tensorflow - 在 keras 的 model.fit 中洗牌

Question

在 model.fit的 keras ，有一个shuffle范围，

shuffle: Boolean (whether to shuffle the training data before each epoch) or str (for 'batch'). 'batch' is a special option for dealing with the limitations of HDF5 data; it shuffles in batch-sized chunks. Has no effect when steps_per_epoch is not  None.

假设训练集是一个带有 50000 的列表元素，所以整个列表将在每个纪元之前随机排列？如果批量大小为 250 ，只有属于每个批次的元素才会被置换？正确的理解应该是什么？

Answer 1

它将 shuffle your entire dataset ( x , y and sample_weight together) first然后 make batches根据 batch_size您传递给 fit 的参数.
编辑
正如@yuk 在评论中指出的那样，自 2018 年以来，代码发生了重大变化。 shuffle 的文档参数现在似乎更清晰。您可以选择打乱整个训练数据或只打乱批次:

        shuffle: Boolean (whether to shuffle the training data
            before each epoch) or str (for 'batch'). This argument is ignored
            when `x` is a generator. 'batch' is a special option for dealing
            with the limitations of HDF5 data; it shuffles in batch-sized
            chunks. Has no effect when `steps_per_epoch` is not `None`.