From eceaf45b11853210dbbf464d10b4fe6c0210a586 Mon Sep 17 00:00:00 2001
From: jgrusewski <jgrusewski@localhost>
Date: Wed, 25 Mar 2026 08:44:41 +0100
Subject: [PATCH] =?UTF-8?q?feat:=20--full=20mode=20for=203-phase=20hyperop?=
 =?UTF-8?q?t=20pipeline=20(BC=20=E2=86=92=20RL=20=E2=86=92=20Refinement)?=
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

- CampaignMode enum: Quick, Standard, Full
- dqn_full() constructor: 20 trials × 100 epochs (covers all 3 phases)
- fxt tune start --full flag: auto-sets 20 trials, 100 epochs
- Phase 1 (BC): MSE warmup + expert demos + DT pretrain
- Phase 2 (RL): all 25 features, C51 ramp, HER
- Phase 3 (Refine): pure C51, shrink-and-perturb

Usage: fxt tune start --model dqn --full --gpu

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
---
 bin/fxt/src/commands/tune.rs       | 14 ++++++++--
 crates/ml/src/hyperopt/campaign.rs | 42 ++++++++++++++++++++++++++++--
 2 files changed, 52 insertions(+), 4 deletions(-)

diff --git a/bin/fxt/src/commands/tune.rs b/bin/fxt/src/commands/tune.rs
index 3c31d626a..00f69a2d3 100644
--- a/bin/fxt/src/commands/tune.rs
+++ b/bin/fxt/src/commands/tune.rs
@@ -31,6 +31,9 @@ enum TuneAction {
         /// Enable GPU
         #[arg(long)]
         gpu: bool,
+        /// Full 3-phase pipeline: BC → Online RL → Refinement (100 epochs/trial)
+        #[arg(long)]
+        full: bool,
     },
     /// Stop a running optimization
     Stop {
@@ -249,16 +252,23 @@ impl TuneCommand {
                 trials,
                 config,
                 gpu,
+                full,
             } => {
+                let description = if *full {
+                    "Full 3-phase pipeline: BC → Online RL → Refinement".to_owned()
+                } else {
+                    String::new()
+                };
+                let num_trials = if *full && *trials == 50 { 20 } else { *trials };
                 let resp = client
                     .ml_training()
                     .start_tuning_job(ml_training::StartTuningJobRequest {
                         model_type: model.to_uppercase(),
-                        num_trials: *trials,
+                        num_trials,
                         config_path: config.clone().unwrap_or_default(),
                         data_source: None,
                         use_gpu: *gpu,
-                        description: String::new(),
+                        description,
                         tags: Default::default(),
                     })
                     .await?
diff --git a/crates/ml/src/hyperopt/campaign.rs b/crates/ml/src/hyperopt/campaign.rs
index f092691f0..b520c8391 100644
--- a/crates/ml/src/hyperopt/campaign.rs
+++ b/crates/ml/src/hyperopt/campaign.rs
@@ -16,6 +16,21 @@ use crate::batch_size_resolver::resolve_batch_size;
 pub use crate::ModelType;
 
 /// Campaign configuration for multi-trial hyperparameter optimization.
+/// Campaign execution mode.
+#[derive(Debug, Clone, Copy, PartialEq, Eq)]
+pub enum CampaignMode {
+    /// Quick: fewer trials, fewer epochs. For local validation.
+    Quick,
+    /// Standard: default trials and epochs. For single-phase hyperopt.
+    Standard,
+    /// Full: all three training phases (BC → Online RL → Refinement).
+    /// Runs max_epochs=100 per trial to cover all phases.
+    /// Phase 1 (BC): epochs 0..c51_warmup_epochs (MSE + expert demos + DT)
+    /// Phase 2 (RL): epochs warmup..80% (all features, C51 ramp, HER)
+    /// Phase 3 (Refine): last 20% (pure C51, shrink-and-perturb)
+    Full,
+}
+
 #[derive(Debug, Clone)]
 pub struct CampaignConfig {
     /// Which model to optimize.
@@ -32,6 +47,8 @@ pub struct CampaignConfig {
     pub max_epochs_per_trial: usize,
     /// Base directory for results output.
     pub results_base_dir: PathBuf,
+    /// Campaign execution mode.
+    pub mode: CampaignMode,
 }
 
 impl CampaignConfig {
@@ -47,6 +64,25 @@ impl CampaignConfig {
             early_stopping_eta: 3,
             max_epochs_per_trial: 81,
             results_base_dir: PathBuf::from("ml/hyperopt_results"),
+            mode: CampaignMode::Standard,
+        }
+    }
+
+    /// DQN full pipeline: all 3 training phases (BC → RL → Refinement).
+    /// 20 trials × 100 epochs — covers Phase 1 (warmup), Phase 2 (full RL),
+    /// and Phase 3 (refinement with shrink-and-perturb).
+    pub fn dqn_full() -> Self {
+        let caps = cached_capabilities();
+        let max_batch = resolve_batch_size(caps, &memory_profile::estimates::DQN, 512);
+        Self {
+            model_type: ModelType::DQN,
+            num_trials: 20,
+            data_dir: PathBuf::from("test_data/real/databento/ml_training"),
+            max_batch_size: max_batch,
+            early_stopping_eta: 3,
+            max_epochs_per_trial: 100,
+            results_base_dir: PathBuf::from("ml/hyperopt_results"),
+            mode: CampaignMode::Full,
         }
     }
 
@@ -62,6 +98,7 @@ impl CampaignConfig {
             early_stopping_eta: 3,
             max_epochs_per_trial: 81,
             results_base_dir: PathBuf::from("ml/hyperopt_results"),
+            mode: CampaignMode::Standard,
         }
     }
 
@@ -280,10 +317,11 @@ mod tests {
             early_stopping_eta: 3,
             max_epochs_per_trial: 50,
             results_base_dir: PathBuf::from("ml/hyperopt_results"),
+            mode: CampaignMode::Full,
         };
 
-        println!("Starting local DQN hyperopt: {} trials × {} epochs",
-            config.num_trials, config.max_epochs_per_trial);
+        println!("Starting local DQN hyperopt: {} trials × {} epochs ({:?} mode)",
+            config.num_trials, config.max_epochs_per_trial, config.mode);
         println!("Data: {}", config.data_dir.display());
 
         let result = run_campaign(&config).expect("campaign should complete");