Overview
The ML Research Pipeline organizes the path from data and features to model comparison, time-based validation, rolling backtests, and execution-oriented experiments. It is a Python research prototype with a connection to LEAN. The project is about making those stages part of a coherent research workflow and examining what each stage can actually tell us.
A prediction is one output in that workflow. Evaluating it also involves its place in time, the cost of turning it into an action, and the connection between the research process and an execution environment.
My contribution
My work centers on organizing data, features, and model comparisons, along with time-split validation and trading-cost evaluation. It also includes connecting prediction and execution stages. Working across these stages keeps the evaluation process connected to the conditions under which a prediction would be used.
- Data & features
- Compare models
- Validate across time
- Backtest with costs
- Connect execution
Research flow
The sequence moves from data and feature preparation through model comparison, then into time-aware evaluation and a cost-aware backtest before the execution connection. The diagram shows how the stages relate, with separate places for assessing a prediction and considering its use in an execution experiment.
Engineering focus
The design questions concern the reliability of an experiment’s interpretation: which information belongs in each time period, whether model comparisons use a consistent evaluation setup, and how costs affect the result being examined. The handoff from a prediction to an execution experiment is another boundary to inspect, rather than assume away.
Status & scope
This is a research prototype for investigating models and execution workflows, not a promise of returns or an established profitable strategy. It is not presented as financial advice or evidence of successful live trading.