Optimize Stock Options AI Training
Budget: $30 – $250 USD
I already have a clean dataset, a working data-loading pipeline, and end-to-end code that trains my stock-options trading model, yet training still takes far too long on my single-GPU machine. My priority is to optimise the existing code for algorithmic efficiency, with most of the pain coming from the hyper-parameter tuning stage.
What I need from you is a careful refactor of the current training loop and tuning routine so that each experiment finishes faster and uses fewer resources without sacrificing predictive accuracy. If Bayesian or bandit-style search with early-stopping and pruning (Optuna, Ray Tune, HyperOpt, etc.) will help, please wire it in and demonstrate the gains. Where possible, streamline any redundant calculations inside the loss, metrics, or data-prep steps so the GPU stays fed but never idle.
Deliverables should include:
• Updated training script(s) and any helper modules
• A short README explaining the changes and the new command line flags
• A before-vs-after benchmark showing wall-clock time, GPU utilisation, and model performance on my validation split
Everything is in Python with typical libraries—NumPy, Pandas, and either PyTorch or TensorFlow (happy to share as soon as we start). Keep all external dependencies lightweight so the project still runs on my current hardware.
Once the revised code reproduces my baseline accuracy in less time, I’ll consider the task complete and release the milestone.
What I need from you is a careful refactor of the current training loop and tuning routine so that each experiment finishes faster and uses fewer resources without sacrificing predictive accuracy. If Bayesian or bandit-style search with early-stopping and pruning (Optuna, Ray Tune, HyperOpt, etc.) will help, please wire it in and demonstrate the gains. Where possible, streamline any redundant calculations inside the loss, metrics, or data-prep steps so the GPU stays fed but never idle.
Deliverables should include:
• Updated training script(s) and any helper modules
• A short README explaining the changes and the new command line flags
• A before-vs-after benchmark showing wall-clock time, GPU utilisation, and model performance on my validation split
Everything is in Python with typical libraries—NumPy, Pandas, and either PyTorch or TensorFlow (happy to share as soon as we start). Keep all external dependencies lightweight so the project still runs on my current hardware.
Once the revised code reproduces my baseline accuracy in less time, I’ll consider the task complete and release the milestone.