实验记录：模型训练时loss为INF(无穷大)
INF(infinity)——无穷大
本次原因：
未进行归一化，损失函数使用log(x)，在log(x)函数中，x越小趋近于零，结果会返回INF
其它原因：
1.除以零错误：
在计算过程中，如果尝试将一个数除以零，将导致无穷大的损失值。
2.数值不稳定性：
在某些情况下，由于数值不稳定性，梯度下降算法可能会产生非常大的梯度，从而导致损失值变得非常大。
3.学习率过高：
如果学习率设置得过高，梯度更新可能会变得非常大，从而在某些迭代中导致损失值变得非常大。
4.模型结构问题：
模型结构可能存在问题，例如，某些层的权重可能设置得过大或过小，导致计算过程中的数值不稳定。
5.初始化问题：
权重的初始化可能不合适，例如，如果权重的初始值过大或过小，可能会导致计算过程中的数值不稳定。
6.输入数据问题：
输入数据可能包含异常值或缺失值，这可能导致模型在计算过程中遇到问题。
可能的解决方法：
1.检查数据：确保输入数据没有异常值或缺失值。
2.调整学习率：尝试降低学习率，以减少梯度更新的幅度。
3.检查模型结构：检查模型结构是否合理，特别是权重和偏置的初始化是否合适。
4.使用梯度裁剪：在训练深度神经网络时，可以使用梯度裁剪来防止梯度爆炸问题。
5.使用更稳定的优化算法：例如，可以使用Adam或RMSProp等更稳定的优化算法来代替SGD。
6.减小批次大小（Batch Size）：减小批次大小可以减少梯度的噪声，但也会增加训练时间。
7.检查硬件和软件：确保硬件和软件环境没有问题，例如内存溢出或GPU内存不足等问题。
结果输出：
幻数:2051, 图片数量: 26843张, 图片大小: 28*28
100%|██████████| 26843/26843 [00:00<00:00, 33193.52it/s]
100%|██████████| 26843/26843 [00:00<00:00, 3341972.22it/s]
魔数:2049, 图片数量: 26843张
幻数:2051, 图片数量: 2982张, 图片大小: 28*28
100%|██████████| 2982/2982 [00:00<00:00, 33162.09it/s]
魔数:2049, 图片数量: 2982张
cpu
100%|██████████| 2982/2982 [00:00<00:00, 3351397.25it/s]
0%| | 0/210 [00:00<?, ?it/s]training on mps:0
100%|██████████| 210/210 [00:02<00:00, 74.55it/s]
100%|██████████| 24/24 [00:00<00:00, 203.93it/s]
Epoch 1: train_loss inf, train_acc 0.936669, test_acc 0.988263, test_loss 0.165266, time 2.9 sec
100%|██████████| 210/210 [00:02<00:00, 80.01it/s]
100%|██████████| 24/24 [00:00<00:00, 233.16it/s]
0%| | 0/210 [00:00<?, ?it/s]Epoch 2: train_loss 0.142747, train_acc 0.985061, test_acc 0.997653, test_loss 0.072634, time 2.7 sec
100%|██████████| 210/210 [00:02<00:00, 79.98it/s]
100%|██████████| 24/24 [00:00<00:00, 236.72it/s]
Epoch 3: train_loss 0.062067, train_acc 0.992773, test_acc 0.997988, test_loss 0.038777, time 2.7 sec
100%|██████████| 210/210 [00:02<00:00, 80.56it/s]
100%|██████████| 24/24 [00:00<00:00, 229.97it/s]
Epoch 4: train_loss 0.052101, train_acc 0.994077, test_acc 0.994299, test_loss 0.086992, time 2.7 sec
100%|██████████| 210/210 [00:02<00:00, 80.91it/s]
100%|██████████| 24/24 [00:00<00:00, 229.63it/s]
0%| | 0/210 [00:00<?, ?it/s]Epoch 5: train_loss 0.045025, train_acc 0.994934, test_acc 0.996311, test_loss 0.081664, time 2.7 sec
100%|██████████| 210/210 [00:02<00:00, 80.65it/s]
100%|██████████| 24/24 [00:00<00:00, 233.24it/s]
0%| | 0/210 [00:00<?, ?it/s]Epoch 6: train_loss 0.014834, train_acc 0.998175, test_acc 0.996311, test_loss 0.079170, time 2.7 sec
100%|██████████| 210/210 [00:02<00:00, 80.94it/s]
100%|██████████| 24/24 [00:00<00:00, 232.75it/s]
0%| | 0/210 [00:00<?, ?it/s]Epoch 7: train_loss 0.020133, train_acc 0.997690, test_acc 0.998994, test_loss 0.052059, time 2.7 sec
100%|██████████| 210/210 [00:02<00:00, 80.86it/s]
100%|██████████| 24/24 [00:00<00:00, 230.38it/s]
0%| | 0/210 [00:00<?, ?it/s]Epoch 8: train_loss 0.005914, train_acc 0.999143, test_acc 0.996982, test_loss 0.070334, time 2.7 sec
100%|██████████| 210/210 [00:02<00:00, 80.47it/s]
100%|██████████| 24/24 [00:00<00:00, 231.28it/s]
0%| | 0/210 [00:00<?, ?it/s]Epoch 9: train_loss inf, train_acc 0.993034, test_acc 0.996311, test_loss inf, time 2.7 sec
100%|██████████| 210/210 [00:02<00:00, 80.33it/s]
100%|██████████| 24/24 [00:00<00:00, 225.85it/s]
Epoch 10: train_loss 0.014424, train_acc 0.998510, test_acc 0.996647, test_loss 0.083575, time 2.7 sec
100%|██████████| 210/210 [00:02<00:00, 75.68it/s]
100%|██████████| 24/24 [00:00<00:00, 230.67it/s]
0%| | 0/210 [00:00<?, ?it/s]Epoch 11: train_loss 0.009379, train_acc 0.998845, test_acc 0.998323, test_loss inf, time 2.9 sec
100%|██████████| 210/210 [00:02<00:00, 80.58it/s]
100%|██████████| 24/24 [00:00<00:00, 236.18it/s]
Epoch 12: train_loss 0.012093, train_acc 0.999329, test_acc 0.998659, test_loss 0.054137, time 2.7 sec
100%|██████████| 210/210 [00:02<00:00, 80.45it/s]
100%|██████████| 24/24 [00:00<00:00, 234.54it/s]
0%| | 0/210 [00:00<?, ?it/s]Epoch 13: train_loss 0.012292, train_acc 0.998696, test_acc 0.998994, test_loss 0.059999, time 2.7 sec
100%|██████████| 210/210 [00:02<00:00, 80.10it/s]
100%|██████████| 24/24 [00:00<00:00, 235.41it/s]
0%| | 0/210 [00:00<?, ?it/s]Epoch 14: train_loss inf, train_acc 0.998957, test_acc 0.998994, test_loss 0.048007, time 2.7 sec
100%|██████████| 210/210 [00:02<00:00, 80.03it/s]
100%|██████████| 24/24 [00:00<00:00, 234.61it/s]
0%| | 0/210 [00:00<?, ?it/s]Epoch 15: train_loss 0.015674, train_acc 0.998361, test_acc 0.997653, test_loss inf, time 2.7 sec
100%|██████████| 210/210 [00:02<00:00, 80.68it/s]
100%|██████████| 24/24 [00:00<00:00, 231.76it/s]
0%| | 0/210 [00:00<?, ?it/s]Epoch 16: train_loss inf, train_acc 0.998547, test_acc 0.996647, test_loss inf, time 2.7 sec
100%|██████████| 210/210 [00:02<00:00, 80.11it/s]
100%|██████████| 24/24 [00:00<00:00, 231.54it/s]
Epoch 17: train_loss inf, train_acc 0.983646, test_acc 0.995305, test_loss inf, time 2.7 sec
100%|██████████| 210/210 [00:02<00:00, 78.57it/s]
100%|██████████| 24/24 [00:00<00:00, 231.14it/s]
Epoch 18: train_loss inf, train_acc 0.999031, test_acc 0.997653, test_loss inf, time 2.8 sec
100%|██████████| 210/210 [00:02<00:00, 79.47it/s]
100%|██████████| 24/24 [00:00<00:00, 235.37it/s]
0%| | 0/210 [00:00<?, ?it/s]Epoch 19: train_loss 0.003552, train_acc 0.999702, test_acc 0.997653, test_loss inf, time 2.7 sec
100%|██████████| 210/210 [00:02<00:00, 79.77it/s]
100%|██████████| 24/24 [00:00<00:00, 232.33it/s]
0%| | 0/210 [00:00<?, ?it/s]Epoch 20: train_loss 0.007358, train_acc 0.999665, test_acc 0.998323, test_loss inf, time 2.7 sec
100%|██████████| 210/210 [00:02<00:00, 80.39it/s]
100%|██████████| 24/24 [00:00<00:00, 229.74it/s]
0%| | 0/210 [00:00<?, ?it/s]Epoch 21: train_loss 0.003276, train_acc 0.999739, test_acc 0.998659, test_loss inf, time 2.7 sec
100%|██████████| 210/210 [00:02<00:00, 76.31it/s]
100%|██████████| 24/24 [00:00<00:00, 231.89it/s]
Epoch 22: train_loss 0.002908, train_acc 0.999814, test_acc 0.998323, test_loss inf, time 2.9 sec
100%|██████████| 210/210 [00:02<00:00, 80.48it/s]
100%|██████████| 24/24 [00:00<00:00, 233.76it/s]
0%| | 0/210 [00:00<?, ?it/s]Epoch 23: train_loss inf, train_acc 0.999516, test_acc 0.998323, test_loss inf, time 2.7 sec
100%|██████████| 210/210 [00:02<00:00, 79.76it/s]
100%|██████████| 24/24 [00:00<00:00, 235.64it/s]
Epoch 24: train_loss 0.005402, train_acc 0.999739, test_acc 0.997653, test_loss inf, time 2.7 sec
100%|██████████| 210/210 [00:02<00:00, 79.92it/s]
100%|██████████| 24/24 [00:00<00:00, 233.85it/s]
Epoch 25: train_loss 0.003059, train_acc 0.999665, test_acc 0.997988, test_loss inf, time 2.7 sec
100%|██████████| 210/210 [00:02<00:00, 80.26it/s]
100%|██████████| 24/24 [00:00<00:00, 235.02it/s]
Epoch 26: train_loss 0.000807, train_acc 0.999851, test_acc 0.997988, test_loss inf, time 2.7 sec
100%|██████████| 210/210 [00:02<00:00, 79.98it/s]
100%|██████████| 24/24 [00:00<00:00, 232.45it/s]
Epoch 27: train_loss inf, train_acc 0.999776, test_acc 0.998659, test_loss inf, time 2.7 sec
100%|██████████| 210/210 [00:02<00:00, 80.17it/s]
100%|██████████| 24/24 [00:00<00:00, 235.01it/s]
Epoch 28: train_loss 0.001759, train_acc 0.999925, test_acc 0.998323, test_loss inf, time 2.7 sec
100%|██████████| 210/210 [00:02<00:00, 79.67it/s]
100%|██████████| 24/24 [00:00<00:00, 235.77it/s]
Epoch 29: train_loss 0.003535, train_acc 0.999404, test_acc 0.997988, test_loss inf, time 2.7 sec
100%|██████████| 210/210 [00:02<00:00, 79.96it/s]
100%|██████████| 24/24 [00:00<00:00, 232.70it/s]
0%| | 0/210 [00:00<?, ?it/s]Epoch 30: train_loss 0.008395, train_acc 0.999553, test_acc 0.997988, test_loss inf, time 2.7 sec
100%|██████████| 210/210 [00:02<00:00, 79.63it/s]
100%|██████████| 24/24 [00:00<00:00, 234.90it/s]
0%| | 0/210 [00:00<?, ?it/s]Epoch 31: train_loss 0.000772, train_acc 0.999851, test_acc 0.996982, test_loss inf, time 2.7 sec
100%|██████████| 210/210 [00:02<00:00, 80.36it/s]
100%|██████████| 24/24 [00:00<00:00, 233.44it/s]
Epoch 32: train_loss 0.007225, train_acc 0.999702, test_acc 0.998994, test_loss inf, time 2.7 sec