def train_keras_classifier():
from cifar10_classifier import Classifier
from keras.optimizers import SGD
from keras import backend as K
batch_size = 128
epochs = 50
data_augmentation = True
opt = SGD(lr=0.01, decay=1e-6, momentum=0.9, nesterov=True)
with tf.variable_scope('conv') as scope:
model = Classifier().model
model.compile(loss='categorical_crossentropy',
optimizer=opt,
metrics=['accuracy'])
# load data and start training
if data_augmentation:
print('Using real-time data augmentation.')
datagen, (x_train,
y_train), (x_test,
y_test) = data_loader.load_augmented_data()
model.fit_generator(datagen.flow(x_train,
y_train,
batch_size=batch_size),
epochs=epochs,
validation_data=(x_test, y_test),
workers=4,
callbacks=[SGDLearningRateTracker()])
else:
print('Not using data augmentation.')
(x_train, y_train), (x_test, y_test) = data_loader.load_original_data()
model.fit(x_train,
y_train,
batch_size=batch_size,
epochs=epochs,
validation_data=(x_test, y_test),
shuffle=True)
# save as tensorflow model
if not os.path.isdir(SAVE_DIR):
os.makedirs(SAVE_DIR)
model.save(CLASSIFIER_PATH)
from keras.backend import get_session
sess = get_session()
saver = tf.train.Saver()
saver.save(sess, PRETRAINED_PATH)
print('Saved trained model at %s ' % (PRETRAINED_PATH))
# evaluate on test set
scores = model.evaluate(x_test, y_test, verbose=1)
print('Test loss:', scores[0])
def kfold(orig_data, orig_labels, test_data, clf: classifiers.Classifier):
results = []
predictions = []
best_k = 0
for k in range(len(orig_data)):
print("Iteration", k + 1, "over", len(orig_data))
data, labels = orig_data[:], orig_labels[:]
val_data = data.pop(k)
val_labels = labels.pop(k)
train_data = np.concatenate(data)
train_labels = np.concatenate(labels)
print("Fitting...")
clf.fit(train_data, train_labels)
print("Evaluating...")
results.append(clf.evaluate(val_data, val_labels))
print(results[k])
print("Predicting...")
predictions.append(clf.predict(test_data))
print(predictions[k])
if results[k]["Accuracy"] > results[best_k]["Accuracy"]:
best_k = k
return results[best_k], predictions[best_k]
