当数据集的数据过大时，采用生成器generator生成数据循环调用

生成

def gen_data(xs, ys, batch_size=100):
    image_buf = list()
    label_buf = list()
    while True:
        for image, label in zip(xs, ys):
            if len(image_buf) < batch_size:
                image_buf.append(cv2.cvtColor(cv2.resize(np.reshape(image,[28,28]),(224,224)),cv2.COLOR_GRAY2BGR))
                label_buf.append(label)
            else:
                yield np.array(image_buf), np.array(label_buf)
                image_buf, label_buf = list(), list()

gen = gen_data(mnist.train.images, mnist.train.labels)

调用

for i in range(num_batches):
        xs,ys=next(gen)
        end=start+batch_size
        end=min(end,max_data_len)
        xx=xs[start:end]
        yy=ys[start:end]