python - 在顺序读取的多个特征文件上训练 Keras 模型以节省内存

Question

我在尝试读取大量特征文件时遇到了内存问题(见下文)。我想我应该拆分培训文件并按顺序阅读它们。 做到这一点的最佳方法是什么？

x_train = np.load(path_features + 'x_train.npy)
y_train = np.load(path_features + 'y_train.npy)
x_test = np.load(path_features + 'x_test.npy)
y_test = np.load(path_features + 'y_test.npy)

path_models = '../pipelines/' + pipeline + '/models/'

# global params
verbose_level = 1
inp_shape = x_train.shape[1:]

# models
if model_type == 'standard_4':
    print('Starting to train ' + feature_type + '_' + model_type + '.')
    num_classes = 1
    dropout_prob = 0.5
    activation_function = 'relu'
    loss_function = 'binary_crossentropy'
    batch_size = 32
    epoch_count = 100
    opt = SGD(lr=0.001)

    model = Sequential()
    model.add(Conv2D(filters=16, kernel_size=(3, 3), input_shape=inp_shape))
    model.add(MaxPooling2D(pool_size=(2, 2)))
    model.add(Conv2D(filters=32, kernel_size=(3, 3)))
    model.add(MaxPooling2D(pool_size=(2, 2)))
    model.add(Flatten())
    model.add(Dense(64, activation=activation_function))
    model.add(Dropout(rate=dropout_prob))
    model.add(Dense(32, activation=activation_function))
    model.add(Dense(num_classes, activation='sigmoid'))
    model.summary()
    model.compile(loss=loss_function, optimizer=opt, metrics=['accuracy'])
    hist = model.fit(x_train, y_train, batch_size=batch_size, epochs=epoch_count,
                     verbose=verbose_level,
                     validation_data=(x_test, y_test))

    model.save(path_models + category + '_' + feature_type + '_' + model_type + '.h5')
    print('Finished training ' + model_type + '.')

    plot_model(hist, path_models, category, feature_type, model_type)
    print('Saved model charts.')

Answer 1

您可以使用 python generator 或 keras sequence .

生成器应该无限期地生成你的批处理:

def myReader(trainOrTest):
    while True:
        do something to define path_features

        x = np.load(path_features + 'x_' + trainOrTest + '.npy')
        y = np.load(path_features + 'y_' + trainOrTest + '.npy')

        #if you're loading them already in a shape accepted by your model:
        yield (x,y)

然后您可以使用 fit_generator 进行训练并使用 predict_generator 来预测值:

model.fit_generator(myReader(trainOrTest),steps_per_epoch=howManyFiles,epochs=.......)