python - Keras:精度非常低，损失非常高，并且每个输入的预测都是相同的

Question

我正在开发一个使用 keras 进行通用音频标记的系统。

我有以下数据输入:x_train 每个输入有 10 个不同的数据(data_leng、max、min 等)，y_train 代表 41 个可能的标签(吉他、贝斯等)

x_train shape = (7104, 10)
y_train shape = (41,)

print(x_train[0])

[ 3.75732000e+05 -2.23437546e-05 -1.17187500e-02  1.30615234e-02
  2.65964586e-03  2.65973969e-03  9.80024859e-02  1.13624850e+00
  1.00003528e+00 -1.11458333e+00] 

print(y_train[0])

[0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0.
 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 1. 0. 0. 0.]

我的模型是:

from keras.models import Sequential
from keras.optimizers import SGD
from keras.layers import Dense, Dropout, Activation

model = Sequential()

model.add(Dense(units=128, activation='relu', input_dim=10))
model.add(Dropout(0.5))
model.add(Dense(units=64, activation='relu'))
model.add(Dropout(0.5))
model.add(Dense(units=32, activation='relu'))
model.add(Dropout(0.5))
model.add(Dense(41, activation='softmax'))

opt = SGD(lr=0.0001, decay=1e-6, momentum=0.9, nesterov=True)

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

model.fit(np.array(x_train), np.array(y_train), epochs=5, batch_size=8)

这是我的结果:

Epoch 1/5
7104/7104 [==============================] - 1s 179us/step - loss: 15.7392 - acc: 0.0235
Epoch 2/5
7104/7104 [==============================] - 1s 132us/step - loss: 15.7369 - acc: 0.0236
Epoch 3/5
7104/7104 [==============================] - 1s 133us/step - loss: 15.7415 - acc: 0.0234
Epoch 4/5
7104/7104 [==============================] - 1s 132us/step - loss: 15.7262 - acc: 0.0242
Epoch 5/5
7104/7104 [==============================] - 1s 132us/step - loss: 15.6484 - acc: 0.0291

正如您所看到的，我的结果显示非常高的数据丢失和非常低的准确性，但主要问题是当我尝试预测结果时，因为每个输入的输出都是相同的。我怎样才能解决这个问题 ？

pre = model.predict(np.array(x_train), batch_size=8, verbose=0)

for i in pre:
    print(i)

[0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0.
 0. 0. 0. 0. 0. 1. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0.]
[0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0.
 0. 0. 0. 0. 0. 1. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0.]
[0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0.
 0. 0. 0. 0. 0. 1. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0.]
[0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0.
 0. 0. 0. 0. 0. 1. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0.]
[0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0.
 0. 0. 0. 0. 0. 1. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0.]
...

Answer 1

在 Dense 层中，您只需为第一层指定 Input_dim。

Keras 负责处理其他层中的 Dim。

所以尝试一下:

model = Sequential()

model.add(Dense(units=128, activation='relu', input_dim=10))
model.add(Dropout(0.5))
model.add(Dense(units=64, activation='relu'))
model.add(Dropout(0.5))
model.add(Dense(units=32, activation='relu'))
model.add(Dropout(0.5))
model.add(Dense(41, activation='softmax'))

也许您的正则化对于此类数据来说太强了，请尝试使用不太强的 dropout 或根本不使用 dropout。

你能做的最后一件事就是提高你的学习率，从 1e-3 之类的东西开始，看看是否有变化。

希望对你有帮助