From b555aafe7707df4fa259b6abf0e2f2446203e6f9 Mon Sep 17 00:00:00 2001
From: jgrusewski <jgrusewski@localhost>
Date: Wed, 25 Mar 2026 20:38:56 +0100
Subject: [PATCH] feat: wire MBP-10 + trades data into hyperopt campaign config

Adds mbp10_data_dir and trades_data_dir to CampaignConfig, passed to
DQNTrainer via with_ofi_data_dirs(). Enables OFI features (VPIN,
Kyle's Lambda, etc.) in hyperopt campaigns. dqn_full() now defaults
to 50 epochs/trial for baseline runs.

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
---
 crates/ml/src/hyperopt/campaign.rs | 27 +++++++++++++++++++++++----
 1 file changed, 23 insertions(+), 4 deletions(-)

diff --git a/crates/ml/src/hyperopt/campaign.rs b/crates/ml/src/hyperopt/campaign.rs
index dff38cd67..a644165c7 100644
--- a/crates/ml/src/hyperopt/campaign.rs
+++ b/crates/ml/src/hyperopt/campaign.rs
@@ -49,6 +49,13 @@ pub struct CampaignConfig {
     pub results_base_dir: PathBuf,
     /// Campaign execution mode.
     pub mode: CampaignMode,
+    /// Optional MBP-10 order book data directory for OFI features.
+    /// When set, enables 8 additional OFI features (VPIN, Kyle's Lambda, etc.)
+    /// producing state_dim=80 instead of 72.
+    pub mbp10_data_dir: Option<PathBuf>,
+    /// Optional trades data directory for VPIN/Kyle's Lambda computation.
+    /// Required alongside mbp10_data_dir for full OFI feature extraction.
+    pub trades_data_dir: Option<PathBuf>,
 }
 
 impl CampaignConfig {
@@ -65,11 +72,13 @@ impl CampaignConfig {
             max_epochs_per_trial: 81,
             results_base_dir: PathBuf::from("ml/hyperopt_results"),
             mode: CampaignMode::Standard,
+            mbp10_data_dir: None,
+            trades_data_dir: None,
         }
     }
 
     /// DQN full pipeline: all 3 training phases (BC → RL → Refinement).
-    /// 20 trials × 100 epochs — covers Phase 1 (warmup), Phase 2 (full RL),
+    /// 20 trials × 50 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();
@@ -80,9 +89,11 @@ impl CampaignConfig {
             data_dir: PathBuf::from("test_data/real/databento/ml_training"),
             max_batch_size: max_batch,
             early_stopping_eta: 3,
-            max_epochs_per_trial: 100,
+            max_epochs_per_trial: 50,
             results_base_dir: PathBuf::from("ml/hyperopt_results"),
             mode: CampaignMode::Full,
+            mbp10_data_dir: None,
+            trades_data_dir: None,
         }
     }
 
@@ -99,6 +110,8 @@ impl CampaignConfig {
             max_epochs_per_trial: 81,
             results_base_dir: PathBuf::from("ml/hyperopt_results"),
             mode: CampaignMode::Standard,
+            mbp10_data_dir: None,
+            trades_data_dir: None,
         }
     }
 
@@ -189,8 +202,12 @@ fn run_dqn_campaign(
         .seed(42)
         .build();
 
-    // Create DQN adapter
-    let adapter = DQNTrainer::new(&config.data_dir, config.max_epochs_per_trial)?;
+    // Create DQN adapter with optional MBP-10 + trades data for OFI features
+    let adapter = DQNTrainer::new(&config.data_dir, config.max_epochs_per_trial)?
+        .with_ofi_data_dirs(
+            config.mbp10_data_dir.as_ref().map(|p| p.to_string_lossy().into_owned()),
+            config.trades_data_dir.as_ref().map(|p| p.to_string_lossy().into_owned()),
+        );
 
     // Run optimization
     let result = optimizer.optimize(adapter)?;
@@ -318,6 +335,8 @@ mod tests {
             max_epochs_per_trial: 10, // Quick validation
             results_base_dir: PathBuf::from("ml/hyperopt_results"),
             mode: CampaignMode::Full,
+            mbp10_data_dir: None,
+            trades_data_dir: None,
         };
 
         println!("Starting local DQN hyperopt: {} trials × {} epochs ({:?} mode)",