rlevo 0.1.0

Deep Reinforcement Learning with Evolutionary Optimization
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
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245

//! End-to-end integration tests for PPG.
//!
//! Budget conventions mirror the PPO tests: default-run tests use a modest
//! 50k-step budget with lax thresholds; heavier macro-convergence and
//! reproducibility checks live behind `#[ignore]`.

use burn::backend::{Autodiff, NdArray};
use burn::module::Module;
use burn::nn::{Linear, LinearConfig};
use burn::tensor::Tensor;
use burn::tensor::activation::tanh;
use burn::tensor::backend::{AutodiffBackend, Backend};

use rand::SeedableRng;
use rand::rngs::StdRng;

use rlevo_environments::classic::cartpole::{
    CartPole, CartPoleAction, CartPoleConfig, CartPoleObservation,
};
use rlevo_environments::wrappers::TimeLimit;
use rlevo_reinforcement_learning::algorithms::ppg::policies::{
    PpgCategoricalPolicyHead, PpgCategoricalPolicyHeadConfig,
};
use rlevo_reinforcement_learning::algorithms::ppg::ppg_agent::PpgAgent;
use rlevo_reinforcement_learning::algorithms::ppg::ppg_config::PpgConfigBuilder;
use rlevo_reinforcement_learning::algorithms::ppg::train::train_discrete;
use rlevo_reinforcement_learning::algorithms::ppo::ppo_config::PpoTrainingConfigBuilder;
use rlevo_reinforcement_learning::algorithms::ppo::ppo_value::PpoValue;

#[derive(Module, Debug)]
struct ValueMlp<B: Backend> {
    fc1: Linear<B>,
    fc2: Linear<B>,
    head: Linear<B>,
}

impl<B: Backend> ValueMlp<B> {
    fn new(obs_dim: usize, hidden: usize, device: &B::Device) -> Self {
        Self {
            fc1: LinearConfig::new(obs_dim, hidden).init(device),
            fc2: LinearConfig::new(hidden, hidden).init(device),
            head: LinearConfig::new(hidden, 1).init(device),
        }
    }

    fn forward_impl(&self, obs: Tensor<B, 2>) -> Tensor<B, 1> {
        let h = tanh(self.fc1.forward(obs));
        let h = tanh(self.fc2.forward(h));
        self.head.forward(h).squeeze_dim::<1>(1)
    }
}

impl<B: AutodiffBackend> PpoValue<B, 2> for ValueMlp<B> {
    fn forward(&self, obs: Tensor<B, 2>) -> Tensor<B, 1> {
        self.forward_impl(obs)
    }
}

type Be = Autodiff<NdArray>;

fn make_cart_pole_agent(
    seed: u64,
    num_steps: usize,
    total_timesteps: usize,
    n_iteration: usize,
) -> PpgAgent<Be, PpgCategoricalPolicyHead<Be>, ValueMlp<Be>, CartPoleObservation, 1, 2> {
    let device = Default::default();
    <Be as Backend>::seed(&device, seed);

    let policy = PpgCategoricalPolicyHeadConfig {
        obs_dim: 4,
        hidden: 64,
        num_actions: 2,
    }
    .init::<Be>(&device);
    let value = ValueMlp::new(4, 64, &device);

    let ppo = PpoTrainingConfigBuilder::new()
        .num_envs(1)
        .num_steps(num_steps)
        .num_minibatches(4)
        .update_epochs(4)
        .learning_rate(2.5e-4)
        .clip_coef(0.2)
        .entropy_coef(0.01)
        .value_coef(0.5)
        .gamma(0.99)
        .gae_lambda(0.95)
        .build();
    let config = PpgConfigBuilder::new()
        .ppo(ppo)
        .n_iteration(n_iteration)
        .e_aux(6)
        .beta_clone(1.0)
        .aux_batch_size(128)
        .build();
    let total_iterations = total_timesteps / config.batch_size().max(1);
    PpgAgent::new(policy, value, config, device, total_iterations)
}

#[test]
fn ppg_cart_pole_reaches_modest_threshold() {
    // CartPole is not PPG's home turf: the auxiliary phase's distillation
    // periodically pulls the policy back, slowing convergence relative to
    // PPO. PPG's sample-efficiency wins live on Procgen-style envs with CNN
    // encoders + vectorised rollout, both deferred. The threshold here is
    // a sanity bar that PPG learns *something* beyond random (random ≈ 20).
    let seed: u64 = 42;
    let total = 50_000_usize;
    let num_steps = 128_usize;

    let mut env = TimeLimit::new(
        CartPole::with_config(CartPoleConfig {
            seed,
            ..CartPoleConfig::default()
        }),
        500,
    );
    let mut rng = StdRng::seed_from_u64(seed);
    let mut agent = make_cart_pole_agent(seed, num_steps, total, 32);
    train_discrete::<Be, _, _, _, _, CartPoleAction, _, 1, 1, 2>(
        &mut agent, &mut env, &mut rng, total, 0,
    )
    .expect("training");
    let avg = agent.stats().avg_score().unwrap_or(0.0);
    assert!(avg >= 25.0, "expected avg reward >= 25, got {avg:.2}");
}

#[test]
fn ppg_without_aux_phase_matches_ppo_baseline() {
    // Sanity check that the policy-phase update is a faithful PPO update:
    // with n_iteration set above the total iteration count the auxiliary
    // phase never fires, and PPG should match PPO's ~50k-step CartPole
    // threshold (80 in `ppo_integration.rs`).
    let seed: u64 = 42;
    let total = 50_000_usize;
    let num_steps = 128_usize;
    let mut env = TimeLimit::new(
        CartPole::with_config(CartPoleConfig {
            seed,
            ..CartPoleConfig::default()
        }),
        500,
    );
    let mut rng = StdRng::seed_from_u64(seed);
    let mut agent = make_cart_pole_agent(seed, num_steps, total, 10_000);
    train_discrete::<Be, _, _, _, _, CartPoleAction, _, 1, 1, 2>(
        &mut agent, &mut env, &mut rng, total, 0,
    )
    .expect("training");
    assert!(
        agent.last_aux_phase().is_none(),
        "aux phase should not have fired with n_iteration=10000"
    );
    let avg = agent.stats().avg_score().unwrap_or(0.0);
    assert!(
        avg >= 80.0,
        "expected avg reward >= 80 (PPO-parity), got {avg:.2}"
    );
}

#[test]
fn ppg_aux_phase_actually_runs() {
    // Use a small n_iteration so the aux phase fires within a tiny budget.
    let seed: u64 = 11;
    let num_steps = 128_usize;
    let total = 2_048_usize; // 16 iterations at num_steps=128.
    let n_iteration = 4_usize;
    let mut env = TimeLimit::new(
        CartPole::with_config(CartPoleConfig {
            seed,
            ..CartPoleConfig::default()
        }),
        500,
    );
    let mut rng = StdRng::seed_from_u64(seed);
    let mut agent = make_cart_pole_agent(seed, num_steps, total, n_iteration);
    train_discrete::<Be, _, _, _, _, CartPoleAction, _, 1, 1, 2>(
        &mut agent, &mut env, &mut rng, total, 0,
    )
    .expect("training");
    assert!(
        agent.iteration() >= n_iteration,
        "should have completed at least n_iteration={n_iteration} policy phases, got {}",
        agent.iteration()
    );
    let aux = agent
        .last_aux_phase()
        .expect("aux phase should have run at least once");
    assert!(aux.minibatches > 0);
    assert!(aux.aux_value_loss.is_finite());
    assert!(aux.policy_kl.is_finite());
}

#[test]
#[ignore = "perturbs Burn's global ndarray RNG; run with --test-threads=1"]
fn ppg_short_run_produces_finite_rewards() {
    let seed: u64 = 7;
    let total = 2_048_usize;
    let num_steps = 128_usize;
    let mut env = TimeLimit::new(
        CartPole::with_config(CartPoleConfig {
            seed,
            ..CartPoleConfig::default()
        }),
        500,
    );
    let mut rng = StdRng::seed_from_u64(seed);
    let mut agent = make_cart_pole_agent(seed, num_steps, total, 4);
    train_discrete::<Be, _, _, _, _, CartPoleAction, _, 1, 1, 2>(
        &mut agent, &mut env, &mut rng, total, 0,
    )
    .expect("training");
    for (i, m) in agent.stats().recent_history.iter().enumerate() {
        assert!(m.reward.is_finite(), "non-finite reward at episode {i}");
        assert!(
            m.policy_loss.is_finite(),
            "non-finite policy_loss at ep {i}"
        );
        assert!(m.value_loss.is_finite(), "non-finite value_loss at ep {i}");
    }
}

#[test]
#[ignore = "macro convergence; ~2-5 min on ndarray"]
fn ppg_cart_pole_reaches_475() {
    let seed: u64 = 42;
    let total = 400_000_usize;
    let num_steps = 128_usize;
    let mut env = TimeLimit::new(
        CartPole::with_config(CartPoleConfig {
            seed,
            ..CartPoleConfig::default()
        }),
        500,
    );
    let mut rng = StdRng::seed_from_u64(seed);
    let mut agent = make_cart_pole_agent(seed, num_steps, total, 32);
    train_discrete::<Be, _, _, _, _, CartPoleAction, _, 1, 1, 2>(
        &mut agent, &mut env, &mut rng, total, 0,
    )
    .expect("training");
    let avg = agent.stats().avg_score().unwrap_or(0.0);
    assert!(avg >= 475.0, "expected avg reward >= 475, got {avg:.2}");
}