神经网络模型的超低精度

Question

我遵循了使用代码交叉验证进行神经网络模型评估的教程：

# Multiclass Classification with the Iris Flowers Dataset 
import numpy 
import pandas 
from keras.models import Sequential 
from keras.layers import Dense 
from keras.wrappers.scikit_learn import KerasClassifier 
from keras.utils import np_utils 
from sklearn.model_selection import cross_val_score 
from sklearn.model_selection import KFold 
from sklearn.preprocessing import LabelEncoder 
from sklearn.pipeline import Pipeline 
# fix random seed for reproducibility 
seed = 7 
numpy.random.seed(seed) 
# load dataset 
dataframe = pandas.read_csv("/content/drive/My Drive/iris.data", header=None) 
dataset = dataframe.values 
X = dataset[:,0:4].astype(float) 
Y = dataset[:,4] 

# encode class values as integers 
encoder = LabelEncoder() 
encoder.fit(Y) 
encoded_Y = encoder.transform(Y) 

# convert integers to dummy variables (i.e. one hot encoded) 
dummy_y = np_utils.to_categorical(encoded_Y) 

# define baseline model 
def baseline_model():

# create model
  model = Sequential()
  model.add(Dense(4, input_dim=4, activation="relu", kernel_initializer="normal"))
  model.add(Dense(3, activation="sigmoid", kernel_initializer="normal"))

# Compile model
  model.compile(loss= 'categorical_crossentropy' , optimizer= 'adam' , metrics=[ 'accuracy' ])

  return model 
estimator = KerasClassifier(build_fn=baseline_model, nb_epoch=200, batch_size=5, verbose=0) 
kfold = KFold(n_splits=10, shuffle=True, random_state=seed) 
results = cross_val_score(estimator, X, dummy_y, cv=kfold) 
print("Accuracy: %.2f%% (%.2f%%)" % (results.mean()*100, results.std()*100))

准确度应该在95.33% (4.27%) 左右，但经过几次尝试我得到了~Accuracy: 34.00% (13.15%)。模型代码似乎完全相同。我按照指示从here 下载了数据。会出什么问题？谢谢

Answer 1

替换这个：

model.add(Dense(4, input_dim=4, activation="relu", kernel_initializer="normal"))

有了这个：

model.add(Dense(16, activation="relu"))
model.add(Dense(32, activation="relu"))

然后，你的输出层为：

model.add(Dense(3, activation="softmax", kernel_initializer="normal"))

你的隐藏层微不足道，你的激活函数是错误的。对于 3 个以上的课程，它必须是 softmax。

完整工作代码：

import numpy
from keras.models import Sequential
from keras.layers import Dense
from keras.wrappers.scikit_learn import KerasClassifier
from keras.utils import np_utils
from sklearn.model_selection import cross_val_score
from sklearn.model_selection import KFold
from sklearn.preprocessing import MinMaxScaler

seed = 7
numpy.random.seed(seed)

from sklearn.datasets import load_iris

X, encoded_Y = load_iris(return_X_y=True)
mms = MinMaxScaler()
X = mms.fit_transform(X)

dummy_y = np_utils.to_categorical(encoded_Y)

def baseline_model():

    model = Sequential()
    model.add(Dense(4, input_dim=4, activation="relu", kernel_initializer="normal"))
    model.add(Dense(8, activation="relu", kernel_initializer="normal"))
    model.add(Dense(3, activation="softmax", kernel_initializer="normal"))

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

    return model

estimator = KerasClassifier(build_fn=baseline_model, epochs=200, verbose=0)
kfold = KFold(n_splits=10, shuffle=True, random_state=seed)
results = cross_val_score(estimator, X, dummy_y, cv=kfold)
print(results)

Out[5]: 
array([0.60000002, 0.93333334, 1.        , 0.66666669, 0.80000001,
       1.        , 1.        , 0.93333334, 0.80000001, 0.86666667])

Answer 2

仅凭机会，您应该获得 33% 的准确率。

如何改进代码：

1. 规范化数据。

from sklearn.preprocessing import StandardScaler, MinMaxScaler

scaler = StandardScaler()
X = scaler.fit_transform(X)

1. 增加层中的神经元数量，
2. 将输出的激活函数从 sigmoid 改为 softmax，
3. 使用categorical_crossentropy作为输出的损失，

# define baseline model 
def baseline_model():

# create model
  model = Sequential()
  model.add(Dense(8, input_dim=4, activation="relu"))
  model.add(Dense(3, activation="softmax"))

# Compile model
  model.compile(loss= 'categorical_crossentropy' , optimizer= 'adam' , metrics=[ 'accuracy' ])

  return model 

5. 将nb_epoch（旧Keras）更改为epochs，

estimator = KerasClassifier(build_fn=baseline_model, epochs=50, batch_size=5, verbose=1) 

这样，您将获得大约 90% 的准确率。如果你运行它超过 50 个 epoch，你最终会过度拟合你的模型，你甚至可以达到 100% 的准确率，但模型不能很好地泛化。

请记住，全连接层并不总是最好的解决方案。