Training CNN with custom miniBatchDatastore thows cryptic error

I am trying to train a CNN model for a classification task. The input to the model is a `1x1024x2` frame and I have the following network:

layers = 
  28×1 Layer array with layers:

     1   'Input Layer'   Image Input             1×1024×2 images
     2   'CNN1'          2-D Convolution         16 1×8 convolutions with stride [1  1] and padding 'same'
     3   'BN1'           Batch Normalization     Batch normalization
     4   'ReLU1'         ReLU                    ReLU
     5   'MaxPool1'      2-D Max Pooling         1×2 max pooling with stride [1  2] and padding [0  0  0  0]
     6   'CNN2'          2-D Convolution         24 1×8 convolutions with stride [1  1] and padding 'same'
     7   'BN2'           Batch Normalization     Batch normalization
     8   'ReLU2'         ReLU                    ReLU
     9   'MaxPool2'      2-D Max Pooling         1×2 max pooling with stride [1  2] and padding [0  0  0  0]
    10   'CNN3'          2-D Convolution         32 1×8 convolutions with stride [1  1] and padding 'same'
    11   'BN3'           Batch Normalization     Batch normalization
    12   'ReLU3'         ReLU                    ReLU
    13   'MaxPool3'      2-D Max Pooling         1×2 max pooling with stride [1  2] and padding [0  0  0  0]
    14   'CNN4'          2-D Convolution         48 1×8 convolutions with stride [1  1] and padding 'same'
    15   'BN4'           Batch Normalization     Batch normalization
    16   'ReLU4'         ReLU                    ReLU
    17   'MaxPool4'      2-D Max Pooling         1×2 max pooling with stride [1  2] and padding [0  0  0  0]
    18   'CNN5'          2-D Convolution         64 1×8 convolutions with stride [1  1] and padding 'same'
    19   'BN5'           Batch Normalization     Batch normalization
    20   'ReLU5'         ReLU                    ReLU
    21   'MaxPool5'      2-D Max Pooling         1×2 max pooling with stride [1  2] and padding [0  0  0  0]
    22   'CNN6'          2-D Convolution         96 1×8 convolutions with stride [1  1] and padding 'same'
    23   'BN6'           Batch Normalization     Batch normalization
    24   'ReLU6'         ReLU                    ReLU
    25   'AP1'           2-D Average Pooling     1×32 average pooling with stride [1  1] and padding [0  0  0  0]
    26   'FC1'           Fully Connected         24 fully connected layer
    27   'SoftMax'       Softmax                 softmax
    28   'Output'        Classification Output   crossentropyex

I'm using a custom minibatch dataset to train the model. The read funtion returns the following table structure:

i'm using these training options:

% Specify training options
options = trainingOptions('sgdm', ...
    'MiniBatchSize', miniBatchSize, ...
    'MaxEpochs', 10, ...
    'Verbose', true, ...
    'Plots', 'training-progress');

% Train the network
net = trainNetwork(mbds, layers, options);

and i get the following error:

Error using trainNetwork
Operands to the logical AND (&&) and OR (||) operators must be convertible to logical scalar values. Use the ANY or ALL functions to reduce operands to logical scalar values.

Caused by:
    Operands to the logical AND (&&) and OR (||) operators must be convertible to logical scalar values. Use the ANY or ALL functions to reduce operands to logical scalar values.

I've been stuck on this for a while now, and theres no way to inspect the trainNetwork code to properly debug. I'd appreciate your suggestions. Thank you.

Paul Osinowo

Graduate Student

University of Strathclyde

Answer 1

[Karan Singh]

Hi @Paul,

I have faced this error before. We can do one of two things, whichever works for you, as I don't have access to your full code:

1. The error occurs even if the operands being evaluated by the short-circuit || and && operators are empty or non-scalar arrays. You can use logical & and | operators for non-scalar operations. Now you can go through the script and check if this is caused by your script or not.
2. It can also happen if the MATLAB path or the toolbox cache is corrupted. You can restore the settings by trying them out.

    >> restoredefaultpath
    >> rehash toolboxcache                                          

It worked for me and may be useful for you.