feat(viz): add population evolution visualization and history interval support

Author: MichalRedm

scripts/experiment_evolution.py

@@ -0,0 +1,97 @@
+import os
+import sys
+import random
+import numpy as np
+from datetime import datetime
+
+# Add project root to sys.path
+sys.path.append(os.path.dirname(os.path.dirname(os.path.abspath(__file__))))
+
+from src.config import (
+    ASSET_NAMES,
+    DATA_FOLDER,
+    OUTPUT_FOLDER,
+    TRAINING_ITERATIONS,
+    NUM_MIXTURES,
+)
+from src.data_handler import load_data
+from src.predictors import ExactGPPredictor
+from src.portfolio import (
+    calculate_expected_returns_and_cov,
+    MOEADOptimizer,
+)
+from src.visualization import plot_population_evolution
+
+
+def seed_everything(seed=42):
+    random.seed(seed)
+    np.random.seed(seed)
+    try:
+        import torch
+
+        torch.manual_seed(seed)
+    except ImportError:
+        pass
+
+
+def run_evolution_demo():
+    seed_everything(42)
+
+    # Setup folders
+    run_id = datetime.now().strftime("%Y%m%d_%H%M%S_moead_evolution")
+    run_folder = os.path.join(OUTPUT_FOLDER, run_id)
+    os.makedirs(run_folder, exist_ok=True)
+
+    print(f"Starting MOEA/D Evolution Demo. Result will be saved to {run_folder}")
+
+    # 1. Load data and calculate expected returns/cov
+    # (Simplified data hash for speed in this demo)
+    raw_data_dict = load_data(DATA_FOLDER, ASSET_NAMES, verbose=False)
+    predictor = ExactGPPredictor(
+        training_iterations=TRAINING_ITERATIONS,
+        num_mixtures=NUM_MIXTURES,
+        verbose=False,
+    )
+
+    sample_timestamps = raw_data_dict[ASSET_NAMES[0]]["timestamps"]
+    target_timestamp = sample_timestamps[-1] + 100.0
+
+    expected_returns, cov_matrix = calculate_expected_returns_and_cov(
+        raw_data_dict, predictor, target_timestamp, ASSET_NAMES, verbose=False
+    )
+
+    data_kwargs = {"expected_returns": expected_returns, "cov_matrix": cov_matrix}
+
+    # 2. Run MOEA/D with history recording
+    print("Running MOEA/D with history recording (interval=10)...")
+    moead = MOEADOptimizer()
+    history_interval = 10
+    generations = 200
+
+    # generate_pareto_front returns: metrics, population, weight_vectors, history (if record_history=True)
+    results = moead.generate_pareto_front(
+        num_points=100,
+        generations=generations,
+        verbose=True,
+        record_history=True,
+        history_interval=history_interval,
+        **data_kwargs,
+    )
+
+    # unpack results
+    moead_metrics, _, _, history = results
+
+    # 3. Plot Population Evolution
+    print("Generating Population Evolution Plot...")
+    obj_names = ["Expected Return", "Expected Risk"]
+    save_path = os.path.join(run_folder, "population_evolution.png")
+
+    plot_population_evolution(
+        history, obj_names, history_interval=history_interval, save_path=save_path
+    )
+
+    print(f"Success! Plot saved to {save_path}")
+
+
+if __name__ == "__main__":
+    run_evolution_demo()

src/portfolio/moead.py

@@ -25,6 +25,7 @@ def generate_pareto_front(
         nr=2,
         verbose=False,
         crossover_operator="sbx",
+        history_interval=10,
         **kwargs,
     ):
         """
@@ -204,7 +205,8 @@ def normalise(f):
                                 z_ideal[k] = off_f[k]
 
             if kwargs.get("record_history", False):
-                history.append(f_phys.copy())
+                if _gen % history_interval == 0 or _gen == generations - 1:
+                    history.append(f_phys.copy())
 
         # ── 6. Return physical-space metrics ─────────────────────────────────
         pareto_metrics = {

src/portfolio/moead_awa.py

@@ -23,7 +23,14 @@ def _get_non_dominated(self, solutions, objectives):
         return is_efficient
 
     def generate_pareto_front(
-        self, num_points=100, generations=100, T=10, nr=2, verbose=False, **kwargs
+        self,
+        num_points=100,
+        generations=100,
+        T=10,
+        nr=2,
+        verbose=False,
+        history_interval=10,
+        **kwargs,
     ):
         """
         MOEA/D-AWA (Adaptive Weight vector Adjustment).
@@ -222,7 +229,8 @@ def normalise(f):
                     neighbors = self._get_neighbors(weight_vectors, T)
 
             if kwargs.get("record_history", False):
-                history.append(f_phys.copy())
+                if gen % history_interval == 0 or gen == generations - 1:
+                    history.append(f_phys.copy())
 
         pareto_metrics = {
             obj.name: f_phys[:, i] for i, obj in enumerate(self.problem.objectives)

src/portfolio/moead_dra.py

@@ -8,7 +8,14 @@ def __init__(self, problem=None):
         super().__init__(problem)
 
     def generate_pareto_front(
-        self, num_points=100, generations=100, T=10, nr=2, verbose=False, **kwargs
+        self,
+        num_points=100,
+        generations=100,
+        T=10,
+        nr=2,
+        verbose=False,
+        history_interval=10,
+        **kwargs,
     ):
         """
         MOEA/D-DRA (Dynamic Resource Allocation).
@@ -163,7 +170,8 @@ def normalise(f):
                                 z_ideal[k] = off_f[k]
 
             if kwargs.get("record_history", False):
-                history.append(f_phys.copy())
+                if gen % history_interval == 0 or gen == generations - 1:
+                    history.append(f_phys.copy())
 
         pareto_metrics = {
             obj.name: f_phys[:, i] for i, obj in enumerate(self.problem.objectives)

src/visualization.py

@@ -395,3 +395,61 @@ def plot_variants_comparison(ref_metrics, variants_results, save_path=None):
     plt.grid(True, alpha=0.3)
     if save_path:
         plt.savefig(save_path)
+
+
+def plot_population_evolution(history, obj_names, history_interval=10, save_path=None):
+    """
+    Plots the evolution of the population through different generations.
+
+    Args:
+        history: Array of shape (n_recorded, n_points, n_objectives)
+        obj_names: List of objective names (e.g., ["Return", "Risk"])
+        history_interval: Interval at which history was recorded
+        save_path: Path to save the plot
+    """
+    n_recorded = len(history)
+    if n_recorded == 0:
+        print("Empty history provided.")
+        return
+
+    plt.figure(figsize=(12, 8))
+    cmap = plt.get_cmap("viridis")
+
+    for i, pop in enumerate(history):
+        # Calculate approximate generation number
+        gen = i * history_interval
+        color = cmap(i / max(1, n_recorded - 1))
+
+        plt.scatter(
+            pop[:, 0],
+            pop[:, 1],
+            color=color,
+            alpha=0.6,
+            s=20,
+            label=f"Gen {gen}" if i % 2 == 0 or i == n_recorded - 1 else "",
+        )
+
+    plt.title("Evolution of Pareto Front (Population Dynamics)")
+    plt.xlabel(obj_names[0])
+    plt.ylabel(obj_names[1])
+    plt.grid(True, alpha=0.3)
+
+    # Add colorbar
+    sm = plt.cm.ScalarMappable(cmap=cmap, norm=plt.Normalize(vmin=0, vmax=gen))
+    plt.colorbar(sm, label="Generation", ax=plt.gca())
+
+    # Limit legend items if too many
+    handles, labels = plt.gca().get_legend_handles_labels()
+    if len(labels) > 10:
+        indices = np.linspace(0, len(labels) - 1, 6, dtype=int)
+        plt.legend(
+            [handles[i] for i in indices],
+            [labels[i] for i in indices],
+            title="Generations Sample",
+        )
+    else:
+        plt.legend(title="Generations Sample")
+
+    plt.tight_layout()
+    if save_path:
+        plt.savefig(save_path)