Adding biobjective example Raytune/Nevergrad

examples/raytune_nevergrad_biobjective.py

@@ -0,0 +1,101 @@
+import nevergrad as ng
+import numpy as np
+import nevergrad as ng
+from ray import train, tune
+from ray.tune.search.nevergrad import NevergradSearch
+
+# Sample data
+expected_returns_simulated = np.array([0.08, 0.12, 0.10, 0.15, 0.09])
+covariance_matrix_simulated = np.array([[0.0004, 0.0003, 0.0002, 0.0001, 0.0002],
+                                        [0.0003, 0.0009, 0.0004, 0.0002, 0.0003],
+                                        [0.0002, 0.0004, 0.0010, 0.0003, 0.0004],
+                                        [0.0001, 0.0002, 0.0003, 0.0008, 0.0002],
+                                        [0.0002, 0.0003, 0.0004, 0.0002, 0.0005]])
+
+# Bi-objective problem: Financial portfolio optimization
+def portfolio_optimization(weights, expected_returns, covariance_matrix):
+    portfolio_return = np.dot(weights, expected_returns)
+    portfolio_volatility = np.sqrt(np.dot(weights, np.dot(covariance_matrix, weights)))
+    # maximize portfolio return and minimize portfolio volatility
+    return [-portfolio_return, portfolio_volatility]
+
+# Using Nevergrad to solve the bi-objective problem
+optimizer = ng.optimizers.CMA(parametrization= 5, budget=100)
+
+initial_weights = np.random.dirichlet(np.ones(5), size=1).flatten()  
+
+optimizer.tell(ng.p.Array(init=initial_weights.tolist()), [0, 1],
+               portfolio_optimization(initial_weights, expected_returns_simulated, covariance_matrix_simulated))
+
+# Minimize the bi-objective problem
+optimizer.minimize(lambda x: portfolio_optimization(x, expected_returns_simulated, covariance_matrix_simulated), verbosity=0, constraint_violation=[lambda x: -np.sum(x) + 1, lambda x: np.min(x) - 0])
+
+print("Pareto front:")
+for param in sorted(optimizer.pareto_front(), key=lambda p: p.losses[0]):
+    print(f"{param} with losses {param.losses}")
+
+# Using Raytune with Nevergrad to solve the same problem
+def evaluate(weight_1, weight_2, weight_3, weight_4, weight_5, alpha1, alpha2):
+    weights = [weight_1, weight_2, weight_3, weight_4, weight_5]
+
+    portfolio_return = sum(weights[i] * expected_returns_simulated[i] for i in range(5))
+
+    portfolio_volatility = sum(
+        weights[i] * weights[j] * covariance_matrix_simulated[i, j] 
+        for i in range(5) for j in range(5)
+    )
+
+    portfolio_volatility = np.sqrt(portfolio_volatility)
+
+    return -alpha1 * portfolio_return + alpha2 * portfolio_volatility
+
+def objective(config):
+    alpha1 = config["alpha1"]
+    alpha2 = 1 - alpha1
+    config["alpha2"] = alpha2
+
+    weights = [config["weight_1"], config["weight_2"], config["weight_3"], config["weight_4"], config["weight_5"]]
+    total_weight = sum(weights)
+    normalized_weights = [w / total_weight for w in weights]
+    config['weight_1'] = normalized_weights[0]
+    config['weight_2'] = normalized_weights[1]
+    config['weight_3'] = normalized_weights[2]
+    config['weight_4'] = normalized_weights[3]
+    config['weight_5'] = normalized_weights[4]
+
+    score = evaluate(config["weight_1"], config["weight_2"], config["weight_3"], config["weight_4"], config["weight_5"], alpha1, alpha2)
+    train.report({"mean_loss": score})
+
+algo = NevergradSearch(
+    optimizer=ng.optimizers.CMA,
+)
+
+algo = tune.search.ConcurrencyLimiter(algo, max_concurrent=4)
+
+num_samples = 100
+
+search_config = {
+    "weight_1": tune.uniform(0, 1),
+    "weight_2": tune.uniform(0, 1),
+    "weight_3": tune.uniform(0, 1),
+    "weight_4": tune.uniform(0, 1),
+    "weight_5": tune.uniform(0, 1),
+    "alpha1": tune.uniform(0, 1),
+    "alpha2": 0
+}
+
+tuner = tune.Tuner(
+    objective,
+    tune_config=tune.TuneConfig(
+        metric="mean_loss",
+        mode="min",
+        search_alg=algo,
+        num_samples=num_samples,
+    ),
+    param_space=search_config,
+)
+results = tuner.fit()
+
+best_result_config = results.get_best_result().config
+formatted_config = {key: round(value, 2) for key, value in best_result_config.items()}
+print("Best results found were: ", formatted_config)