怠け者のためのディープラーニング入門 - ハイパーパラメータのチューニング ref: http://qiita.com/akimach/items/a10154e0f0e6dcaafdd0

file0.py

import numpy as np
np.random.seed(1234)
from keras.models import Sequential
from keras.layers import Dense, Activation
from keras import optimizers
from keras.wrappers.scikit_learn import KerasClassifier

from sklearn.utils import shuffle
from sklearn.model_selection import GridSearchCV

param_grid = {
    "activation": ["relu", "tanh", "sigmoid", ],
    "learning_rate": [0.0001, 0.001, 0.01],
    "units": [64, 128, 256],
}

def mlp(activation="relu", learning_rate=0.001, units=128):
    model = Sequential()
    model.add(Dense(units=units, input_dim=2))
    model.add(Activation(activation))
    model.add(Dense(units=units))
    model.add(Activation(activation))
    model.add(Dense(units=1))
    model.add(Activation('sigmoid'))
    model.compile(loss='binary_crossentropy',
              optimizer=optimizers.SGD(lr=learning_rate),
              metrics=['accuracy',])
    return model

n_data = 200
n_class = n_data//2
X_positive = np.random.normal(loc=1.0, scale=1.0, size=(n_class, 2))
X_negative = np.random.normal(loc=-1.0, scale=1.0, size=(n_class, 2))
X = np.r_[X_positive, X_negative]
y = np.r_[[1]*n_class, [0]*n_class]
X, y = shuffle(X, y)

model = KerasClassifier(build_fn=mlp, nb_epoch=50, batch_size=20, verbose=0)
clf = GridSearchCV(estimator=model, param_grid=param_grid, cv=4, scoring='accuracy')
res = clf.fit(X, y)
print ("Accuracy:", res.best_score_)
print ("Hyper Parameters:", res.best_params_)

file1.txt

Accuracy: 0.915
Hyper Parameters: {'activation': 'relu', 'learning_rate': 0.01, 'units': 64}