Layer Input Expectation Doesn't Match (Neural Network)

Question

Candace 2024-5-2

0
链接

此问题的直接链接

https://ww2.mathworks.cn/matlabcentral/answers/2114111-layer-input-expectation-doesn-t-match-neural-network

评论： Matt J 2024-5-3

Creating a dlnetwork with three inputs and two outputs. The inputs are x, y, z coordinates in the form of UTMN, UTME, and height. The outputs are pressure and temperature. Data is from multiple well logs and is organized by well ID, UTMN, UTME, height, temp, and pressure.

I seperated the data into training and validation sets. Training data: train_x_array (3 feature inputs) and train_y_array (2 feature outputs). Validation: val_x_array and val_y_array.

Keep getting errors such as:

Training stopped: Error occurred

Error using trainnet

Layer 'input': Invalid input data. Invalid size of channel dimension. Layer expects input with channel dimension size 3

but received input with size 189.

Error in final3 (line 25)

netTrained = trainnet(train_x_array_transposed, train_y_array_transposed, net,"mse", opts);

In an effort to fix the issue, I transposed the x and y arrays for both the training and validation sets. However, I am still getting the same error. I am at a loss for how to fix the error. If I try to use the Deep Network Designer, I receive an error stating their are multiple observations. The x and y values are the same for different heights/temps/pressures from the same well. I am assuming this is were the issue arises, but I do not understand why the designer would produce a different error.

Example data (before transposing):

x y z T P

4407300 327880 2796 344.3 501

4407300 327880 2746 356.5 521

4407300 327880 2696 357 541

Where train_x would be the first 3 columns, and train_y is the last 2.

Full Code:

inputSize = 3; 
hiddenLayerSize = 20; 
outputSize = 2; 
layers = [
    featureInputLayer(inputSize)
    fullyConnectedLayer(hiddenLayerSize, 'Name', 'fc1')
    reluLayer('Name', 'relu1')
    fullyConnectedLayer(outputSize, 'Name', 'output')
];
net = dlnetwork(layers);
opts = trainingOptions('adam', ... 
    'MaxEpochs', 100, ...
    'MiniBatchSize', 64, ...
    'ValidationData', {val_x_array_transposed', val_y_array_transposed'}, ... 
    'ValidationFrequency', 10, ... 
    'Verbose', true ...
     );
netTrained = trainnet(train_x_array_transposed, train_y_array_transposed, net,"mse", opts);
Y_pred_val = predict(netTrained, val_x_array'); 

0 个评论
显示 -2更早的评论隐藏 -2更早的评论

请先登录，再进行评论。

请先登录，再回答此问题。

Answer 1

Matt J 2024-5-2

0
链接

此回答的直接链接

https://ww2.mathworks.cn/matlabcentral/answers/2114111-layer-input-expectation-doesn-t-match-neural-network#answer_1451006

在 MATLAB Online 中打开

'ValidationData', {val_x_array_transposed', val_y_array_transposed'}

From the naming you have used here, it appears that you have transposed your validation data twice (and hence untransposed them).

5 个评论
显示 3更早的评论隐藏 3更早的评论

Matt J 2024-5-2

编辑：Matt J 2024-5-2

在 MATLAB Online 中打开

The array sizes you mention are reflected in the test code below, which gives me absolutely no errors. So, I think you just have to check that your arrays are truly the sizes that you say they should be,

inputSize = 3; 
hiddenLayerSize = 20; 
outputSize = 2; 
layers = [
    featureInputLayer(inputSize)
    fullyConnectedLayer(hiddenLayerSize, 'Name', 'fc1')
    reluLayer('Name', 'relu1')
    fullyConnectedLayer(outputSize, 'Name', 'output')
];
net = dlnetwork(layers);
opts = trainingOptions('adam', ... 
    'MaxEpochs', 100, ...
    'MiniBatchSize', 64, ...
    'ValidationData', {rand(81,3), rand(81,2)}, ... 
    'ValidationFrequency', 10, ... 
    'Verbose', true ...
     );
netTrained = trainnet(rand(189,3), rand(189,2), net,"mse", opts);
    Iteration    Epoch    TimeElapsed    LearnRate    TrainingLoss    ValidationLoss
    _________    _____    ___________    _________    ____________    ______________
            0        0       00:00:01        0.001                           0.77353
            1        1       00:00:01        0.001         0.72641                  
           10        5       00:00:02        0.001         0.79818           0.63803
           20       10       00:00:02        0.001         0.65905           0.52162
           30       15       00:00:02        0.001         0.54222           0.42349
           40       20       00:00:02        0.001         0.44389           0.34247
           50       25       00:00:02        0.001          0.3608           0.27561
           60       30       00:00:02        0.001         0.29201           0.22095
           70       35       00:00:02        0.001          0.2377           0.17871
           80       40       00:00:02        0.001         0.19629           0.14782
           90       45       00:00:02        0.001         0.16601           0.12666
          100       50       00:00:02        0.001         0.14579            0.1128
          110       55       00:00:02        0.001         0.13273           0.10406
          120       60       00:00:02        0.001         0.12485          0.098747
          130       65       00:00:03        0.001         0.11987           0.09504
          140       70       00:00:03        0.001         0.11638          0.092164
          150       75       00:00:03        0.001         0.11361          0.089783
          160       80       00:00:03        0.001         0.11121          0.087748
          170       85       00:00:03        0.001         0.10903          0.085977
          180       90       00:00:03        0.001         0.10699          0.084425
          190       95       00:00:03        0.001         0.10507          0.083059
          200      100       00:00:03        0.001          0.1033          0.081858
Training stopped: Max epochs completed
Y_pred_val = predict(netTrained, rand(81,3)); 
whos Y_pred_val
  Name             Size            Bytes  Class     Attributes

  Y_pred_val      81x2               648  single              