Task Scheduling with ML, DL and DRL

Learned schedulers are usually reported against their own baselines, which makes it hard to tell whether they beat first-come-first-served by a margin worth the complexity.

A simulated multi-tasking environment runs FCFS, shortest job first and shortest remaining time first alongside machine learning, deep learning and deep reinforcement learning approaches, all measured on the same average waiting time and turnaround time.

A like-for-like comparison across every paradigm on one workload.

Python, PyTorch, Jupyter