Slice of 20 elements of rank1 tensor then reshaping throws "Input to reshape is tensor with 10272 values, but requested shape requires multiple of 20"

user_11590962 0
Viewed 43

My input tensor Data = Input(shape=(856,)) is a vector of float32 values concatenated from many different devices. I am trying to apply different TensorFlow functions to different subslices of each input chunk. Some of these functions include a 1D Convolution which requires a reshape.

slice = Data[:20]
reshape = tf.reshape(slice, (-1, 20, 1))
...

Doing this crashes after trying to fit my model. It throws the following errors:

tensorflow.python.framework.errors_impl.InvalidArgumentError:  Input to reshape is a tensor with 10272 values, but the requested shape requires a multiple of 20
         [[node model/tf.reshape_1/Reshape
 (defined at /home/.local/lib/python3.8/site-packages/keras/layers/core/tf_op_layer.py:261)
]] [Op:__inference_train_function_1858]

Errors may have originated from an input operation.
Input Source operations connected to node model/tf.reshape_1/Reshape:
In[0] model/tf.__operators__.getitem_1/strided_slice:
In[1] model/tf.reshape_1/Reshape/shape:

I am not sure how slicing 20 elements from a tensor of 856 could result in a tensor of 10272 values.

I have also tried using the tf.slice function a couple of different ways; both fail. Referencing the docs: https://www.tensorflow.org/guide/tensor_slicing

slice = tf.slice(Data, begin=[0], size=[20]) 
...

And fails, stating:

Shape must be rank 1 but is rank 2 for '{{node tf.slice/Slice}} = Slice[Index=DT_INT32, T=DT_FLOAT](Placeholder, tf.slice/Slice/begin, tf.slice/Slice/size)' with input shapes: [?,856], [1], [1].

For reference, here is what some of the values look like in the input data

array([-9.55784683e+01, -1.70557899e+01,  2.95967350e+01,  7.81378937e+00,
        9.02729130e+00,  5.49621725e+00,  4.19811630e+00,  5.84186697e+00,
        4.90438080e+00,  3.73845983e+00,  5.12300587e+00,  2.61530232e+00,
        2.67061424e+00,  3.91038632e+00,  2.31110978e+00,  4.20644665e+00,
        4.50000000e+00,  9.87345278e-01,  1.59740388e+00,  6.30727148e+00,
...
keras tensorflow machine-learning numpy python
1 Answers

When you slice data like in Data[:20] it will produce a sequence with length min(20, len(Data)). So I guess your data has length less than 20.

Other message says it has rank 2, so I guess it has one of the following shapes

       1   10272
       2    5136
       3    3424
       4    2568
       6    1712
       8    1284
      12     856
      16     642

Any of those result in a tensor with 10272 elements as your first message shows, and that's not a multiple of 20.

user_127503530
Related