The dataset seems to be missing the batch dimension - AKA how to construct a dataset with mutiple ou-CodePudding

I'm having a hard time understanding what the problem is. Consider the following model:

Model: "model_8"
__________________________________________________________________________________________________
 Layer (type)                   Output Shape         Param #     Connected to                     
==================================================================================================
 input (InputLayer)             [(None, 15)]         0           []                               
                                                                                                  
 dense_1 (Dense)                (None, 128)          2048        ['input[0][0]']                  
                                                                                                  
 dense_2 (Dense)                (None, 1024)         132096      ['dense_1[0][0]']                
                                                                                                  
 dense_3 (Dense)                (None, 5120)         5248000     ['dense_2[0][0]']                
                                                                                                  
 a_out (Dense)                  (None, 17)           87057       ['dense_3[0][0]']                
                                                                                                  
 b_out (Dense)                  (None, 27)           138267      ['dense_3[0][0]']                
                                                                                                  
 c_out (Dense)                  (None, 71)           363591      ['dense_3[0][0]']                
                                                                                                  
 d_out (Dense)                  (None, 29)           148509      ['dense_3[0][0]']                
                                                                                                  
==================================================================================================
Total params: 6,119,568
Trainable params: 6,119,568
Non-trainable params: 0

It's a rather simple model with one input and 4 outputs (a_out, b_out, c_out, and d_out). I'm trying to fit the model by feeding it some dataset:

dataset = tf.data.Dataset.from_tensor_slices((inputs, {'a_out': targets[:, 0],
                                                       'b_out': targets[:, 1],
                                                       'c_out': targets[:, 2],
                                                       'd_out': targets[:, 3]}))

The inputs and targets are two numpy arrays with shapes: (525081, 15) and (525081, 4) respectively. When I run the fit method:

model.fit(dataset, epochs=10, batch_size=128)

I get the following error:

ValueError: Exception encountered when calling layer "model_8" (type Functional).
    
    Input 0 of layer "dense_1" is incompatible with the layer: expected min_ndim=2, found ndim=1. Full shape received: (15,)
    
    Call arguments received:
      • inputs=tf.Tensor(shape=(15,), dtype=float64)
      • training=True
      • mask=None

It seems to me like the tensor sent to the layer_1 is missing the batch dimension which does not make sense to me. Am I constructing my dataset wrong?

CodePudding user response：

When using the tf.data, batch_size parameter in model.fit is ignored.

Batching should be done with .batch() method of tf.data.Dataset.

In your case it should be dataset.batch(batch_size).