Merge pull request #179 from hrntsm/featura/updata-optuna-v3.2

Featura/updata optuna v3.2

.gitignore

@@ -451,4 +451,8 @@ $RECYCLE.BIN/
 !.vscode/settings.json
 !.vscode/tasks.json
 !.vscode/launch.json
-!.vscode/extensions.json
+!.vscode/extensions.json
+
+**/venv
+*.db
+*.log

CHANGELOG.md

@@ -15,13 +15,17 @@ Please see [here](https://github.com/hrntsm/Tunny/releases) for the data release
   - Add 2 sample gh file
 - Support CMA-ES with Margin
   - It allows for more efficient optimization in mixed integer problems.
+- Support NSGA-III
+  - For more than 3 objective optimization.
+- Python sample code
 - Open optuna dashboard menu item
 
+
 ### Changed
 
 - When optimizing with CMA-ES, the with Margin option is enabled by default.
-- Bump up optuna v3.1.1
-- Bump up optuna-dashboard v0.9.2
+- Bump up optuna v3.2.0
+- Bump up optuna-dashboard v0.10.2
 
 ### Deprecated
 
@@ -36,6 +40,8 @@ Please see [here](https://github.com/hrntsm/Tunny/releases) for the data release
 
 ### Security
 
+- Bump up scipy v1.10.0
+
 ## [v0.7.2] -2023-03-22
 
 ### Added

Samples/Python/compare_sampler.py

@@ -0,0 +1,78 @@
+# #############################################################################
+# This is an example of how optuna compares each sampler it can handle.
+# #############################################################################
+
+import optuna
+
+n_trials = 50
+
+
+def objective(trial):
+    x = trial.suggest_float("x", -5, 5, step=0.1)
+    y = trial.suggest_int("y", -5, 5)
+    return x**2 + y**2
+
+
+studies = []
+
+# compare samplers
+cmaes = optuna.samplers.CmaEsSampler(with_margin=True)
+study_cmaes = optuna.create_study(
+    sampler=cmaes, direction="minimize", study_name="cmaes"
+)
+study_cmaes.optimize(objective, n_trials=n_trials)
+studies.append(study_cmaes)
+
+nsgaii = optuna.samplers.NSGAIISampler(population_size=10)
+study_nsgaii = optuna.create_study(
+    sampler=nsgaii, direction="minimize", study_name="nsgaii"
+)
+study_nsgaii.optimize(objective, n_trials=n_trials)
+studies.append(study_nsgaii)
+
+nsgaiii = optuna.samplers.NSGAIIISampler(population_size=10)
+study_nsgaiii = optuna.create_study(
+    sampler=nsgaiii, direction="minimize", study_name="nsgaiii"
+)
+study_nsgaiii.optimize(objective, n_trials=n_trials)
+studies.append(study_nsgaiii)
+
+tpe = optuna.samplers.TPESampler()
+study_tpe = optuna.create_study(sampler=tpe, direction="minimize", study_name="tpe")
+study_tpe.optimize(objective, n_trials=n_trials)
+studies.append(study_tpe)
+
+bo = optuna.integration.BoTorchSampler()
+study_bo = optuna.create_study(sampler=bo, direction="minimize", study_name="bo")
+study_bo.optimize(objective, n_trials=n_trials)
+studies.append(study_bo)
+
+random = optuna.samplers.RandomSampler()
+study_random = optuna.create_study(
+    sampler=random, direction="minimize", study_name="random"
+)
+study_random.optimize(objective, n_trials=n_trials)
+
+qmc = optuna.samplers.QMCSampler()
+study_qmc = optuna.create_study(sampler=qmc, direction="minimize", study_name="qmc")
+study_qmc.optimize(objective, n_trials=n_trials)
+
+brute = optuna.samplers.BruteForceSampler()
+study_brute = optuna.create_study(
+    sampler=brute, direction="minimize", study_name="brute"
+)
+study_brute.optimize(objective, n_trials=n_trials)
+
+print("Result")
+print("  true value       :  0.0 {'x': 0.0, 'y': 0}")
+print("  CmaEsSampler     : ", study_cmaes.best_value, study_cmaes.best_params)
+print("  NSGAIISampler    : ", study_nsgaii.best_value, study_nsgaii.best_params)
+print("  NSGAIIISampler   : ", study_nsgaiii.best_value, study_nsgaiii.best_params)
+print("  TPESampler       : ", study_tpe.best_value, study_tpe.best_params)
+print("  BoTorchSampler   : ", study_bo.best_value, study_bo.best_params)
+print("  RandomSampler    : ", study_random.best_value, study_random.best_params)
+print("  QMCSampler       : ", study_qmc.best_value, study_qmc.best_params)
+print("  BruteForceSampler: ", study_brute.best_value, study_brute.best_params)
+
+fig = optuna.visualization.plot_optimization_history(studies, error_bar=False)
+fig.show()

Samples/Python/create_storage.py

@@ -0,0 +1,29 @@
+# #############################################################################
+# This is an example of creating STORAGE.
+# #############################################################################
+
+import optuna
+
+
+def objective(trial):
+    x = trial.suggest_float("x", -100, 100)
+    return x**2
+
+
+# SQLite3 storage
+storage = optuna.storages.RDBStorage(
+    url="sqlite:///test.db",
+)
+study_db = optuna.create_study(storage=storage)
+study_db.optimize(objective, n_trials=10)
+
+
+# Journal storage
+file_path = "test.log"
+lock_obj = optuna.storages.JournalFileOpenLock(file_path)
+
+storage = optuna.storages.JournalStorage(
+    optuna.storages.JournalFileStorage(file_path, lock_obj=lock_obj),
+)
+study_journal = optuna.create_study(storage=storage)
+study_journal.optimize(objective, n_trials=10)

Samples/Python/cull_trial.py

@@ -0,0 +1,43 @@
+# ##############################################################################
+# This is an example of how to cull unintended trial results from STUDY
+# ##############################################################################
+
+import optuna
+
+storage_path = "fish.log"
+target_study_name = "study_target"
+culled_study_name = "study_cull"
+cull_trial_number = [10, 15, 17]
+
+# If you use .log file, use blow.
+lock_obj = optuna.storages.JournalFileOpenLock(storage_path)
+storage = optuna.storages.JournalStorage(
+    optuna.storages.JournalFileStorage(storage_path, lock_obj=lock_obj),
+)
+
+# If you use RDB, use blow.
+# storage = optuna.storages.RDBStorage("sqlite:///" + storage_path)
+
+# Load storage
+study = optuna.load_study(storage=storage, study_name=target_study_name)
+usr_attr = study.user_attrs
+trials = study.get_trials()
+directions = study.directions
+
+# Create new study to save cull result
+cull_study = optuna.create_study(
+    storage=storage, study_name=culled_study_name, directions=directions
+)
+
+# If you want to read this result file from Tunny, you need to set the following.
+# Tunny needs to read some attributes
+for key, value in usr_attr.items():
+    cull_study.set_user_attr(key, value)
+
+# Cull trials
+for num in cull_trial_number:
+    trials = list(filter(lambda trial: trial.number != num, trials))
+cull_study.add_trials(trials=trials)
+
+# visualize result(if you want)
+optuna.visualization.plot_pareto_front(cull_study).show()

Samples/Python/plot.py

@@ -0,0 +1,63 @@
+# #############################################################################
+# This is an example of how to use the visualization function.
+# #############################################################################
+
+
+import optuna
+
+n_trials = 50
+
+
+def objective(trial):
+    # Rosenbrock function
+    x = trial.suggest_float("x", -5, 5, step=0.1)
+    y = trial.suggest_int("y", -5, 5)
+    trial.set_user_attr("too_long_str", "too_long_str, " * 100)
+
+    return [(1 - x) ** 2 + 100 * (y - x**2) ** 2, x]
+
+
+tpe = optuna.samplers.TPESampler()
+study = optuna.create_study(sampler=tpe, directions=["minimize", "minimize"])
+study.optimize(objective, n_trials=n_trials)
+
+name = "Rosenbrock function"
+
+optuna.visualization.plot_slice(
+    study,
+    params=["x", "y"],
+    target=lambda t: t.values[0],
+    target_name=name,
+).show()
+
+optuna.visualization.plot_pareto_front(
+    study,
+    target_names=[name, "x"],
+).show()
+
+optuna.visualization.plot_param_importances(
+    study,
+    target=lambda t: t.values[0],
+    target_name=name,
+).show()
+
+optuna.visualization.plot_contour(
+    study,
+    params=["x", "y"],
+    target=lambda t: t.values[0],
+    target_name=name,
+).show()
+
+optuna.visualization.plot_optimization_history(
+    study, target=lambda t: t.values[0], target_name=name
+).show()
+
+optuna.visualization.plot_parallel_coordinate(
+    study, params=["x", "y"], target=lambda t: t.values[0], target_name=name
+).show()
+
+optuna.visualization.plot_edf(
+    study,
+    target=lambda t: t.values[0],
+    target_name=name,
+).show()

Samples/Python/plot_with_coloring_of_results.py

@@ -0,0 +1,85 @@
+# #############################################################################
+# This example shows how to coloring and plot the optimized results.
+# #############################################################################
+
+import optuna
+import plotly.graph_objects as go
+
+
+def objective(trial):
+    x = trial.suggest_float("x", -10, 10)
+    y = trial.suggest_float("y", -10, 10)
+    return [x + y, x - y]
+
+
+study = optuna.create_study(directions=["minimize", "minimize"])
+study.optimize(objective, n_trials=100)
+
+criteria_x = 0
+criteria_value = 5
+
+good_trials = []
+no_good_trials = []
+
+for trial in filter(
+    lambda t: t.state == optuna.trial.TrialState.COMPLETE, study.trials
+):
+    if trial.values[0] < criteria_value and trial.params["x"] < criteria_x:
+        good_trials.append(trial)
+    else:
+        no_good_trials.append(trial)
+
+traces = []
+
+traces.append(
+    go.Scatter(
+        x=[t.values[0] for t in good_trials],
+        y=[t.values[1] for t in good_trials],
+        mode="markers",
+        name="good",
+        marker={"color": "blue"},
+    )
+)
+
+traces.append(
+    go.Scatter(
+        x=[t.values[0] for t in no_good_trials],
+        y=[t.values[1] for t in no_good_trials],
+        mode="markers",
+        name="no good",
+        marker={"color": "#cccccc"},
+    )
+)
+
+fig = go.Figure(traces)
+fig.update_layout(
+    plot_bgcolor="white",
+    xaxis=dict(
+        title="x+y",
+        showline=True,
+        linewidth=1,
+        linecolor="black",
+        zeroline=True,
+        zerolinecolor="black",
+        zerolinewidth=1,
+        showgrid=True,
+        gridcolor="lightgray",
+        range=[-10, 10],
+    ),
+    yaxis=dict(
+        title="x-y",
+        showline=True,
+        linewidth=1,
+        linecolor="black",
+        zeroline=True,
+        zerolinecolor="black",
+        zerolinewidth=1,
+        showgrid=True,
+        gridcolor="lightgray",
+        range=[-10, 10],
+    ),
+    width=640,
+    height=480,
+)
+
+fig.show()

Tunny/Lib/requirements.txt

@@ -13,15 +13,15 @@ MarkupSafe==2.1.2
 multipledispatch==0.6.0
 numpy==1.24.1
 opt-einsum==3.3.0
-optuna==3.1.1
-optuna-dashboard==0.9.2
+optuna==3.2.0
+optuna-dashboard==0.10.2
 packaging==23.0
 plotly==5.9.0
 pyro-api==0.1.2
 pyro-ppl==1.8.4
 PyYAML==6.0
 scikit-learn==1.2.0
-scipy==1.8.1
+scipy==1.10.0
 six==1.16.0
 SQLAlchemy==1.4.46
 tenacity==8.1.0

Tunny/Settings/Sampler/NSGAIII.cs

@@ -0,0 +1,11 @@
+namespace Tunny.Settings.Sampler
+{
+    ///  <summary>
+    /// https://optuna.readthedocs.io/en/stable/reference/generated/optuna.samplers.NSGAIIISampler.html
+    /// </summary>
+    public class NSGAIII : NSGAII
+    {
+        public double[] ReferencePoints { get; set; }
+        public int DividingParameter { get; set; } = 3;
+    }
+}

Tunny/Settings/Sampler/SamplerSettings.cs

@@ -9,6 +9,7 @@ public class SamplerSettings
         public Tpe Tpe { get; set; } = new Tpe();
         public CmaEs CmaEs { get; set; } = new CmaEs();
         public NSGAII NsgaII { get; set; } = new NSGAII();
+        public NSGAIII NsgaIII { get; set; } = new NSGAIII();
         public QuasiMonteCarlo QMC { get; set; } = new QuasiMonteCarlo();
         public BoTorch BoTorch { get; set; } = new BoTorch();
     }