Created by: yeyupiaoling

• 在使用自定义图片数据集训练时,在训练到一半是突然报错,错误信息如下:

Pass 0, Batch 0, Cost 14.597490, {'classification_error_evaluator': 0.7265625}

Pass 0, Batch 1, Cost 39.000000, {'classification_error_evaluator': 0.609375}

Pass 0, Batch 2, Cost 36.000000, {'classification_error_evaluator': 0.5625}
Thread [140661301901056] Forwarding __conv_1__, 
*** Aborted at 1515036867 (unix time) try "date -d @1515036867" if you are using GNU date ***

PC: @                0x0 (unknown)

*** SIGFPE (@0x7fee67489187) received by PID 18909 (TID 0x7fee42eef700) from PID 1732809095; stack trace: ***

    @     0x7fee82e4d390 (unknown)

    @     0x7fee67489187 sgemm_kernel_NEHALEM

• 这是什么原因,我图片的命名是用uuid的没有用到时间毫秒,部分数据如下:

../data/image/fee16631-f096-11e7-aff8-2cd05afbe631.png	1
../data/image/fee16632-f096-11e7-aff8-2cd05afbe631.png	1
../data/image/fee16633-f096-11e7-aff8-2cd05afbe631.png	1
../data/image/fee16634-f096-11e7-aff8-2cd05afbe631.png	1

• 读取方式如下:

    train_reader = paddle.batch(
        paddle.reader.shuffle(
            reader.train_reader('../data/trainer.list'),
            buf_size=1000),
        batch_size=BATCH_SIZE)