503cb27b99c2ed29c893ee8946d97c1792a96950,examples/ale/train_nsq_ale.py,,main,#,28

Before Change

            eval_stats["stdev"]))
    else:
        explorer = explorers.ConstantEpsilonGreedy(0.05, action_space.sample)
        experiments.train_agent_async(
            outdir=args.outdir,
            processes=args.processes,
            make_env=make_env,
            make_agent=make_agent,
            profile=args.profile,
            steps=args.steps,
            eval_n_runs=args.eval_n_runs,
            eval_interval=args.eval_interval,
            eval_explorer=explorer)

if __name__ == "__main__":
    main()

After Change

        def lr_setter(env, agent, value):
            agent.optimizer.lr = value

        lr_decay_hook = experiments.LinearInterpolationHook(
            args.steps, args.lr, 0, lr_setter)

        experiments.train_agent_async(
            outdir=args.outdir,
            processes=args.processes,
            make_env=make_env,
            make_agent=make_agent,
            profile=args.profile,
            steps=args.steps,
            eval_n_runs=args.eval_n_runs,
            eval_interval=args.eval_interval,
            eval_explorer=explorer,
            global_step_hooks=[lr_decay_hook])

if __name__ == "__main__":
    main()

Instances