claude (dqn_rs trainer simplifié, compilation still fails)

bot/Cargo.toml

@@ -13,6 +13,10 @@ path = "src/bin/train_dqn.rs"
 name = "train_burn_dqn"
 path = "src/bin/train_burn_dqn.rs"
 
+[[bin]]
+name = "simple_burn_train"
+path = "src/bin/simple_burn_train.rs"
+
 [dependencies]
 pretty_assertions = "1.4.0"
 serde = { version = "1.0", features = ["derive"] }

bot/src/bin/simple_burn_train.rs

@@ -0,0 +1,83 @@
+use bot::strategy::burn_dqn::{BurnDqnAgent, BurnDqnConfig, Experience};
+use rand::Rng;
+
+fn main() -> Result<(), Box<dyn std::error::Error>> {
+    env_logger::init();
+
+    println!("Entraînement DQN simplifié avec Burn");
+
+    // Configuration DQN simple
+    let config = BurnDqnConfig {
+        state_size: 10,
+        action_size: 4,
+        hidden_size: 64,
+        learning_rate: 0.001,
+        gamma: 0.99,
+        epsilon: 1.0,
+        epsilon_decay: 0.995,
+        epsilon_min: 0.01,
+        replay_buffer_size: 1000,
+        batch_size: 16,
+        target_update_freq: 50,
+    };
+
+    let mut agent = BurnDqnAgent::new(config);
+    let mut rng = rand::thread_rng();
+
+    println!("Début de l'entraînement simple...");
+
+    for episode in 1..=100 {
+        let mut total_reward = 0.0;
+        
+        for step in 1..=50 {
+            // État aléatoire simple
+            let state: Vec<f32> = (0..10).map(|_| rng.gen::<f32>()).collect();
+            
+            // Actions valides (toutes les actions pour simplifier)
+            let valid_actions: Vec<usize> = vec![0, 1, 2, 3];
+            
+            // Sélectionner une action
+            let action = agent.select_action(&state, &valid_actions);
+            
+            // Récompense simulée
+            let reward = rng.gen::<f32>() - 0.5; // Récompense entre -0.5 et 0.5
+            
+            // État suivant aléatoire
+            let next_state: Vec<f32> = (0..10).map(|_| rng.gen::<f32>()).collect();
+            
+            // Fin d'épisode aléatoire
+            let done = step >= 50 || rng.gen::<f32>() < 0.1;
+            
+            // Ajouter l'expérience
+            let experience = Experience {
+                state: state.clone(),
+                action,
+                reward,
+                next_state: if done { None } else { Some(next_state) },
+                done,
+            };
+            agent.add_experience(experience);
+            
+            // Entraîner
+            if let Some(loss) = agent.train_step() {
+                if step % 25 == 0 {
+                    println!("Episode {}, Step {}, Loss: {:.4}, Epsilon: {:.3}", 
+                             episode, step, loss, agent.get_epsilon());
+                }
+            }
+            
+            total_reward += reward;
+            
+            if done {
+                break;
+            }
+        }
+        
+        if episode % 10 == 0 {
+            println!("Episode {} terminé. Récompense totale: {:.2}", episode, total_reward);
+        }
+    }
+
+    println!("Entraînement terminé !");
+    Ok(())
+}

bot/src/strategy/burn_dqn.rs

@@ -117,9 +117,7 @@ impl BurnDqnAgent {
             &device,
         );
         
-        let optimizer = AdamConfig::new()
+        let optimizer = AdamConfig::new().init();
-            .with_learning_rate(config.learning_rate)
-            .init();
 
         Self {
             config: config.clone(),
@@ -147,10 +145,9 @@ impl BurnDqnAgent {
         }
 
         // Exploitation : choisir la meilleure action selon le Q-network
-        let state_tensor = Tensor::<MyBackend, 2>::from_data(
+        // Utiliser from_floats avec un vecteur 2D pour Burn 0.17
-            burn::tensor::Data::new(state.to_vec(), burn::tensor::Shape::new([1, state.len()])),
+        let state_2d = vec![state.to_vec()];
-            &self.device
+        let state_tensor = Tensor::<MyBackend, 2>::from_floats(state_2d, &self.device);
-        );
         
         let q_values = self.q_network.forward(state_tensor);
         let q_data = q_values.into_data().to_vec::<f32>().unwrap();
@@ -195,17 +192,9 @@ impl BurnDqnAgent {
             .collect();
         
         // Convertir en format compatible avec Burn
-        let state_data: Vec<f32> = states.into_iter().flatten().collect();
+        let state_tensor = Tensor::<MyBackend, 2>::from_floats(states, &self.device);
-        let state_tensor = Tensor::<MyBackend, 2>::from_data(
-            burn::tensor::Data::new(state_data, burn::tensor::Shape::new([batch.len(), self.config.state_size])),
-            &self.device
-        );
         let next_state_tensor = if !next_states.is_empty() {
-            let next_state_data: Vec<f32> = next_states.into_iter().flatten().collect();
+            Some(Tensor::<MyBackend, 2>::from_floats(next_states, &self.device))
-            Some(Tensor::<MyBackend, 2>::from_data(
-                burn::tensor::Data::new(next_state_data, burn::tensor::Shape::new([batch.len(), self.config.state_size])),
-                &self.device
-            ))
         } else {
             None
         };