Open augustkx opened 7 years ago
@augustkx 有个函数是category_crossentropy. 我觉得这个binary_crossentropy可以看做是category_crossentropy的特例, 即只有两个分类.
谢谢!初学者能再弱问下吗? 其实不太理解metrics 下的函数,究竟跟loss有什么区别,我看见说“ Metric functions are to be supplied in the metrics parameter when a model is compiled.” 那为什么不跟loss保持一致呢?
我理解的Metrics属于监督机制, 即你在模型搭建, 选择能量函数, 优化方法, 进行模型训练的时候, metrics可以不提供. 但涉及回调函数的时候, Metrics提供的值就会起到帮助.
比如说你期望在loss不在下降的时候, 减小搜索步长, 那么你就需要监测loss. 即metrics=['loss']
documentation里都搜不到。。