无法将 1D 信号前馈到 Keras 中的 1D 卷积

Question

我是 Keras 的新手。给定输入数组，我想执行一维卷积。我该怎么做？我已经写了代码，但它无法运行

from keras.models import Sequential
from keras.layers import Conv1D
import numpy as np
import tensorflow as tf
from keras.models import Model

input = np.array(tf.constant([1,2,3,4,5,6,7,8]))   
model = Sequential()
model.add(Conv1D(1,3,strides=1,padding='same', name='conv'))

layer_model = Model(inputs=input,outputs=model.get_layer('conv').output)
conv_output = layer_model.predict(f)    
print (conv_output)

这是错误

    AttributeError                            Traceback (most recent call last)
<ipython-input-13-31db6e7dcb90> in <module>()
     10 
     11 layer_model = Model(inputs=input,
---> 12                                 outputs=model.get_layer('conv').output)
     13 conv_output = layer_model.predict(f)
     14 print (conv_output)

/usr/local/lib/python3.6/dist-packages/keras/engine/base_layer.py in output(self)
    808         if not self._inbound_nodes:
    809             raise AttributeError('Layer ' + self.name +
--> 810                                  ' has no inbound nodes.')
    811         if len(self._inbound_nodes) > 1:
    812             raise AttributeError('Layer ' + self.name +

AttributeError: Layer conv has no inbound nodes.

Answer 1

您不应该通过混合功能和顺序 API 来复杂化自己，更简单的方法是：

from keras.models import Sequential
from keras.layers import Conv1D
import numpy as np

input = np.array([1,2,3,4,5,6,7,8])
input = input.reshape(1, 8, 1) # (samples, width, channels)   

model = Sequential()
model.add(Conv1D(1,3,padding='same', input_shape=(8, 1)))

conv_output = model.predict(input)    
print(conv_output)

请注意，我们没有设置卷积权重，因此它们会自动初始化为随机值。