我试图为二进制分类任务创建一个神经网络模型。模型是这样的
LEARNING_RATE_INIT = 0.001
LEARNING_RATE_END = 0.0001
BATCH_SIZE = 20000
EPOCHS = 2
EMBEDDING_N = 50
DENSE_N = 1024
SPATIAL_DROPOUT_1D = 0.2
DROPOUT_1 = 0.2
DROPOUT_2 = 0.2
in_machine = Input(shape=[1], name='machine')
emb_machine = Embedding(max_machine, EMBEDDING_N)(in_machine)
in_windspeed = Input(shape=[1], name='windspeed')
emb_windspeed = Embedding(max_windspeed, EMBEDDING_N)(in_windspeed)
in_activepower = Input(shape=[1], name='activepower')
emb_activepower = Embedding(max_activepower, EMBEDDING_N)(in_activepower)
in_pitchangle = Input(shape=[1], name='pitchangle')
emb_pitchangle = Embedding(max_pitchangle, EMBEDDING_N)(in_pitchangle)
in_genspeed = Input(shape=[1], name='genspeed')
emb_genspeed = Embedding(max_genspeed, EMBEDDING_N)(in_genspeed)
in_temp = Input(shape=[1], name='temp')
emb_temp = Embedding(max_temp, EMBEDDING_N)(in_temp)
in_turbine = Input(shape=[1], name='turbine')
emb_turbine = Embedding(max_turbine, EMBEDDING_N)(in_turbine)
print('Create RNN Layers...')
fe = concatenate([
(emb_machine),
(emb_windspeed),
(emb_activepower),
(emb_pitchangle),
(emb_genspeed),
(emb_temp),
(emb_turbine)
])
s_dout = SpatialDropout1D(SPATIAL_DROPOUT_1D)(fe)
x = Flatten()(s_dout)
x = Dropout(0.2)(Dense(1024,activation='relu')(x))
x = Dropout(0.2)(Dense(256,activation='relu')(x))
outp = Dense(1,activation='sigmoid')(x)
model = Model(inputs=[in_machine, in_windspeed, in_activepower, in_pitchangle,
in_genspeed, in_temp, in_turbine], outputs=outp)
print('Model made')
它是这样编译的
model.compile(loss='binary_crossentropy', optimizer=optimizer_adam, metrics=['accuracy'])
model.summary()
模型架构如下所示
__________________________________________________________________________________________________
Layer (type) Output Shape Param # Connected to
==================================================================================================
machine (InputLayer) (None, 1) 0
__________________________________________________________________________________________________
windspeed (InputLayer) (None, 1) 0
__________________________________________________________________________________________________
activepower (InputLayer) (None, 1) 0
__________________________________________________________________________________________________
pitchangle (InputLayer) (None, 1) 0
__________________________________________________________________________________________________
genspeed (InputLayer) (None, 1) 0
__________________________________________________________________________________________________
temp (InputLayer) (None, 1) 0
__________________________________________________________________________________________________
turbine (InputLayer) (None, 1) 0
__________________________________________________________________________________________________
embedding_48 (Embedding) (None, 1, 50) 3700 machine[0][0]
__________________________________________________________________________________________________
embedding_49 (Embedding) (None, 1, 50) 1200 windspeed[0][0]
__________________________________________________________________________________________________
embedding_50 (Embedding) (None, 1, 50) 108000 activepower[0][0]
__________________________________________________________________________________________________
embedding_51 (Embedding) (None, 1, 50) 1500 pitchangle[0][0]
__________________________________________________________________________________________________
embedding_52 (Embedding) (None, 1, 50) 78350 genspeed[0][0]
__________________________________________________________________________________________________
embedding_53 (Embedding) (None, 1, 50) 3050 temp[0][0]
__________________________________________________________________________________________________
embedding_54 (Embedding) (None, 1, 50) 100 turbine[0][0]
__________________________________________________________________________________________________
concatenate_6 (Concatenate) (None, 1, 350) 0 embedding_48[0][0]
embedding_49[0][0]
embedding_50[0][0]
embedding_51[0][0]
embedding_52[0][0]
embedding_53[0][0]
embedding_54[0][0]
__________________________________________________________________________________________________
spatial_dropout1d_3 (SpatialDro (None, 1, 350) 0 concatenate_6[0][0]
__________________________________________________________________________________________________
flatten_17 (Flatten) (None, 350) 0 spatial_dropout1d_3[0][0]
__________________________________________________________________________________________________
dense_24 (Dense) (None, 1024) 359424 flatten_17[0][0]
__________________________________________________________________________________________________
dropout_18 (Dropout) (None, 1024) 0 dense_24[0][0]
__________________________________________________________________________________________________
dense_25 (Dense) (None, 256) 262400 dropout_18[0][0]
__________________________________________________________________________________________________
dropout_19 (Dropout) (None, 256) 0 dense_25[0][0]
__________________________________________________________________________________________________
dense_26 (Dense) (None, 1) 257 dropout_19[0][0]
==================================================================================================
Total params: 817,981
Trainable params: 817,981
Non-trainable params: 0
__________________________________________________________________________________________________
在尝试拟合模型时，出现以下错误。
InvalidArgumentError: indices[19577,0] = -3 is not in [0, 2160)
[[Node: embedding_50/GatherV2 = GatherV2[Taxis=DT_INT32, Tindices=DT_INT32, Tparams=DT_FLOAT, _class=["loc:@training_3/Adam/gradients/embedding_50/GatherV2_grad/Reshape"], _device="/job:localhost/replica:0/task:0/device:CPU:0"](embedding_50/embeddings/read, embedding_50/Cast, embedding_48/GatherV2/axis)]]
此错误的原因是什么？该如何解决？？
我的输入数据包含负值是问题所在。那么有没有办法合并负值？？
答案 0 :(得分：0)
您对Embedding的使用毫无意义，您说您的输入具有负值。如果您查看documentation for Embedding，它会显示：
将正整数（索引）转换为固定大小的密集向量。 例如。 [[4]，[20]]-> [[0.25，0.1]，[0.6，-0.2]]
这使我感到您的输入不是正整数，这意味着在这些输入上使用嵌入是没有意义的。嵌入用于语言建模，例如，将单词索引转换为向量，它应与任何类型的正整数输入一起使用，但不适用于常规输入。