relearn 0.3.1

A Reinforcement Learning library
Documentation 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52

use env_logger::Env;
use rand::SeedableRng;
use relearn::agents::{ActorMode, Agent, BuildAgent, TabularQLearningAgentConfig};
use relearn::envs::{BuildEnv, Chain, Environment};
use relearn::logging::{ByCounter, DisplayLogger};
use relearn::simulation::{train_parallel, SimSeed, StepsIter, TrainParallelConfig};
use relearn::Prng;

fn main() {
    env_logger::Builder::from_env(Env::default().default_filter_or("info")).init();

    let env_config = Chain::default();
    // let env_config = Chain::default().with_step_limit(100);
    let agent_config = TabularQLearningAgentConfig::default();
    let training_config = TrainParallelConfig {
        num_periods: 10,
        num_threads: num_cpus::get(),
        min_worker_steps: 10_000,
    };

    let mut rng = Prng::seed_from_u64(0);
    let env = env_config.build_env(&mut rng).unwrap();
    let mut agent = agent_config.build_agent(&env, &mut rng).unwrap();

    {
        let summary = env
            .run(&agent.actor(ActorMode::Evaluation), SimSeed::Root(0), ())
            .take(10_000)
            .summarize();
        println!("Initial Stats\n{:.3}", summary);
    }

    {
        // This block ensures the logger is dropped before `Final Stats` are printed
        // so that the flushed outputs appear in-order.
        let mut logger = DisplayLogger::new(ByCounter::of_path(["agent_update", "count"], 10));
        train_parallel(
            &mut agent,
            &env,
            &training_config,
            &mut Prng::from_rng(&mut rng).unwrap(),
            &mut rng,
            &mut logger,
        );
    }

    let summary = env
        .run(&agent.actor(ActorMode::Evaluation), SimSeed::Root(0), ())
        .take(10_000)
        .summarize();
    println!("\nFinal Stats\n{:.3}", summary);
}