呜呜一天不到,竟然写出来了
修了一堆一堆bug,发现自己神经网络的基础是不够好的
能跑了!虽然我怀疑这玩意儿不对劲,测试集正确率卡在70%(这可是个二分类啊,最低正确率也应该是50%)
训练 Epoch: 1000 [0/209 (0%)] Loss: 0.034381
训练 Epoch: 1000 [28/209 (14%)] Loss: 0.060970
训练 Epoch: 1000 [56/209 (29%)] Loss: 0.034686
训练 Epoch: 1000 [84/209 (43%)] Loss: 0.037471
训练 Epoch: 1000 [112/209 (57%)] Loss: 0.019298
训练 Epoch: 1000 [140/209 (71%)] Loss: 0.028701
训练 Epoch: 1000 [168/209 (86%)] Loss: 0.026698
测试 set: Average loss: 0.8679, Accuracy: 35/50 #(70%)
明天试着写一个卷积神经网络?别搞这不靠谱的全连接了