keras CNN 训练时 loss值为nan & val_acc不变

最新推荐文章于 2024-06-19 15:31:46 发布
最新推荐文章于 2024-06-19 15:31:46 发布
阅读量6.1k 收藏 37
点赞数 6
分类专栏： 默认分类 文章标签： keras cnn loss nan binary_crossentropy
版权声明：本文为博主原创文章，遵循 CC 4.0 BY 版权协议，转载请附上原文出处链接和本声明。
本文链接：https://blog.csdn.net/Drifter_Galaxy/article/details/104004267
版权

初始网络代码（数据处理部分未晒出，传给网络的参数就是x_test,y_test,x_train,y_train）


import tensorflow
import keras
from keras.layers.normalization import BatchNormalization
from keras.utils import np_utils
import numpy as np 
from keras.layers import *
from keras.models import *
from keras.objectives import *

input_shape = (300,36)
batch_size = 25

epochs = 10

model = Sequential()
#一层卷积层，包含了32个卷积核，大小为3*3
model.add(Conv1D(16,16, activation='relu', input_shape=input_shape))
model.add(Conv1D(16,16, activation='relu'))
#一个最大池化层，池化大小为2*2
model.add(MaxPooling1D(pool_size=2))
#遗忘层，遗忘速率为0.25
model.add(Dropout(0.25))
# #添加一个卷积层，包含64个卷积和，每个卷积和仍为3*3
# model.add(Conv1D(64,4, activation='relu'))
# model.add(Conv1D(64,4, activation='relu'))
# #来一个池化层
# model.add(MaxPooling1D(pool_size=2))
# model.add(Dropout(0.25))
#压平层
model.add(Flatten())
#来一个全连接层
model.add(Dense(100, activation='relu'))
#来一个遗忘层
model.add(Dropout(0.5))
#最后为分类层
model.add(Dense(1, activation='relu'))

model.compile(loss='sparse_categorical_crossentropy', optimizer='adam', metrics=['accuracy'])

# model.add(Conv1D(301, 24, strides=2, padding = 'same', activation='relu',input_shape = input_shape))
# model.add(Dropout(0.3))
# #model.add(MaxPooling1D(2))

# model.add(Conv1D(400,12, strides=1 , padding = 'same', activation='relu'))
# model.add(Dropout(0.3))
# #model.add(MaxPooling1D(2))


# model.add(Flatten())
# model.add(Dense(33))
# model.add(Activation ('relu'))
# model.add(Dropout(0.3))
# model.add(Dense(1))

# model.compile(loss='binary_crossentropy',
#               optimizer='adam',
#               metrics=['accuracy'])
model.summary()
model.fit(x_train, y_train,
          batch_size=batch_size,
          epochs=epochs,
          verbose =1,
          validation_data=(x_test, y_test))
print("part 2 over")

运行结果出现俩问题

Train on 6877 samples, validate on 16464 samples
Epoch 1/10
6877/6877 [==============================] - ETA: 1:16 - loss: nan - acc: 0.0000e+ - ETA: 7s - loss: nan - acc: 0.0000e+00 - ETA: 4s - loss: nan - acc: 0.0019   - ETA: 3s - loss: nan - acc: 0.00 - ETA: 2s - loss: nan - acc: 0.00 - ETA: 2s - loss: nan - acc: 0.00 - ETA: 2s - loss: nan - acc: 0.00 - ETA: 1s - loss: nan - acc: 0.00 - ETA: 1s - loss: nan - acc: 0.00 - ETA: 1s - loss: nan - acc: 0.00 - ETA: 1s - loss: nan - acc: 0.00 - ETA: 1s - loss: nan - acc: 0.00 - ETA: 1s - loss: nan - acc: 0.00 - ETA: 1s - loss: nan - acc: 0.00 - ETA: 0s - loss: nan - acc: 0.02 - ETA: 0s - loss: nan - acc: 0.06 - ETA: 0s - loss: nan - acc: 0.09 - ETA: 0s - loss: nan - acc: 0.11 - ETA: 0s - loss: nan - acc: 0.13 - ETA: 0s - loss: nan - acc: 0.15 - ETA: 0s - loss: nan - acc: 0.17 - ETA: 0s - loss: nan - acc: 0.18 - ETA: 0s - loss: nan - acc: 0.20 - ETA: 0s - loss: nan - acc: 0.21 - ETA: 0s - loss: nan - acc: 0.22 - ETA: 0s - loss: nan - acc: 0.23 - ETA: 0s - loss: nan - acc: 0.24 - ETA: 0s - loss: nan - acc: 0.25 - 3s 469us/step - loss: nan - acc: 0.2601 - val_loss: nan - val_acc: 0.5000
Epoch 2/10
6877/6877 [==============================] - ETA: 1s - loss: nan - acc: 0.48 - ETA: 1s - loss: nan - acc: 0.53 - ETA: 1s - loss: nan - acc: 0.52 - ETA: 1s - loss: nan - acc: 0.50 - ETA: 1s - loss: nan - acc: 0.50 - ETA: 1s - loss: nan - acc: 0.50 - ETA: 1s - loss: nan - acc: 0.51 - ETA: 1s - loss: nan - acc: 0.51 - ETA: 1s - loss: nan - acc: 0.51 - ETA: 0s - loss: nan - acc: 0.50 - ETA: 0s - loss: nan - acc: 0.50 - ETA: 0s - loss: nan - acc: 0.50 - ETA: 0s - loss: nan - acc: 0.50 - ETA: 0s - loss: nan - acc: 0.50 - ETA: 0s - loss: nan - acc: 0.49 - ETA: 0s - loss: nan - acc: 0.49 - ETA: 0s - loss: nan - acc: 0.50 - ETA: 0s - loss: nan - acc: 0.49 - ETA: 0s - loss: nan - acc: 0.49 - ETA: 0s - loss: nan - acc: 0.49 - ETA: 0s - loss: nan - acc: 0.49 - ETA: 0s - loss: nan - acc: 0.49 - ETA: 0s - loss: nan - acc: 0.49 - ETA: 0s - loss: nan - acc: 0.49 - ETA: 0s - loss: nan - acc: 0.49 - ETA: 0s - loss: nan - acc: 0.49 - ETA: 0s - loss: nan - acc: 0.50 - ETA: 0s - loss: nan - acc: 0.50 - 3s 428us/step - loss: nan - acc: 0.5001 - val_loss: nan - val_acc: 0.5000
Epoch 3/10
6877/6877 [==============================] - ETA: 1s - loss: nan - acc: 0.52 - ETA: 1s - loss: nan - acc: 0.50 - ETA: 1s - loss: nan - acc: 0.49 - ETA: 1s - loss: nan - acc: 0.48 - ETA: 1s - loss: nan - acc: 0.48 - ETA: 1s - loss: nan - acc: 0.50 - ETA: 1s - loss: nan - acc: 0.51 - ETA: 1s - loss: nan - acc: 0.51 - ETA: 1s - loss: nan - acc: 0.51 - ETA: 0s - loss: nan - acc: 0.50 - ETA: 0s - loss: nan - acc: 0.50 - ETA: 0s - loss: nan - acc: 0.50 - ETA: 0s - loss: nan - acc: 0.50 - ETA: 0s - loss: nan - acc: 0.50 - ETA: 0s - loss: nan - acc: 0.50 - ETA: 0s - loss: nan - acc: 0.49 - ETA: 0s - loss: nan - acc: 0.49 - ETA: 0s - loss: nan - acc: 0.49 - ETA: 0s - loss: nan - acc: 0.49 - ETA: 0s - loss: nan - acc: 0.50 - ETA: 0s - loss: nan - acc: 0.49 - ETA: 0s - loss: nan - acc: 0.49 - ETA: 0s - loss: nan - acc: 0.49 - ETA: 0s - loss: nan - acc: 0.49 - ETA: 0s - loss: nan - acc: 0.49 - ETA: 0s - loss: nan - acc: 0.49 - ETA: 0s - loss: nan - acc: 0.49 - ETA: 0s - loss: nan - acc: 0.49 - 3s 423us/step - loss: nan - acc: 0.5001 - val_loss: nan - val_acc: 0.5000
Epoch 4/10
6877/6877 [==============================] - ETA: 1s - loss: nan - acc: 0.52 - ETA: 1s - loss: nan - acc: 0.53 - ETA: 1s - loss: nan - acc: 0.52 - ETA: 1s - loss: nan - acc: 0.51 - ETA: 1s - loss: nan - acc: 0.50 - ETA: 1s - loss: nan - acc: 0.50 - ETA: 1s - loss: nan - acc: 0.51 - ETA: 1s - loss: nan - acc: 0.50 - ETA: 1s - loss: nan - acc: 0.50 - ETA: 0s - loss: nan - acc: 0.50 - ETA: 0s - loss: nan - acc: 0.51 - ETA: 0s - loss: nan - acc: 0.50 - ETA: 0s - loss: nan - acc: 0.51 - ETA: 0s - loss: nan - acc: 0.50 - ETA: 0s - loss: nan - acc: 0.50 - ETA: 0s - loss: nan - acc: 0.50 - ETA: 0s - loss: nan - acc: 0.50 - ETA: 0s - loss: nan - acc: 0.50 - ETA: 0s - loss: nan - acc: 0.50 - ETA: 0s - loss: nan - acc: 0.49 - ETA: 0s - loss: nan - acc: 0.49 - ETA: 0s - loss: nan - acc: 0.50 - ETA: 0s - loss: nan - acc: 0.50 - ETA: 0s - loss: nan - acc: 0.50 - ETA: 0s - loss: nan - acc: 0.50 - ETA: 0s - loss: nan - acc: 0.50 - ETA: 0s - loss: nan - acc: 0.50 - ETA: 0s - loss: nan - acc: 0.49 - 3s 426us/step - loss: nan - acc: 0.5001 - val_loss: nan - val_acc: 0.5000
Epoch 5/10
6877/6877 [==============================] - ETA: 1s - loss: nan - acc: 0.40 - ETA: 1s - loss: nan - acc: 0.53 - ETA: 1s - loss: nan - acc: 0.48 - ETA: 1s - loss: nan - acc: 0.50 - ETA: 1s - loss: nan - acc: 0.50 - ETA: 1s - loss: nan - acc: 0.50 - ETA: 1s - loss: nan - acc: 0.50 - ETA: 1s - loss: nan - acc: 0.51 - ETA: 1s - loss: nan - acc: 0.51 - ETA: 0s - loss: nan - acc: 0.51 - ETA: 0s - loss: nan - acc: 0.51 - ETA: 0s - loss: nan - acc: 0.51 - ETA: 0s - loss: nan - acc: 0.51 - ETA: 0s - loss: nan - acc: 0.50 - ETA: 0s - loss: nan - acc: 0.50 - ETA: 0s - loss: nan - acc: 0.50 - ETA: 0s - loss: nan - acc: 0.50 - ETA: 0s - loss: nan - acc: 0.50 - ETA: 0s - loss: nan - acc: 0.50 - ETA: 0s - loss: nan - acc: 0.50 - ETA: 0s - loss: nan - acc: 0.50 - ETA: 0s - loss: nan - acc: 0.50 - ETA: 0s - loss: nan - acc: 0.50 - ETA: 0s - loss: nan - acc: 0.50 - ETA: 0s - loss: nan - acc: 0.50 - ETA: 0s - loss: nan - acc: 0.50 - ETA: 0s - loss: nan - acc: 0.50 - ETA: 0s - loss: nan - acc: 0.50 - 3s 423us/step - loss: nan - acc: 0.5001 - val_loss: nan - val_acc: 0.5000
Epoch 6/10
6877/6877 [==============================] - ETA: 1s - loss: nan - acc: 0.60 - ETA: 1s - loss: nan - acc: 0.51 - ETA: 1s - loss: nan - acc: 0.47 - ETA: 1s - loss: nan - acc: 0.49 - ETA: 1s - loss: nan - acc: 0.48 - ETA: 1s - loss: nan - acc: 0.48 - ETA: 1s - loss: nan - acc: 0.49 - ETA: 1s - loss: nan - acc: 0.49 - ETA: 1s - loss: nan - acc: 0.49 - ETA: 0s - loss: nan - acc: 0.49 - ETA: 0s - loss: nan - acc: 0.49 - ETA: 0s - loss: nan - acc: 0.49 - ETA: 0s - loss: nan - acc: 0.49 - ETA: 0s - loss: nan - acc: 0.49 - ETA: 0s - loss: nan - acc: 0.50 - ETA: 0s - loss: nan - acc: 0.50 - ETA: 0s - loss: nan - acc: 0.50 - ETA: 0s - loss: nan - acc: 0.49 - ETA: 0s - loss: nan - acc: 0.49 - ETA: 0s - loss: nan - acc: 0.49 - ETA: 0s - loss: nan - acc: 0.49 - ETA: 0s - loss: nan - acc: 0.50 - ETA: 0s - loss: nan - acc: 0.50 - ETA: 0s - loss: nan - acc: 0.50 - ETA: 0s - loss: nan - acc: 0.50 - ETA: 0s - loss: nan - acc: 0.50 - ETA: 0s - loss: nan - acc: 0.50 - ETA: 0s - loss: nan - acc: 0.50 - 3s 422us/step - loss: nan - acc: 0.5001 - val_loss: nan - val_acc: 0.5000
Epoch 7/10
6877/6877 [==============================] - ETA: 1s - loss: nan - acc: 0.52 - ETA: 1s - loss: nan - acc: 0.52 - ETA: 1s - loss: nan - acc: 0.54 - ETA: 1s - loss: nan - acc: 0.54 - ETA: 1s - loss: nan - acc: 0.52 - ETA: 1s - loss: nan - acc: 0.52 - ETA: 1s - loss: nan - acc: 0.52 - ETA: 1s - loss: nan - acc: 0.51 - ETA: 0s - loss: nan - acc: 0.51 - ETA: 0s - loss: nan - acc: 0.50 - ETA: 0s - loss: nan - acc: 0.50 - ETA: 0s - loss: nan - acc: 0.50 - ETA: 0s - loss: nan - acc: 0.50 - ETA: 0s - loss: nan - acc: 0.51 - ETA: 0s - loss: nan - acc: 0.51 - ETA: 0s - loss: nan - acc: 0.50 - ETA: 0s - loss: nan - acc: 0.50 - ETA: 0s - loss: nan - acc: 0.50 - ETA: 0s - loss: nan - acc: 0.50 - ETA: 0s - loss: nan - acc: 0.50 - ETA: 0s - loss: nan - acc: 0.50 - ETA: 0s - loss: nan - acc: 0.50 - ETA: 0s - loss: nan - acc: 0.50 - ETA: 0s - loss: nan - acc: 0.50 - ETA: 0s - loss: nan - acc: 0.50 - ETA: 0s - loss: nan - acc: 0.50 - ETA: 0s - loss: nan - acc: 0.50 - ETA: 0s - loss: nan - acc: 0.50 - 3s 425us/step - loss: nan - acc: 0.5001 - val_loss: nan - val_acc: 0.5000
Epoch 8/10
6877/6877 [==============================] - ETA: 1s - loss: nan - acc: 0.56 - ETA: 1s - loss: nan - acc: 0.50 - ETA: 1s - loss: nan - acc: 0.47 - ETA: 1s - loss: nan - acc: 0.47 - ETA: 1s - loss: nan - acc: 0.48 - ETA: 1s - loss: nan - acc: 0.48 - ETA: 1s - loss: nan - acc: 0.47 - ETA: 1s - loss: nan - acc: 0.48 - ETA: 1s - loss: nan - acc: 0.48 - ETA: 0s - loss: nan - acc: 0.48 - ETA: 0s - loss: nan - acc: 0.49 - ETA: 0s - loss: nan - acc: 0.49 - ETA: 0s - loss: nan - acc: 0.48 - ETA: 0s - loss: nan - acc: 0.49 - ETA: 0s - loss: nan - acc: 0.49 - ETA: 0s - loss: nan - acc: 0.49 - ETA: 0s - loss: nan - acc: 0.49 - ETA: 0s - loss: nan - acc: 0.49 - ETA: 0s - loss: nan - acc: 0.49 - ETA: 0s - loss: nan - acc: 0.50 - ETA: 0s - loss: nan - acc: 0.49 - ETA: 0s - loss: nan - acc: 0.49 - ETA: 0s - loss: nan - acc: 0.49 - ETA: 0s - loss: nan - acc: 0.49 - ETA: 0s - loss: nan - acc: 0.49 - ETA: 0s - loss: nan - acc: 0.50 - ETA: 0s - loss: nan - acc: 0.49 - ETA: 0s - loss: nan - acc: 0.50 - 3s 418us/step - loss: nan - acc: 0.5001 - val_loss: nan - val_acc: 0.5000
Epoch 9/10
6877/6877 [==============================] - ETA: 1s - loss: nan - acc: 0.52 - ETA: 1s - loss: nan - acc: 0.54 - ETA: 1s - loss: nan - acc: 0.53 - ETA: 1s - loss: nan - acc: 0.52 - ETA: 1s - loss: nan - acc: 0.52 - ETA: 1s - loss: nan - acc: 0.52 - ETA: 1s - loss: nan - acc: 0.52 - ETA: 1s - loss: nan - acc: 0.51 - ETA: 0s - loss: nan - acc: 0.50 - ETA: 0s - loss: nan - acc: 0.50 - ETA: 0s - loss: nan - acc: 0.50 - ETA: 0s - loss: nan - acc: 0.50 - ETA: 0s - loss: nan - acc: 0.49 - ETA: 0s - loss: nan - acc: 0.49 - ETA: 0s - loss: nan - acc: 0.49 - ETA: 0s - loss: nan - acc: 0.49 - ETA: 0s - loss: nan - acc: 0.49 - ETA: 0s - loss: nan - acc: 0.49 - ETA: 0s - loss: nan - acc: 0.49 - ETA: 0s - loss: nan - acc: 0.50 - ETA: 0s - loss: nan - acc: 0.50 - ETA: 0s - loss: nan - acc: 0.50 - ETA: 0s - loss: nan - acc: 0.49 - ETA: 0s - loss: nan - acc: 0.49 - ETA: 0s - loss: nan - acc: 0.49 - ETA: 0s - loss: nan - acc: 0.50 - ETA: 0s - loss: nan - acc: 0.49 - 3s 417us/step - loss: nan - acc: 0.5001 - val_loss: nan - val_acc: 0.5000
Epoch 10/10
6877/6877 [==============================] - ETA: 1s - loss: nan - acc: 0.56 - ETA: 1s - loss: nan - acc: 0.50 - ETA: 1s - loss: nan - acc: 0.51 - ETA: 1s - loss: nan - acc: 0.51 - ETA: 1s - loss: nan - acc: 0.51 - ETA: 1s - loss: nan - acc: 0.50 - ETA: 1s - loss: nan - acc: 0.50 - ETA: 0s - loss: nan - acc: 0.50 - ETA: 0s - loss: nan - acc: 0.49 - ETA: 0s - loss: nan - acc: 0.49 - ETA: 0s - loss: nan - acc: 0.48 - ETA: 0s - loss: nan - acc: 0.49 - ETA: 0s - loss: nan - acc: 0.49 - ETA: 0s - loss: nan - acc: 0.49 - ETA: 0s - loss: nan - acc: 0.49 - ETA: 0s - loss: nan - acc: 0.49 - ETA: 0s - loss: nan - acc: 0.50 - ETA: 0s - loss: nan - acc: 0.50 - ETA: 0s - loss: nan - acc: 0.50 - ETA: 0s - loss: nan - acc: 0.50 - ETA: 0s - loss: nan - acc: 0.50 - ETA: 0s - loss: nan - acc: 0.50 - ETA: 0s - loss: nan - acc: 0.50 - ETA: 0s - loss: nan - acc: 0.50 - ETA: 0s - loss: nan - acc: 0.50 - ETA: 0s - loss: nan - acc: 0.50 - ETA: 0s - loss: nan - acc: 0.49 - 3s 419us/step - loss: nan - acc: 0.5001 - val_loss: nan - val_acc: 0.5000

 

1、loss值为nan

有两种可能：

1. 梯度爆炸，解决方法：降低学习率、梯度剪裁、归一化

2. 计算loss的时候有log0，可能是初始化的问题，也可能是数据的问题

我主要讲第二种情况，当loss函数为 categorical_crossentropy 经常会出现这种情况，

网上大多数的解决办法是：

在pycharm中双击shift键，寻找该函数，会出现keras.loss模块中有该函数，进入该函数后，

原函数为：

def categorical_crossentropy(y_true, y_pred):
    return K.categorical_crossentropy(y_true, y_pred)

修改为：
def categorical_crossentropy(y_true, y_pred):
    return K.categorical_crossentropy(y_true, y_pred+1e-5)

我的办法：

在我的CNN网络中是 loss='sparse_categorical_crossentropy'，将其更换为binary_crossentropy, nan问题解决！

此时运行结果是这样的：

Train on 6877 samples, validate on 16464 samples
Epoch 1/10
6877/6877 [==============================] - ETA: 3:05 - loss: 10.2995 - acc: 0.36 - ETA: 19s - loss: 8.2500 - acc: 0.1000 - ETA: 10s - loss: 8.1701 - acc: 0.06 - ETA: 7s - loss: 8.0500 - acc: 0.0486 - ETA: 5s - loss: 8.1273 - acc: 0.038 - ETA: 4s - loss: 8.2305 - acc: 0.039 - ETA: 3s - loss: 8.0206 - acc: 0.086 - ETA: 3s - loss: 8.0284 - acc: 0.124 - ETA: 2s - loss: 8.0805 - acc: 0.139 - ETA: 2s - loss: 8.0746 - acc: 0.135 - ETA: 2s - loss: 8.0856 - acc: 0.127 - ETA: 2s - loss: 8.0500 - acc: 0.115 - ETA: 1s - loss: 8.0767 - acc: 0.104 - ETA: 1s - loss: 8.0508 - acc: 0.097 - ETA: 1s - loss: 8.0478 - acc: 0.090 - ETA: 1s - loss: 8.0135 - acc: 0.084 - ETA: 1s - loss: 8.0318 - acc: 0.079 - ETA: 1s - loss: 8.0201 - acc: 0.074 - ETA: 0s - loss: 8.0210 - acc: 0.069 - ETA: 0s - loss: 8.0183 - acc: 0.066 - ETA: 0s - loss: 8.0191 - acc: 0.062 - ETA: 0s - loss: 8.0090 - acc: 0.059 - ETA: 0s - loss: 7.9662 - acc: 0.057 - ETA: 0s - loss: 7.9942 - acc: 0.054 - ETA: 0s - loss: 7.9919 - acc: 0.052 - ETA: 0s - loss: 7.9883 - acc: 0.050 - ETA: 0s - loss: 7.9928 - acc: 0.048 - ETA: 0s - loss: 7.9994 - acc: 0.046 - ETA: 0s - loss: 8.0428 - acc: 0.045 - ETA: 0s - loss: 8.0509 - acc: 0.043 - 4s 546us/step - loss: 8.0508 - acc: 0.0435 - val_loss: 7.9712 - val_acc: 0.0000e+00
Epoch 2/10
6877/6877 [==============================] - ETA: 1s - loss: 8.9277 - acc: 0.0000e+0 - ETA: 1s - loss: 8.2321 - acc: 0.0000e+0 - ETA: 1s - loss: 7.9256 - acc: 0.0000e+0 - ETA: 1s - loss: 8.3723 - acc: 0.0000e+0 - ETA: 1s - loss: 8.1501 - acc: 0.0000e+0 - ETA: 1s - loss: 8.1650 - acc: 0.0000e+0 - ETA: 1s - loss: 8.0601 - acc: 0.0000e+0 - ETA: 1s - loss: 7.9218 - acc: 0.0000e+0 - ETA: 1s - loss: 8.0066 - acc: 0.0000e+0 - ETA: 0s - loss: 7.9186 - acc: 0.0000e+0 - ETA: 0s - loss: 7.8670 - acc: 0.0000e+0 - ETA: 0s - loss: 7.8879 - acc: 0.0000e+0 - ETA: 0s - loss: 7.8842 - acc: 0.0000e+0 - ETA: 0s - loss: 7.9249 - acc: 0.0000e+0 - ETA: 0s - loss: 7.9192 - acc: 0.0000e+0 - ETA: 0s - loss: 7.9480 - acc: 0.0000e+0 - ETA: 0s - loss: 7.9653 - acc: 0.0000e+0 - ETA: 0s - loss: 7.9619 - acc: 0.0000e+0 - ETA: 0s - loss: 7.9870 - acc: 0.0000e+0 - ETA: 0s - loss: 7.9795 - acc: 0.0000e+0 - ETA: 0s - loss: 7.9918 - acc: 0.0000e+0 - ETA: 0s - loss: 7.9606 - acc: 0.0000e+0 - ETA: 0s - loss: 7.9697 - acc: 0.0000e+0 - ETA: 0s - loss: 7.9296 - acc: 0.0000e+0 - ETA: 0s - loss: 7.9606 - acc: 0.0000e+0 - ETA: 0s - loss: 7.9788 - acc: 0.0000e+0 - ETA: 0s - loss: 7.9908 - acc: 0.0000e+0 - ETA: 0s - loss: 7.9854 - acc: 0.0000e+0 - 3s 437us/step - loss: 7.9724 - acc: 0.0000e+00 - val_loss: 7.9712 - val_acc: 0.0000e+00
Epoch 3/10
6877/6877 [==============================] - ETA: 1s - loss: 8.2900 - acc: 0.0000e+0 - ETA: 1s - loss: 7.7683 - acc: 0.0000e+0 - ETA: 1s - loss: 7.5005 - acc: 0.0000e+0 - ETA: 1s - loss: 7.7552 - acc: 0.0000e+0 - ETA: 1s - loss: 7.9393 - acc: 0.0000e+0 - ETA: 1s - loss: 7.9584 - acc: 0.0000e+0 - ETA: 1s - loss: 7.8901 - acc: 0.0000e+0 - ETA: 1s - loss: 7.9481 - acc: 0.0000e+0 - ETA: 1s - loss: 7.8945 - acc: 0.0000e+0 - ETA: 1s - loss: 7.8530 - acc: 0.0000e+0 - ETA: 0s - loss: 7.8606 - acc: 0.0000e+0 - ETA: 0s - loss: 7.8118 - acc: 0.0000e+0 - ETA: 0s - loss: 7.7995 - acc: 0.0000e+0 - ETA: 0s - loss: 7.8833 - acc: 0.0000e+0 - ETA: 0s - loss: 7.8572 - acc: 0.0000e+0 - ETA: 0s - loss: 7.9083 - acc: 0.0000e+0 - ETA: 0s - loss: 7.9245 - acc: 0.0000e+0 - ETA: 0s - loss: 7.8675 - acc: 0.0000e+0 - ETA: 0s - loss: 7.8734 - acc: 0.0000e+0 - ETA: 0s - loss: 7.8683 - acc: 0.0000e+0 - ETA: 0s - loss: 7.8638 - acc: 0.0000e+0 - ETA: 0s - loss: 7.8752 - acc: 0.0000e+0 - ETA: 0s - loss: 7.8856 - acc: 0.0000e+0 - ETA: 0s - loss: 7.9514 - acc: 0.0000e+0 - ETA: 0s - loss: 7.9928 - acc: 0.0000e+0 - ETA: 0s - loss: 8.0115 - acc: 0.0000e+0 - ETA: 0s - loss: 8.0000 - acc: 0.0000e+0 - ETA: 0s - loss: 7.9796 - acc: 0.0000e+0 - ETA: 0s - loss: 7.9724 - acc: 0.0000e+0 - 3s 440us/step - loss: 7.9724 - acc: 0.0000e+00 - val_loss: 7.9712 - val_acc: 0.0000e+00
Epoch 4/10
6877/6877 [==============================] - ETA: 1s - loss: 8.9277 - acc: 0.0000e+0 - ETA: 1s - loss: 8.1741 - acc: 0.0000e+0 - ETA: 1s - loss: 8.1989 - acc: 0.0000e+0 - ETA: 1s - loss: 8.2489 - acc: 0.0000e+0 - ETA: 1s - loss: 8.1190 - acc: 0.0000e+0 - ETA: 1s - loss: 7.9649 - acc: 0.0000e+0 - ETA: 1s - loss: 8.1228 - acc: 0.0000e+0 - ETA: 1s - loss: 8.1643 - acc: 0.0000e+0 - ETA: 1s - loss: 8.1090 - acc: 0.0000e+0 - ETA: 1s - loss: 8.0938 - acc: 0.0000e+0 - ETA: 0s - loss: 8.0943 - acc: 0.0000e+0 - ETA: 0s - loss: 8.0373 - acc: 0.0000e+0 - ETA: 0s - loss: 8.0529 - acc: 0.0000e+0 - ETA: 0s - loss: 8.0466 - acc: 0.0000e+0 - ETA: 0s - loss: 8.0232 - acc: 0.0000e+0 - ETA: 0s - loss: 8.0662 - acc: 0.0000e+0 - ETA: 0s - loss: 8.0484 - acc: 0.0000e+0 - ETA: 0s - loss: 7.9954 - acc: 0.0000e+0 - ETA: 0s - loss: 8.0011 - acc: 0.0000e+0 - ETA: 0s - loss: 8.0697 - acc: 0.0000e+0 - ETA: 0s - loss: 8.0584 - acc: 0.0000e+0 - ETA: 0s - loss: 7.9908 - acc: 0.0000e+0 - ETA: 0s - loss: 7.9813 - acc: 0.0000e+0 - ETA: 0s - loss: 7.9809 - acc: 0.0000e+0 - ETA: 0s - loss: 7.9831 - acc: 0.0000e+0 - ETA: 0s - loss: 7.9979 - acc: 0.0000e+0 - ETA: 0s - loss: 7.9993 - acc: 0.0000e+0 - ETA: 0s - loss: 7.9818 - acc: 0.0000e+0 - 3s 439us/step - loss: 7.9724 - acc: 0.0000e+00 - val_loss: 7.9712 - val_acc: 0.0000e+00
Epoch 5/10
6877/6877 [==============================] - ETA: 1s - loss: 6.3770 - acc: 0.0000e+0 - ETA: 1s - loss: 8.5799 - acc: 0.0000e+0 - ETA: 1s - loss: 8.3508 - acc: 0.0000e+0 - ETA: 1s - loss: 8.2695 - acc: 0.0000e+0 - ETA: 1s - loss: 8.3211 - acc: 0.0000e+0 - ETA: 1s - loss: 8.2525 - acc: 0.0000e+0 - ETA: 1s - loss: 8.2273 - acc: 0.0000e+0 - ETA: 1s - loss: 8.1374 - acc: 0.0000e+0 - ETA: 1s - loss: 8.1168 - acc: 0.0000e+0 - ETA: 0s - loss: 8.1078 - acc: 0.0000e+0 - ETA: 0s - loss: 8.0501 - acc: 0.0000e+0 - ETA: 0s - loss: 8.0639 - acc: 0.0000e+0 - ETA: 0s - loss: 7.9021 - acc: 0.0000e+0 - ETA: 0s - loss: 7.8731 - acc: 0.0000e+0 - ETA: 0s - loss: 7.8634 - acc: 0.0000e+0 - ETA: 0s - loss: 7.8877 - acc: 0.0000e+0 - ETA: 0s - loss: 7.9211 - acc: 0.0000e+0 - ETA: 0s - loss: 7.8976 - acc: 0.0000e+0 - ETA: 0s - loss: 7.9088 - acc: 0.0000e+0 - ETA: 0s - loss: 7.9256 - acc: 0.0000e+0 - ETA: 0s - loss: 7.9087 - acc: 0.0000e+0 - ETA: 0s - loss: 7.9178 - acc: 0.0000e+0 - ETA: 0s - loss: 7.9406 - acc: 0.0000e+0 - ETA: 0s - loss: 7.9376 - acc: 0.0000e+0 - ETA: 0s - loss: 7.9524 - acc: 0.0000e+0 - ETA: 0s - loss: 7.9609 - acc: 0.0000e+0 - ETA: 0s - loss: 7.9687 - acc: 0.0000e+0 - ETA: 0s - loss: 7.9641 - acc: 0.0000e+0 - 3s 435us/step - loss: 7.9724 - acc: 0.0000e+00 - val_loss: 7.9712 - val_acc: 0.0000e+00
Epoch 6/10
6877/6877 [==============================] - ETA: 1s - loss: 5.7393 - acc: 0.0000e+0 - ETA: 1s - loss: 8.0582 - acc: 0.0000e+0 - ETA: 1s - loss: 8.0775 - acc: 0.0000e+0 - ETA: 1s - loss: 7.9198 - acc: 0.0000e+0 - ETA: 1s - loss: 8.0101 - acc: 0.0000e+0 - ETA: 1s - loss: 7.8899 - acc: 0.0000e+0 - ETA: 1s - loss: 7.8719 - acc: 0.0000e+0 - ETA: 1s - loss: 7.8769 - acc: 0.0000e+0 - ETA: 1s - loss: 7.9830 - acc: 0.0000e+0 - ETA: 0s - loss: 7.9537 - acc: 0.0000e+0 - ETA: 0s - loss: 8.0059 - acc: 0.0000e+0 - ETA: 0s - loss: 7.9913 - acc: 0.0000e+0 - ETA: 0s - loss: 8.0002 - acc: 0.0000e+0 - ETA: 0s - loss: 7.9347 - acc: 0.0000e+0 - ETA: 0s - loss: 7.9735 - acc: 0.0000e+0 - ETA: 0s - loss: 8.0198 - acc: 0.0000e+0 - ETA: 0s - loss: 8.0326 - acc: 0.0000e+0 - ETA: 0s - loss: 8.0125 - acc: 0.0000e+0 - ETA: 0s - loss: 7.9818 - acc: 0.0000e+0 - ETA: 0s - loss: 7.9511 - acc: 0.0000e+0 - ETA: 0s - loss: 7.9202 - acc: 0.0000e+0 - ETA: 0s - loss: 7.9287 - acc: 0.0000e+0 - ETA: 0s - loss: 7.9451 - acc: 0.0000e+0 - ETA: 0s - loss: 7.9906 - acc: 0.0000e+0 - ETA: 0s - loss: 7.9659 - acc: 0.0000e+0 - ETA: 0s - loss: 7.9788 - acc: 0.0000e+0 - ETA: 0s - loss: 7.9712 - acc: 0.0000e+0 - ETA: 0s - loss: 7.9523 - acc: 0.0000e+0 - 3s 437us/step - loss: 7.9724 - acc: 0.0000e+00 - val_loss: 7.9712 - val_acc: 0.0000e+00
Epoch 7/10
6877/6877 [==============================] - ETA: 1s - loss: 10.8408 - acc: 0.0000e+ - ETA: 1s - loss: 8.8698 - acc: 0.0000e+00 - ETA: 1s - loss: 8.3204 - acc: 0.0000e+0 - ETA: 1s - loss: 8.2263 - acc: 0.0000e+0 - ETA: 1s - loss: 8.3718 - acc: 0.0000e+0 - ETA: 1s - loss: 8.3551 - acc: 0.0000e+0 - ETA: 1s - loss: 8.3008 - acc: 0.0000e+0 - ETA: 1s - loss: 8.2069 - acc: 0.0000e+0 - ETA: 1s - loss: 8.1528 - acc: 0.0000e+0 - ETA: 1s - loss: 8.1324 - acc: 0.0000e+0 - ETA: 0s - loss: 8.1354 - acc: 0.0000e+0 - ETA: 0s - loss: 8.1555 - acc: 0.0000e+0 - ETA: 0s - loss: 8.0810 - acc: 0.0000e+0 - ETA: 0s - loss: 8.0972 - acc: 0.0000e+0 - ETA: 0s - loss: 8.0240 - acc: 0.0000e+0 - ETA: 0s - loss: 7.9776 - acc: 0.0000e+0 - ETA: 0s - loss: 7.9973 - acc: 0.0000e+0 - ETA: 0s - loss: 8.0070 - acc: 0.0000e+0 - ETA: 0s - loss: 7.9837 - acc: 0.0000e+0 - ETA: 0s - loss: 7.9796 - acc: 0.0000e+0 - ETA: 0s - loss: 7.9920 - acc: 0.0000e+0 - ETA: 0s - loss: 7.9666 - acc: 0.0000e+0 - ETA: 0s - loss: 7.9668 - acc: 0.0000e+0 - ETA: 0s - loss: 7.9628 - acc: 0.0000e+0 - ETA: 0s - loss: 7.9605 - acc: 0.0000e+0 - ETA: 0s - loss: 7.9583 - acc: 0.0000e+0 - ETA: 0s - loss: 7.9687 - acc: 0.0000e+0 - ETA: 0s - loss: 7.9807 - acc: 0.0000e+0 - 3s 436us/step - loss: 7.9724 - acc: 0.0000e+00 - val_loss: 7.9712 - val_acc: 0.0000e+00
Epoch 8/10
6877/6877 [==============================] - ETA: 1s - loss: 8.9277 - acc: 0.0000e+0 - ETA: 1s - loss: 7.5364 - acc: 0.0000e+0 - ETA: 1s - loss: 8.0167 - acc: 0.0000e+0 - ETA: 1s - loss: 7.8169 - acc: 0.0000e+0 - ETA: 1s - loss: 7.8856 - acc: 0.0000e+0 - ETA: 1s - loss: 8.0525 - acc: 0.0000e+0 - ETA: 1s - loss: 8.2369 - acc: 0.0000e+0 - ETA: 1s - loss: 8.2994 - acc: 0.0000e+0 - ETA: 1s - loss: 8.1989 - acc: 0.0000e+0 - ETA: 1s - loss: 8.1215 - acc: 0.0000e+0 - ETA: 0s - loss: 8.1717 - acc: 0.0000e+0 - ETA: 0s - loss: 8.2066 - acc: 0.0000e+0 - ETA: 0s - loss: 8.2028 - acc: 0.0000e+0 - ETA: 0s - loss: 8.1043 - acc: 0.0000e+0 - ETA: 0s - loss: 8.1457 - acc: 0.0000e+0 - ETA: 0s - loss: 8.0818 - acc: 0.0000e+0 - ETA: 0s - loss: 8.0504 - acc: 0.0000e+0 - ETA: 0s - loss: 7.9922 - acc: 0.0000e+0 - ETA: 0s - loss: 8.0378 - acc: 0.0000e+0 - ETA: 0s - loss: 7.9952 - acc: 0.0000e+0 - ETA: 0s - loss: 8.0153 - acc: 0.0000e+0 - ETA: 0s - loss: 7.9759 - acc: 0.0000e+0 - ETA: 0s - loss: 7.9578 - acc: 0.0000e+0 - ETA: 0s - loss: 7.9641 - acc: 0.0000e+0 - ETA: 0s - loss: 7.9861 - acc: 0.0000e+0 - ETA: 0s - loss: 7.9621 - acc: 0.0000e+0 - ETA: 0s - loss: 7.9549 - acc: 0.0000e+0 - ETA: 0s - loss: 7.9724 - acc: 0.0000e+0 - ETA: 0s - loss: 7.9712 - acc: 0.0000e+0 - 3s 439us/step - loss: 7.9724 - acc: 0.0000e+00 - val_loss: 7.9712 - val_acc: 0.0000e+00
Epoch 9/10
6877/6877 [==============================] - ETA: 1s - loss: 11.4785 - acc: 0.0000e+ - ETA: 1s - loss: 8.1625 - acc: 0.0000e+00 - ETA: 1s - loss: 7.4929 - acc: 0.0000e+0 - ETA: 1s - loss: 7.6098 - acc: 0.0000e+0 - ETA: 1s - loss: 7.6205 - acc: 0.0000e+0 - ETA: 1s - loss: 7.6141 - acc: 0.0000e+0 - ETA: 1s - loss: 7.6842 - acc: 0.0000e+0 - ETA: 1s - loss: 7.8983 - acc: 0.0000e+0 - ETA: 1s - loss: 7.9393 - acc: 0.0000e+0 - ETA: 1s - loss: 7.9358 - acc: 0.0000e+0 - ETA: 0s - loss: 7.9839 - acc: 0.0000e+0 - ETA: 0s - loss: 8.0002 - acc: 0.0000e+0 - ETA: 0s - loss: 7.9606 - acc: 0.0000e+0 - ETA: 0s - loss: 7.9908 - acc: 0.0000e+0 - ETA: 0s - loss: 7.9712 - acc: 0.0000e+0 - ETA: 0s - loss: 8.0222 - acc: 0.0000e+0 - ETA: 0s - loss: 8.0031 - acc: 0.0000e+0 - ETA: 0s - loss: 7.9824 - acc: 0.0000e+0 - ETA: 0s - loss: 7.9181 - acc: 0.0000e+0 - ETA: 0s - loss: 7.9358 - acc: 0.0000e+0 - ETA: 0s - loss: 7.9615 - acc: 0.0000e+0 - ETA: 0s - loss: 7.9666 - acc: 0.0000e+0 - ETA: 0s - loss: 7.9991 - acc: 0.0000e+0 - ETA: 0s - loss: 7.9895 - acc: 0.0000e+0 - ETA: 0s - loss: 7.9968 - acc: 0.0000e+0 - ETA: 0s - loss: 7.9544 - acc: 0.0000e+0 - ETA: 0s - loss: 7.9476 - acc: 0.0000e+0 - ETA: 0s - loss: 7.9616 - acc: 0.0000e+0 - 3s 439us/step - loss: 7.9724 - acc: 0.0000e+00 - val_loss: 7.9712 - val_acc: 0.0000e+00
Epoch 10/10
6877/6877 [==============================] - ETA: 1s - loss: 6.3770 - acc: 0.0000e+0 - ETA: 1s - loss: 7.4205 - acc: 0.0000e+0 - ETA: 1s - loss: 7.9560 - acc: 0.0000e+0 - ETA: 1s - loss: 7.9403 - acc: 0.0000e+0 - ETA: 1s - loss: 7.9945 - acc: 0.0000e+0 - ETA: 1s - loss: 7.8274 - acc: 0.0000e+0 - ETA: 1s - loss: 7.7151 - acc: 0.0000e+0 - ETA: 1s - loss: 7.6973 - acc: 0.0000e+0 - ETA: 1s - loss: 7.6838 - acc: 0.0000e+0 - ETA: 0s - loss: 7.7785 - acc: 0.0000e+0 - ETA: 0s - loss: 7.7863 - acc: 0.0000e+0 - ETA: 0s - loss: 7.8279 - acc: 0.0000e+0 - ETA: 0s - loss: 7.7917 - acc: 0.0000e+0 - ETA: 0s - loss: 7.8303 - acc: 0.0000e+0 - ETA: 0s - loss: 7.8726 - acc: 0.0000e+0 - ETA: 0s - loss: 7.9305 - acc: 0.0000e+0 - ETA: 0s - loss: 7.9451 - acc: 0.0000e+0 - ETA: 0s - loss: 7.9542 - acc: 0.0000e+0 - ETA: 0s - loss: 7.9160 - acc: 0.0000e+0 - ETA: 0s - loss: 7.8919 - acc: 0.0000e+0 - ETA: 0s - loss: 7.8799 - acc: 0.0000e+0 - ETA: 0s - loss: 7.8964 - acc: 0.0000e+0 - ETA: 0s - loss: 7.9173 - acc: 0.0000e+0 - ETA: 0s - loss: 7.9587 - acc: 0.0000e+0 - ETA: 0s - loss: 7.9592 - acc: 0.0000e+0 - ETA: 0s - loss: 7.9545 - acc: 0.0000e+0 - ETA: 0s - loss: 7.9453 - acc: 0.0000e+0 - ETA: 0s - loss: 7.9700 - acc: 0.0000e+0 - 3s 438us/step - loss: 7.9724 - acc: 0.0000e+00 - val_loss: 7.9712 - val_acc: 0.0000e+00

此时出现了第二个问题：

2、acc和loss波动，但是val_acc不变（我之前调试的时候val_acc一直是0.5,重现不了了...）

这个时候从这篇博客得到了启发，上图中,Dense层之后的激活层使用relu函数.我使用binary_crossentropy作为损失函数,使用Adam作为编译keras模型的优化器。问题在这里：

softmax激活确保输出的总和为1.它有助于确保只输出许多类中的一个类.

由于你只有1个输出(只有一个类),这当然是个坏主意.对于所有样本,您可能最终得到1.

（Dense层之后的激活层使用relu函数）->  改用sigmoid.它与binary_crossentropy相处得很好.

于是我将网络中所有激活函数都换成了sigmoid

此时运行结果偏向了正常输出：

Train on 6877 samples, validate on 16464 samples
Epoch 1/10
6877/6877 [==============================] - ETA: 4:33 - loss: 6.7649 - acc: 0.480 - ETA: 31s - loss: 7.1418 - acc: 0.506 - ETA: 15s - loss: 7.7072 - acc: 0.47 - ETA: 10s - loss: 7.2519 - acc: 0.50 - ETA: 8s - loss: 7.2190 - acc: 0.5122 - ETA: 6s - loss: 7.4603 - acc: 0.504 - ETA: 5s - loss: 7.4128 - acc: 0.507 - ETA: 4s - loss: 7.3817 - acc: 0.510 - ETA: 4s - loss: 7.2286 - acc: 0.520 - ETA: 3s - loss: 6.9389 - acc: 0.536 - ETA: 3s - loss: 6.6594 - acc: 0.554 - ETA: 2s - loss: 6.3475 - acc: 0.574 - ETA: 2s - loss: 6.0860 - acc: 0.591 - ETA: 2s - loss: 5.8544 - acc: 0.606 - ETA: 2s - loss: 5.6991 - acc: 0.616 - ETA: 1s - loss: 5.6773 - acc: 0.619 - ETA: 1s - loss: 5.5422 - acc: 0.629 - ETA: 1s - loss: 5.5123 - acc: 0.631 - ETA: 1s - loss: 5.4530 - acc: 0.636 - ETA: 1s - loss: 5.3710 - acc: 0.641 - ETA: 1s - loss: 5.2486 - acc: 0.650 - ETA: 0s - loss: 5.2133 - acc: 0.652 - ETA: 0s - loss: 5.1799 - acc: 0.654 - ETA: 0s - loss: 5.1696 - acc: 0.655 - ETA: 0s - loss: 5.1072 - acc: 0.658 - ETA: 0s - loss: 5.0458 - acc: 0.662 - ETA: 0s - loss: 4.9425 - acc: 0.668 - ETA: 0s - loss: 4.8641 - acc: 0.672 - ETA: 0s - loss: 4.7479 - acc: 0.676 - ETA: 0s - loss: 4.6099 - acc: 0.674 - ETA: 0s - loss: 4.4781 - acc: 0.673 - 4s 638us/step - loss: 4.4080 - acc: 0.6699 - val_loss: 0.7621 - val_acc: 0.4636
Epoch 2/10
6877/6877 [==============================] - ETA: 1s - loss: 0.7424 - acc: 0.560 - ETA: 1s - loss: 0.6807 - acc: 0.552 - ETA: 1s - loss: 0.6852 - acc: 0.545 - ETA: 1s - loss: 0.6680 - acc: 0.585 - ETA: 1s - loss: 0.6833 - acc: 0.620 - ETA: 1s - loss: 0.6702 - acc: 0.627 - ETA: 1s - loss: 0.6632 - acc: 0.624 - ETA: 1s - loss: 0.6645 - acc: 0.619 - ETA: 1s - loss: 0.6607 - acc: 0.614 - ETA: 1s - loss: 0.6703 - acc: 0.611 - ETA: 1s - loss: 0.6712 - acc: 0.603 - ETA: 1s - loss: 0.6724 - acc: 0.598 - ETA: 0s - loss: 0.6738 - acc: 0.593 - ETA: 0s - loss: 0.6757 - acc: 0.584 - ETA: 0s - loss: 0.6766 - acc: 0.580 - ETA: 0s - loss: 0.6776 - acc: 0.573 - ETA: 0s - loss: 0.6776 - acc: 0.574 - ETA: 0s - loss: 0.6780 - acc: 0.571 - ETA: 0s - loss: 0.6785 - acc: 0.569 - ETA: 0s - loss: 0.6766 - acc: 0.572 - ETA: 0s - loss: 0.6774 - acc: 0.578 - ETA: 0s - loss: 0.6758 - acc: 0.581 - ETA: 0s - loss: 0.6748 - acc: 0.587 - ETA: 0s - loss: 0.6749 - acc: 0.592 - ETA: 0s - loss: 0.6718 - acc: 0.597 - ETA: 0s - loss: 0.6703 - acc: 0.601 - ETA: 0s - loss: 0.6686 - acc: 0.602 - ETA: 0s - loss: 0.6663 - acc: 0.605 - ETA: 0s - loss: 0.6667 - acc: 0.608 - ETA: 0s - loss: 0.6640 - acc: 0.611 - ETA: 0s - loss: 0.6624 - acc: 0.613 - 3s 470us/step - loss: 0.6628 - acc: 0.6145 - val_loss: 0.9911 - val_acc: 0.4840
Epoch 3/10
6877/6877 [==============================] - ETA: 1s - loss: 0.5304 - acc: 0.640 - ETA: 1s - loss: 0.5601 - acc: 0.684 - ETA: 1s - loss: 0.6144 - acc: 0.671 - ETA: 1s - loss: 0.6165 - acc: 0.654 - ETA: 1s - loss: 0.6074 - acc: 0.670 - ETA: 1s - loss: 0.6062 - acc: 0.673 - ETA: 1s - loss: 0.6063 - acc: 0.675 - ETA: 1s - loss: 0.5971 - acc: 0.685 - ETA: 1s - loss: 0.6012 - acc: 0.680 - ETA: 1s - loss: 0.5959 - acc: 0.687 - ETA: 1s - loss: 0.6013 - acc: 0.687 - ETA: 1s - loss: 0.6016 - acc: 0.688 - ETA: 0s - loss: 0.6000 - acc: 0.691 - ETA: 0s - loss: 0.5937 - acc: 0.696 - ETA: 0s - loss: 0.5918 - acc: 0.702 - ETA: 0s - loss: 0.5947 - acc: 0.699 - ETA: 0s - loss: 0.5924 - acc: 0.699 - ETA: 0s - loss: 0.5944 - acc: 0.697 - ETA: 0s - loss: 0.5998 - acc: 0.689 - ETA: 0s - loss: 0.5999 - acc: 0.687 - ETA: 0s - loss: 0.5975 - acc: 0.691 - ETA: 0s - loss: 0.5963 - acc: 0.692 - ETA: 0s - loss: 0.5982 - acc: 0.690 - ETA: 0s - loss: 0.5968 - acc: 0.690 - ETA: 0s - loss: 0.5994 - acc: 0.688 - ETA: 0s - loss: 0.6014 - acc: 0.685 - ETA: 0s - loss: 0.6037 - acc: 0.682 - ETA: 0s - loss: 0.6040 - acc: 0.680 - ETA: 0s - loss: 0.6039 - acc: 0.682 - ETA: 0s - loss: 0.6013 - acc: 0.684 - ETA: 0s - loss: 0.5981 - acc: 0.686 - 3s 475us/step - loss: 0.5988 - acc: 0.6866 - val_loss: 0.8766 - val_acc: 0.5480
Epoch 4/10
6877/6877 [==============================] - ETA: 1s - loss: 0.5636 - acc: 0.720 - ETA: 1s - loss: 0.6279 - acc: 0.672 - ETA: 1s - loss: 0.5901 - acc: 0.709 - ETA: 1s - loss: 0.5690 - acc: 0.728 - ETA: 1s - loss: 0.5498 - acc: 0.744 - ETA: 1s - loss: 0.5385 - acc: 0.758 - ETA: 1s - loss: 0.5313 - acc: 0.757 - ETA: 1s - loss: 0.5296 - acc: 0.758 - ETA: 1s - loss: 0.5243 - acc: 0.761 - ETA: 1s - loss: 0.5194 - acc: 0.763 - ETA: 1s - loss: 0.5142 - acc: 0.766 - ETA: 1s - loss: 0.5135 - acc: 0.768 - ETA: 0s - loss: 0.5086 - acc: 0.771 - ETA: 0s - loss: 0.5048 - acc: 0.774 - ETA: 0s - loss: 0.5072 - acc: 0.777 - ETA: 0s - loss: 0.5050 - acc: 0.778 - ETA: 0s - loss: 0.5053 - acc: 0.779 - ETA: 0s - loss: 0.5018 - acc: 0.778 - ETA: 0s - loss: 0.5002 - acc: 0.781 - ETA: 0s - loss: 0.4976 - acc: 0.781 - ETA: 0s - loss: 0.4973 - acc: 0.782 - ETA: 0s - loss: 0.4929 - acc: 0.782 - ETA: 0s - loss: 0.4898 - acc: 0.783 - ETA: 0s - loss: 0.4876 - acc: 0.786 - ETA: 0s - loss: 0.4854 - acc: 0.788 - ETA: 0s - loss: 0.4828 - acc: 0.790 - ETA: 0s - loss: 0.4791 - acc: 0.793 - ETA: 0s - loss: 0.4769 - acc: 0.793 - ETA: 0s - loss: 0.4727 - acc: 0.795 - ETA: 0s - loss: 0.4700 - acc: 0.797 - ETA: 0s - loss: 0.4711 - acc: 0.799 - 3s 471us/step - loss: 0.4709 - acc: 0.7999 - val_loss: 0.7022 - val_acc: 0.5863
Epoch 5/10
6877/6877 [==============================] - ETA: 1s - loss: 0.6417 - acc: 0.640 - ETA: 1s - loss: 0.4173 - acc: 0.820 - ETA: 1s - loss: 0.4044 - acc: 0.829 - ETA: 1s - loss: 0.4059 - acc: 0.825 - ETA: 1s - loss: 0.4072 - acc: 0.830 - ETA: 1s - loss: 0.3987 - acc: 0.833 - ETA: 1s - loss: 0.3995 - acc: 0.825 - ETA: 1s - loss: 0.3995 - acc: 0.826 - ETA: 1s - loss: 0.3904 - acc: 0.833 - ETA: 1s - loss: 0.3927 - acc: 0.830 - ETA: 1s - loss: 0.3880 - acc: 0.833 - ETA: 1s - loss: 0.3818 - acc: 0.837 - ETA: 0s - loss: 0.3754 - acc: 0.843 - ETA: 0s - loss: 0.3743 - acc: 0.844 - ETA: 0s - loss: 0.3710 - acc: 0.846 - ETA: 0s - loss: 0.3686 - acc: 0.849 - ETA: 0s - loss: 0.3672 - acc: 0.850 - ETA: 0s - loss: 0.3712 - acc: 0.847 - ETA: 0s - loss: 0.3669 - acc: 0.850 - ETA: 0s - loss: 0.3621 - acc: 0.853 - ETA: 0s - loss: 0.3598 - acc: 0.854 - ETA: 0s - loss: 0.3579 - acc: 0.856 - ETA: 0s - loss: 0.3549 - acc: 0.858 - ETA: 0s - loss: 0.3534 - acc: 0.858 - ETA: 0s - loss: 0.3530 - acc: 0.858 - ETA: 0s - loss: 0.3562 - acc: 0.855 - ETA: 0s - loss: 0.3554 - acc: 0.856 - ETA: 0s - loss: 0.3523 - acc: 0.857 - ETA: 0s - loss: 0.3489 - acc: 0.859 - ETA: 0s - loss: 0.3470 - acc: 0.860 - ETA: 0s - loss: 0.3442 - acc: 0.861 - 3s 469us/step - loss: 0.3425 - acc: 0.8624 - val_loss: 0.6694 - val_acc: 0.6505
Epoch 6/10
6877/6877 [==============================] - ETA: 1s - loss: 0.3975 - acc: 0.840 - ETA: 1s - loss: 0.3331 - acc: 0.872 - ETA: 1s - loss: 0.2968 - acc: 0.890 - ETA: 1s - loss: 0.3055 - acc: 0.881 - ETA: 1s - loss: 0.3044 - acc: 0.880 - ETA: 1s - loss: 0.2955 - acc: 0.883 - ETA: 1s - loss: 0.2912 - acc: 0.887 - ETA: 1s - loss: 0.2929 - acc: 0.885 - ETA: 1s - loss: 0.2988 - acc: 0.883 - ETA: 1s - loss: 0.3079 - acc: 0.879 - ETA: 1s - loss: 0.3057 - acc: 0.880 - ETA: 1s - loss: 0.3032 - acc: 0.882 - ETA: 0s - loss: 0.3054 - acc: 0.882 - ETA: 0s - loss: 0.3073 - acc: 0.882 - ETA: 0s - loss: 0.3073 - acc: 0.881 - ETA: 0s - loss: 0.3103 - acc: 0.880 - ETA: 0s - loss: 0.3085 - acc: 0.880 - ETA: 0s - loss: 0.3051 - acc: 0.883 - ETA: 0s - loss: 0.3023 - acc: 0.884 - ETA: 0s - loss: 0.2985 - acc: 0.887 - ETA: 0s - loss: 0.2969 - acc: 0.887 - ETA: 0s - loss: 0.2942 - acc: 0.888 - ETA: 0s - loss: 0.2923 - acc: 0.889 - ETA: 0s - loss: 0.2918 - acc: 0.888 - ETA: 0s - loss: 0.2913 - acc: 0.889 - ETA: 0s - loss: 0.2885 - acc: 0.891 - ETA: 0s - loss: 0.2857 - acc: 0.892 - ETA: 0s - loss: 0.2847 - acc: 0.893 - ETA: 0s - loss: 0.2875 - acc: 0.891 - ETA: 0s - loss: 0.2859 - acc: 0.891 - ETA: 0s - loss: 0.2864 - acc: 0.891 - 3s 469us/step - loss: 0.2866 - acc: 0.8917 - val_loss: 0.5661 - val_acc: 0.7211
Epoch 7/10
6877/6877 [==============================] - ETA: 1s - loss: 0.1464 - acc: 0.960 - ETA: 1s - loss: 0.3060 - acc: 0.854 - ETA: 1s - loss: 0.2915 - acc: 0.878 - ETA: 1s - loss: 0.2817 - acc: 0.878 - ETA: 1s - loss: 0.2623 - acc: 0.888 - ETA: 1s - loss: 0.2587 - acc: 0.892 - ETA: 1s - loss: 0.2607 - acc: 0.892 - ETA: 1s - loss: 0.2610 - acc: 0.892 - ETA: 1s - loss: 0.2569 - acc: 0.896 - ETA: 1s - loss: 0.2527 - acc: 0.897 - ETA: 1s - loss: 0.2500 - acc: 0.898 - ETA: 1s - loss: 0.2501 - acc: 0.899 - ETA: 0s - loss: 0.2517 - acc: 0.898 - ETA: 0s - loss: 0.2525 - acc: 0.897 - ETA: 0s - loss: 0.2530 - acc: 0.895 - ETA: 0s - loss: 0.2503 - acc: 0.897 - ETA: 0s - loss: 0.2495 - acc: 0.898 - ETA: 0s - loss: 0.2505 - acc: 0.896 - ETA: 0s - loss: 0.2504 - acc: 0.897 - ETA: 0s - loss: 0.2467 - acc: 0.899 - ETA: 0s - loss: 0.2471 - acc: 0.899 - ETA: 0s - loss: 0.2499 - acc: 0.898 - ETA: 0s - loss: 0.2509 - acc: 0.898 - ETA: 0s - loss: 0.2497 - acc: 0.898 - ETA: 0s - loss: 0.2491 - acc: 0.899 - ETA: 0s - loss: 0.2466 - acc: 0.901 - ETA: 0s - loss: 0.2481 - acc: 0.900 - ETA: 0s - loss: 0.2483 - acc: 0.900 - ETA: 0s - loss: 0.2495 - acc: 0.900 - ETA: 0s - loss: 0.2508 - acc: 0.900 - ETA: 0s - loss: 0.2526 - acc: 0.899 - 3s 476us/step - loss: 0.2535 - acc: 0.8991 - val_loss: 0.6633 - val_acc: 0.7014
Epoch 8/10
6877/6877 [==============================] - ETA: 1s - loss: 0.4436 - acc: 0.800 - ETA: 1s - loss: 0.2287 - acc: 0.896 - ETA: 1s - loss: 0.2133 - acc: 0.913 - ETA: 1s - loss: 0.2047 - acc: 0.917 - ETA: 1s - loss: 0.2043 - acc: 0.918 - ETA: 1s - loss: 0.1947 - acc: 0.926 - ETA: 1s - loss: 0.1940 - acc: 0.925 - ETA: 1s - loss: 0.1961 - acc: 0.924 - ETA: 1s - loss: 0.2002 - acc: 0.922 - ETA: 1s - loss: 0.2050 - acc: 0.920 - ETA: 1s - loss: 0.2047 - acc: 0.920 - ETA: 1s - loss: 0.2019 - acc: 0.921 - ETA: 0s - loss: 0.2000 - acc: 0.922 - ETA: 0s - loss: 0.2008 - acc: 0.922 - ETA: 0s - loss: 0.2074 - acc: 0.918 - ETA: 0s - loss: 0.2075 - acc: 0.917 - ETA: 0s - loss: 0.2153 - acc: 0.915 - ETA: 0s - loss: 0.2224 - acc: 0.911 - ETA: 0s - loss: 0.2265 - acc: 0.909 - ETA: 0s - loss: 0.2258 - acc: 0.910 - ETA: 0s - loss: 0.2260 - acc: 0.910 - ETA: 0s - loss: 0.2250 - acc: 0.910 - ETA: 0s - loss: 0.2216 - acc: 0.912 - ETA: 0s - loss: 0.2214 - acc: 0.912 - ETA: 0s - loss: 0.2211 - acc: 0.912 - ETA: 0s - loss: 0.2203 - acc: 0.912 - ETA: 0s - loss: 0.2187 - acc: 0.913 - ETA: 0s - loss: 0.2191 - acc: 0.913 - ETA: 0s - loss: 0.2193 - acc: 0.913 - ETA: 0s - loss: 0.2178 - acc: 0.914 - ETA: 0s - loss: 0.2176 - acc: 0.915 - 3s 478us/step - loss: 0.2176 - acc: 0.9155 - val_loss: 0.6637 - val_acc: 0.7144
Epoch 9/10
6877/6877 [==============================] - ETA: 1s - loss: 0.1912 - acc: 0.960 - ETA: 1s - loss: 0.2035 - acc: 0.928 - ETA: 1s - loss: 0.2160 - acc: 0.917 - ETA: 1s - loss: 0.1937 - acc: 0.921 - ETA: 1s - loss: 0.1857 - acc: 0.925 - ETA: 1s - loss: 0.1924 - acc: 0.922 - ETA: 1s - loss: 0.1937 - acc: 0.920 - ETA: 1s - loss: 0.1929 - acc: 0.918 - ETA: 1s - loss: 0.1930 - acc: 0.921 - ETA: 1s - loss: 0.1884 - acc: 0.924 - ETA: 1s - loss: 0.1845 - acc: 0.927 - ETA: 1s - loss: 0.1846 - acc: 0.926 - ETA: 0s - loss: 0.1897 - acc: 0.923 - ETA: 0s - loss: 0.1907 - acc: 0.922 - ETA: 0s - loss: 0.1890 - acc: 0.923 - ETA: 0s - loss: 0.1913 - acc: 0.923 - ETA: 0s - loss: 0.1941 - acc: 0.922 - ETA: 0s - loss: 0.1896 - acc: 0.925 - ETA: 0s - loss: 0.1891 - acc: 0.925 - ETA: 0s - loss: 0.1888 - acc: 0.925 - ETA: 0s - loss: 0.1881 - acc: 0.925 - ETA: 0s - loss: 0.1878 - acc: 0.926 - ETA: 0s - loss: 0.1896 - acc: 0.926 - ETA: 0s - loss: 0.1894 - acc: 0.925 - ETA: 0s - loss: 0.1915 - acc: 0.924 - ETA: 0s - loss: 0.1917 - acc: 0.924 - ETA: 0s - loss: 0.1947 - acc: 0.922 - ETA: 0s - loss: 0.1982 - acc: 0.921 - ETA: 0s - loss: 0.1984 - acc: 0.921 - ETA: 0s - loss: 0.1986 - acc: 0.920 - ETA: 0s - loss: 0.1990 - acc: 0.920 - 3s 466us/step - loss: 0.1990 - acc: 0.9203 - val_loss: 0.6975 - val_acc: 0.7060
Epoch 10/10
6877/6877 [==============================] - ETA: 1s - loss: 0.3060 - acc: 0.880 - ETA: 1s - loss: 0.1647 - acc: 0.938 - ETA: 1s - loss: 0.1785 - acc: 0.937 - ETA: 1s - loss: 0.1654 - acc: 0.944 - ETA: 1s - loss: 0.1560 - acc: 0.945 - ETA: 1s - loss: 0.1521 - acc: 0.945 - ETA: 1s - loss: 0.1528 - acc: 0.944 - ETA: 1s - loss: 0.1473 - acc: 0.946 - ETA: 1s - loss: 0.1461 - acc: 0.947 - ETA: 1s - loss: 0.1476 - acc: 0.948 - ETA: 1s - loss: 0.1624 - acc: 0.941 - ETA: 0s - loss: 0.1651 - acc: 0.940 - ETA: 0s - loss: 0.1679 - acc: 0.936 - ETA: 0s - loss: 0.1664 - acc: 0.937 - ETA: 0s - loss: 0.1664 - acc: 0.936 - ETA: 0s - loss: 0.1668 - acc: 0.936 - ETA: 0s - loss: 0.1660 - acc: 0.937 - ETA: 0s - loss: 0.1659 - acc: 0.938 - ETA: 0s - loss: 0.1632 - acc: 0.938 - ETA: 0s - loss: 0.1646 - acc: 0.938 - ETA: 0s - loss: 0.1685 - acc: 0.937 - ETA: 0s - loss: 0.1679 - acc: 0.937 - ETA: 0s - loss: 0.1666 - acc: 0.937 - ETA: 0s - loss: 0.1670 - acc: 0.937 - ETA: 0s - loss: 0.1668 - acc: 0.937 - ETA: 0s - loss: 0.1680 - acc: 0.937 - ETA: 0s - loss: 0.1702 - acc: 0.936 - ETA: 0s - loss: 0.1702 - acc: 0.936 - ETA: 0s - loss: 0.1706 - acc: 0.935 - ETA: 0s - loss: 0.1682 - acc: 0.936 - ETA: 0s - loss: 0.1674 - acc: 0.936 - 3s 468us/step - loss: 0.1670 - acc: 0.9372 - val_loss: 0.6829 - val_acc: 0.7318

 我相信继续调参之后就可以辽！

 

 

说明：我的train和test比例不对，这个是我故意的，我也不懂机器学习，这都是带着猜测的心理改了四个多小时才让这个网络正常的额，如有错漏，轻喷~

参考链接：

https://www.zhihu.com/question/49346370

https://www.cnblogs.com/kjkj/p/10528259.html

http://www.voidcn.com/article/p-emsxaiqs-buk.html

 
Masked5
关注
  • 6
    点赞
  • 踩
  • 37
    收藏
    觉得还不错? 一键收藏
  • 3
    评论
专栏目录
深度学习网络模型训练中loss为nans的总结
神秘少年的博客
06-18 4237
在网络训练中经常出现loss为nans的情况。 主要原因有如下： 梯度爆炸 原因：巨大的梯度值使model的学习过程off-track。 检测方法：查看运行时日志，您应该查看每个迭代的损失值。您会注意到，损失开始从迭代到迭代显著地增长，最终损失将太大，无法用浮点变量表示，它将成为nan。 措施：减少base_lr (在solver.prototxt中) 至少一个数量级。如果mod
keras的验证损失val_loss不减反增和验证准确率val_accuracy不增反减
Mr.horse的博客
09-21 1944
换一种优化器或调整学习率
【机器学习】模型训练loss损失值正常下降，但是acc准确率精度却为零0
最新发布
qq_33700934的博客
06-19 653
如果模型在训练时使用了不同的标签集，或者预测时标签的处理方式与训练时不一致，那么精度可能会受到影响。2.类别不平衡：如果数据集中某些类别的样本数量远远多于其他类别，而模型又没有得到适当的处理（例如使用权重调整、重采样等），那么模型可能会偏向于预测数量最多的类别，导致其他类别的精度很低甚至为0。例如，在分类任务中，你可能需要确保你的代码正确地计算了精度（即正确分类的样本数除以总样本数）。1.模型性能差：模型可能还没有学习到任何有用的特征或模式，导致它的预测几乎总是错误的。
python 关于训练集loss不降问题的经验分享
做减求空
05-12 2096
如何尽量避免这种问题的出现： 大部分情况是函数默认参数不同、或者是代码所处环境不同产生的bug 1. 使用相同的环境：系统、库的版本要一致 2. 设定固定的随机种子，可以规避一些不确定性问题 3. 在函数调用时主动填写默认参数(后续看代码也方便)
Keras-训练网络时的问题：loss一直为nan，accuracy一直为一个固定的数
q2972112的博客
12-25 3712
在使用VGG19做分类任务时，遇到一个问题：loss一直为nan，accuracy一直为一个固定的数，如下输出所示，即使加入了自动调整学习率 (ReduceLROnPlateau) 也没法解决问题。
Tensorflow 2.* keras 训练开始 loss 为NAN
Forrest97的博客
06-09 1811
问题描述 在网络训练过程中，如果一开始就为nan， 问题成因 说明loss function的定义出现了问题,即·log(0), x/0这样一类计算错误 解决方案 在损失函数的定义方程式中，避免此类问题。比如：加一个小常数，tf.keras.backend.epsilon等于1e-7 eps = tf.keras.backend.epsilon ...
浅谈keras使用中val_acc和acc值不同步的思考
09-16
主要介绍了浅谈keras使用中val_acc和acc值不同步的思考，具有很好的参考价值，希望对大家有所帮助。一起跟随小编过来看看吧
浅谈keras中loss与val_loss的关系
09-16
在Keras中，训练模型时我们经常关注两个关键指标：`loss`和`val_loss`。它们是评估模型性能的重要参数。 `loss`是模型在训练数据上的损失，它反映了模型预测输出与实际标签之间的差异。Keras提供了一系列内置的损失...
keras中epoch,batch,loss,val_loss用法说明
09-16
本文将深入解析Keras中的一些关键概念：epoch、batch、loss以及val_loss，并讨论它们在训练过程中的作用。 1. **epoch** - **定义**：一个epoch是指在训练过程中，神经网络完整地遍历一次数据集的过程。在这个过程...
keras做CNN的训练误差loss的下降操作
09-16
在使用Keras构建卷积神经网络（CNN）时，训练误差(loss)的下降是一个关键问题。本篇文章将探讨如何在Keras中优化CNN的训练过程，以降低损失并提高模型性能。 首先，针对训练误差(loss)无法下降的问题，可能是由于...
keras训练模型遇到loss:nan 解决办法
shuzip的博客
10-31 9044
训练神经网络的时候，增大了隐藏层的神经元个数，突然间loss值变成了nan 经过多方搜索，找到了原因 tensorflow解决方法网上已有，这里贴keras的解决办法 在pycharm里 import keras keras.losses ctrl + 左键 losses 进入损失函数模块 找到调用的损失函数，这里是 def categorical_crossentropy(y_true, y...
解决tensorflow-gpu版本训练loss一直为nan，或者loss，accuracy数值明显不对问题
Sciws的博客
06-15 1911
首先说下我的电脑是有y9000p，win11系统，3060显卡之前装了好几个版本都不行 。python=3.6 CUDA=10.1 cuDNN=7.6 tensorflow-gpu=2.2.0或者2.3.0python=3.8 CUDA=10.1 cuDNN=7.6 tensorflow-gpu=2.3.0都出现了loss一直为nan，或者loss，accuracy数值明显不对的问题尝试了一下用CPU tensorflow跑是正常的，并且也在服务器上用GPU跑了显示正常，说明不是学习率，网络模型，数据集等的
模型训练中出现NaN Loss的原因及解决方法
豆芽菜
05-23 1万+
1 梯度爆炸 原因：学习的过程中，梯度变得非常大，使得学习的过程偏离了正常的轨迹。 症状：观察每次迭代的loss值，会发现loss明显增长，最后因为loss值太大以至于不能用浮点去表示，所以变成了Nan。 可采取的措施：1 降低学习速率，2 如果模型中有多个loss层，就需要找到梯度爆炸的层，然后降低该层的loss weight。 2 学习率过高 原因：过高的学习率乘上所有的梯度使得所有参数变成无效的值。 症状：观察输出日志，会发现学习率变成nan 可采取的措施：设置合适的学习速率 3 损失函数
keras训练出现nan的解决流程
永远飞翔的鸟
09-29 6319
出现nan可能是在train的loss阶段，也可能是train的metric阶段，还可能是validation阶段，反正都一样。 在写毕设的过程里面，用学弟提取的特征做DNN的训练，loss没有出现nan，但是反而是metric（MSE）里面出现了nan，predict的结果也全是nan。尝试了很多方法来调bug，最后采用的方法是将训练集二分的方法来调试。 先将训练集划成（...
python prediction_Python Keras预测返回nan(Python Keras Prediction returning nan)
weixin_39703982的博客
12-09 327
我在理解Keras如何处理数据以及为什么我的模型不能正常工作方面遇到了问题。我正在尝试建立一个小模型，可以根据经度和纬度的输入来预测城市。我想看到的是，当我做一个预测时，例如，城市数组的第一个索引，我想看到的是输出数组索引0有最大的激活值。我目前的型号与Keras &Tensorflow < /强>< /强> 数据cities = [];cities.append(...
Keras学习笔记：Chapter2-初探Keras模型
DFan的NoteBook
10-28 4621
初探Sequential模型Sequential模型就是简单的“线性”堆叠，一层接着一层的连接，可通过.add()方法将一个一个层添加至模型，例： network = Sequential() network.add(Dense(512, activation='relu', input_shape=(28*28,))) network.add(Dense(10,activation='soft
（记录）tensorflow.keras.models.Sequential和输入层InputLayer以及dense层个人理解
weixin_53481887的博客
04-13 2827
model = models.Sequential() model.add(layers.Dense(64, activation='relu', input_shape=(10000,))) model.add(layers.Dense(64, activation='relu')) model.add(layers.Dense(46, activation='softmax')) 神经网络模型的构建通常由输入层，隐藏层和输出层构成 其中，输出层的激活函数通常为softmax 对神经网络全连接层.
Keras深度学习之示例应用
bgbgssh1314的博客
07-21 2893
4.1 输入——预测数值模型示例 4.1.1 模型准备 线性回归模型 w = np.cov(X,Y,bias=1)[0,1]/np.var(X) b = np.average(Y) - w * np.average(X) 感应器神经网络模型 是最基本的感应器神经网络模型，只有一个Dense层，一个神经元 model = Sequential() model.add(Dense(1,input_dim=1)) 多层感应器神经网络模型 是由两个De
深度模型 loss为nan解决方案详解
热门推荐
bitcarmanlee的博客
03-21 1万+
1.loss为nan 深度模型进行训练的时候，经常出现loss为nan的情况。比如某一次训练过程的输出如下： Epoch 1/50 1/303 [..............................] - ETA: 0s - loss: 278.5685 - accuracy: 0.0000e+00 47/303 [===>..........................] - ETA: 0s - loss: nan - accuracy: 0.5745 95/3
利用tensorboard分别将train_acc和val_acc，train_loss和val_loss画在一张图上代码
03-24
很高兴为您解答这个问题！以下是利用TensorBoard将train_acc和val_acc、train_loss和val_loss画在一张图上的代码： ```python import tensorflow as tf from tensorflow.keras.callbacks import TensorBoard # 定义TensorBoard回调函数 tensorboard_callback = TensorBoard(log_dir='./logs', histogram_freq=1) # 训练模型 model.fit(x_train, y_train, epochs=10, validation_data=(x_val, y_val), callbacks=[tensorboard_callback]) # 启动TensorBoard %load_ext tensorboard %tensorboard --logdir=./logs ``` 在训练模型时，我们将TensorBoard回调函数传递给了`callbacks`参数。这个回调函数会在每个epoch结束时自动记录训练和验证的准确率和损失，并将它们写入TensorBoard日志文件中。 最后，我们使用`%tensorboard`命令启动TensorBoard，并指定日志文件所在的目录。TensorBoard会自动读取日志文件，并在网页中展示训练和验证的准确率和损失曲线。 希望这个回答能够帮助您！
评论 3
添加红包

请填写红包祝福语或标题

个

红包个数最小为10个

元

红包金额最低5元

当前余额3.43元 前往充值 >
需支付：10.00元
成就一亿技术人!
领取后你会自动成为博主和红包主的粉丝 规则
hope_wisdom
发出的红包
实付元
使用余额支付
点击重新获取
扫码支付
钱包余额 0

抵扣说明：

1.余额是钱包充值的虚拟货币，按照1:1的比例进行支付金额的抵扣。
2.余额无法直接购买下载，可以购买VIP、付费专栏及课程。

余额充值