为向量构建 keras CNN

Question

我是 ML 的初学者，正在尝试解决问题。 我有N个向量作为输入。 每个向量的长度为10 。 我得到了N个 0/1 的标签。 我需要在这些数据上训练一个 CNN（按提到的顺序）：

1. 卷积层
2. 最大池化
3. 致密层
4. 获取二进制 output 的 Softmax 激活

我做了：

def get_model(inputShape, filters = 32, kernel_size = 3, pool_size = 4, strides = 1):
    model = models.Sequential()
    model.add(layers.Conv1D(filters, kernel_size))
    model.add(layers.MaxPooling1D(pool_size, strides))
    model.add(layers.Dense(1))
    model.add(layers.Activation(activation='softmax'))
    model.build(inputShape)
    print(model.summary())
    return model

model = get_model((None, None, 10))

然后，我测试我的 model 是否正确构建，我这样做：（未经培训）

x = np.array([[1, 1, 1, 1, 1, 1, 1, 1, 1, 1], [1, 1, 1, 1, 1, 1, 1, 1, 1, 1]])
x = x.reshape(1, x.shape[0], x.shape[1])
print(model.predict(x))

我得到这个输出/错误：

_________________________________________________________________
Layer (type)                 Output Shape              Param #   
=================================================================
conv1d_1 (Conv1D)            (None, None, 32)          992       
_________________________________________________________________
max_pooling1d_1 (MaxPooling1 (None, None, 32)          0         
_________________________________________________________________
dense_1 (Dense)              (None, None, 1)           33        
_________________________________________________________________
activation_1 (Activation)    (None, None, 1)           0         
=================================================================
Total params: 1,025
Trainable params: 1,025
Non-trainable params: 0
_________________________________________________________________
None
2020-06-30 19:44:01.763782: W tensorflow/core/framework/op_kernel.cc:1753] OP_REQUIRES failed at pooling_ops_common.cc:91 : Invalid argument: Computed output size would be negative: -3 [input_size: 0, effective_filter_size: 4, stride: 1]
Traceback (most recent call last):
  File "model_predict.py", line 41, in <module>
    model.predict(x)
  File "/usr/local/lib/python3.8/dist-packages/keras/engine/training.py", line 1458, in predict
    return training_arrays.predict_loop(self, f, ins,
  File "/usr/local/lib/python3.8/dist-packages/keras/engine/training_arrays.py", line 324, in predict_loop
    batch_outs = f(ins_batch)
  File "/usr/local/lib/python3.8/dist-packages/tensorflow/python/keras/backend.py", line 3792, in __call__
    outputs = self._graph_fn(*converted_inputs)
  File "/usr/local/lib/python3.8/dist-packages/tensorflow/python/eager/function.py", line 1605, in __call__
    return self._call_impl(args, kwargs)
  File "/usr/local/lib/python3.8/dist-packages/tensorflow/python/eager/function.py", line 1645, in _call_impl
    return self._call_flat(args, self.captured_inputs, cancellation_manager)
  File "/usr/local/lib/python3.8/dist-packages/tensorflow/python/eager/function.py", line 1745, in _call_flat
    return self._build_call_outputs(self._inference_function.call(
  File "/usr/local/lib/python3.8/dist-packages/tensorflow/python/eager/function.py", line 593, in call
    outputs = execute.execute(
  File "/usr/local/lib/python3.8/dist-packages/tensorflow/python/eager/execute.py", line 59, in quick_execute
    tensors = pywrap_tfe.TFE_Py_Execute(ctx._handle, device_name, op_name,
tensorflow.python.framework.errors_impl.InvalidArgumentError:  Computed output size would be negative: -3 [input_size: 0, effective_filter_size: 4, stride: 1]
     [[node max_pooling1d_1/MaxPool (defined at /usr/local/lib/python3.8/dist-packages/keras/backend/tensorflow_backend.py:3007) ]] [Op:__inference_keras_scratch_graph_136]

Function call stack:
keras_scratch_graph

请建议我如何纠正这个问题。

Answer 1

问题在于x的维度。 它对于Conv1D(kernel_size=3)和MaxPooling1D(pool_size=4)了。

我向x添加了一个维度并减少kernel_size和pool_size ：

x = np.array([[1, 1, 1, 1, 1, 1, 1, 1, 1, 1], 
              [1, 1, 1, 1, 1, 1, 1, 1, 1, 1],
              [1, 1, 1, 1, 1, 1, 1, 1, 1, 1]], dtype=np.float32)
x = x.reshape(1, x.shape[0], x.shape[1])

model = Sequential()
model.add(layers.Conv1D(filters=10, kernel_size=2)) 
model.add(layers.MaxPooling1D(pool_size=2, strides=1))
model.add(layers.Dense(1))
model.add(layers.Activation(activation='softmax'))
model.build((None, None, 10))

model.predict(x)

如果您需要，我可以分享指向 google colab 文件的链接。