalancing internally, computes transaction costs asymmetrically for buys and sells, and clips infeasible trades against available cash and holdings.

Reward function. Composite shaping designed to balance profitability and risk-aware behavior:

1. Log-return of total asset (primary signal)
2. Period-based PnL bonus via tanh (long-horizon optimization, e.g. every 24 hours)
3. Rolling Sharpe ratio term (volatility awareness)
4. Penalties for extreme allocations, unfilled orders, and prolonged inactivity
5. Small activity bonus to discourage idle cash-holding

Observation space. Normalized cash ratio, log price-change, externally standardized technical indicators, current coin weight, and buy/sell capacities - all in stable numerical ranges.

The environment follows the Gymnasium API .

Please visit my repo https://github.com/anarlavrenov/TradeNinja to look through implementation. The description is not in English but I believe it won't be the blocker.

Please let me know guys if you have any questions