MLP example

examples/MLPClassifier.py

@@ -0,0 +1,49 @@
+import argparse
+import pickle
+
+from digen import Benchmark
+
+# Load a package with DIGEN benchmark
+benchmark = Benchmark()
+
+# seedmap=dict(map(lambda x : (x.split('_')[0],x.split('_')[1]), benchmark.list_datasets()))
+parser = argparse.ArgumentParser()
+parser.add_argument("-d", "--dataset", default=None, help="Specify a dataset (otherwise all datasets are used)",
+                    required=False, nargs='?')
+args = parser.parse_args()
+
+datasets = args.dataset
+if args.dataset is None:
+    datasets = benchmark.list_datasets()
+
+# Create your default class here or import from the package. As an example, we re benchmarking ExtraTreesClassifier from scikit-learn:
+from sklearn.neural_network import MLPClassifier
+
+est = MLPClassifier()
+
+
+# In order to properly benchmark a method, we need to define its parameters and their values.
+# Please set the expected range of hyper parameters for your method below. For details, please refer to Optuna.
+def params_myParamScope(trial):
+
+    n_layers = trial.suggest_int('n_layers', 1, 3)
+    layers = []
+    for i in range(n_layers):
+        layers.append(trial.suggest_int(f'n_neurons_{i}', 4, 128))
+
+    params = {
+        'activation':  trial.suggest_categorical(name='activation', choices=['identity', 'logistic', 'tanh', 'relu']),
+        'solver':  trial.suggest_categorical(name='solver', choices=['lbfgs', 'sgd', 'adam']),
+        'alpha':  trial.suggest_loguniform('alpha', 0.0001, 1.0),
+        'hidden_layer_sizes':  tuple(layers),
+        'max_iter' : 10000
+    }
+
+    return params
+
+
+
+# Perform optimization of the method on DIGEN datasets
+results = benchmark.optimize(est=est, datasets=datasets, parameter_scopes=params_myParamScope,
+                             storage='sqlite:///'+datasets+'-mlp.db', local_cache_dir='.')
+pickle.dump( results, open( datasets+".pkl", "wb" ) )