Question

以下是我正在使用的代码。请让我知道为什么我的验证和培训准确度如此之低？验证准确度仅为0.0000e + 00，培训准确率约为37％。什么可能出错？我的训练集有10500行和172列我的测试集有3150行和172列我的第一列是响应（类），因此我仅将其用作Y，其余列用作X. 我的回答是3个类：默认，LF和RF

from __future__ import print_function
import numpy as np
import pandas
from keras.models import Sequential
from keras.layers.core import Dense, Activation
from keras.optimizers import SGD
from keras.utils import np_utils
from sklearn.preprocessing import LabelEncoder
np.random.seed(1671)
NB_EPOCH = 5
BATCH_SIZE = 128
VERBOSE = 1
NB_CLASSES = 3
OPTIMIZER = SGD()
N_HIDDEN = 128
VALIDATION_SPLIT=0.1
RESHAPED = 171
dataframe_train = pandas.read_csv("TrainingEdgesToAction.csv", header=None)
dataset_train = dataframe_train.values
X_train = dataset_train[1:,1:172].astype(float)
#X_train = dataset_train[1:,0:172]
Y_train = dataset_train[1:,0]

dataframe_test = pandas.read_csv("TestingEdgesToAction.csv", header=None)
dataset_test = dataframe_test.values
X_test = dataset_test[1:,1:172].astype(float)
#X_test = dataset_test[1:,0:172]
Y_test = dataset_test[1:,0]

X_train = X_train.reshape(10500,RESHAPED)
X_test = X_test.reshape(3150,RESHAPED)
X_train /= 255
X_test /= 255
print(X_train.shape[0],'train samples')
print(X_test.shape[0],'test samples')

encoder = LabelEncoder()
encoder.fit(Y_train)
encoded_Y_train = encoder.transform(Y_train)
# convert integers to dummy variables (i.e. one hot encoded)
dummy_y_train = np_utils.to_categorical(encoded_Y_train)
print(dummy_y_train)

encoder = LabelEncoder()
encoder.fit(Y_test)
encoded_Y_test = encoder.transform(Y_test)
# convert integers to dummy variables (i.e. one hot encoded)
dummy_y_test = np_utils.to_categorical(encoded_Y_test)
print(dummy_y_test)

#Y_train = np_utils.to_categorical(Y_train,NB_CLASSES)
#Y_test = np_utils.to_categorical(Y_test, NB_CLASSES)

model = Sequential()
model.add(Dense(N_HIDDEN,input_shape=(RESHAPED,)))
model.add(Activation('relu'))
model.add(Dense(N_HIDDEN))
model.add(Activation('relu'))
model.add(Dense(NB_CLASSES))
model.add(Activation('softmax'))
model.summary()
model.compile(loss='categorical_crossentropy',optimizer=OPTIMIZER,metrics=
['accuracy'])
history = model.fit(X_train,dummy_y_train,batch_size=BATCH_SIZE,epochs=NB_EPOCH,shuffle=True,verbose=VERBOSE,validation_split=VALIDATION_SPLIT)
score = model.evaluate(X_test,dummy_y_test,verbose=VERBOSE)

print("\nTest score:",score[0])
print("Test accuracy:",score[1])

10500 train samples
3150 test samples
[[ 1.  0.  0.]
[ 1.  0.  0.]
[ 1.  0.  0.]
..., 
[ 0.  0.  1.]
[ 0.  0.  1.]
[ 0.  0.  1.]]
[[ 1.  0.  0.]
[ 1.  0.  0.]
[ 1.  0.  0.]
..., 
[ 0.  0.  1.]
[ 0.  0.  1.]
[ 0.  0.  1.]]
_________________________________________________________________
Layer (type)                 Output Shape              Param #   
=================================================================
dense_49 (Dense)             (None, 128)               22016     
_________________________________________________________________
activation_49 (Activation)   (None, 128)               0         
_________________________________________________________________
dense_50 (Dense)             (None, 128)               16512     
_________________________________________________________________
activation_50 (Activation)   (None, 128)               0         
_________________________________________________________________
dense_51 (Dense)             (None, 3)                 387       
_________________________________________________________________
activation_51 (Activation)   (None, 3)                 0         
=================================================================
Total params: 38,915
Trainable params: 38,915
Non-trainable params: 0
_________________________________________________________________
Train on 9450 samples, validate on 1050 samples
Epoch 1/5
9450/9450 [==============================] - 2s - loss: 1.0944 - acc: 0.3618 
- val_loss: 1.1809 - val_acc: 0.0000e+00
Epoch 2/5
9450/9450 [==============================] - 1s - loss: 1.0895 - acc: 0.3704 
- val_loss: 1.2344 - val_acc: 0.0000e+00
Epoch 3/5
9450/9450 [==============================] - 0s - loss: 1.0874 - acc: 0.3704 
- val_loss: 1.2706 - val_acc: 0.0000e+00
Epoch 4/5
9450/9450 [==============================] - 0s - loss: 1.0864 - acc: 0.3878 
- val_loss: 1.2955 - val_acc: 0.0000e+00
Epoch 5/5
9450/9450 [==============================] - 0s - loss: 1.0860 - acc: 0.3761 
- val_loss: 1.3119 - val_acc: 0.0000e+00
2848/3150 [==========================>...] - ETA: 0s
Test score: 1.10844093784
Test accuracy: 0.333333333333

Answer 1

我决定总结一下我们的“聊天”。

那么，如果你的测试精度很低（大约≈0.1％）该怎么办，这里有一些一般的建议：

尝试不同的优化器，根据我的经验，NAdam是一个很好的起点。
尝试不同的激活功能;我建议你从“relu”开始，然后尝试“selu”和“elu”。
添加正则化。 Dropout和BatchNormalization可能会提高您的测试准确性。
给你的网络一些时间，训练更长时间。
使用多参数，例如多个图层，批量大小，多个时代，学习率等等...
最后，始终 规范化您的数据 ，然后再将其提供给NN。

Answer 2

所以对我来说，我一直在寻找一种指标，以了解我的回归模型的效果，并且对于我来说，以R²作为主要指标最有效。 R²可以描述模型在进行预测时的“良好”程度。正如@Paddy已经提到的，您需要花一些时间。您的情况至少需要30个纪元。所以现在当：

R²= 0意味着模型总是无法预测正确的目标变量 R²= 1意味着模型可以完美地预测目标变量。

在代码中，它看起来像这样：

def det_coeff(y_true, y_pred):
u = K.sum(K.square(y_true - y_pred))
v = K.sum(K.square(y_true - K.mean(y_true)))
return K.ones_like(v) - (u / v)

这是确定指标的功能。

model.compile(optimizer="SGD", loss="mse", metrics=[det_coeff])

Answer 3

我遇到了类似的问题。请尝试洗牌您的数据，这可以解决您的问题。

from sklearn.utils import shuffle
Xtrain, ytrain = shuffle(Xtrain, ytrain)

使用keras和验证精度的神经网络精度非常低0.0000e + 00

3 个答案: