In Deep Learning Toolbox, what input layer should I use for simple dataframe-type input?

In Deep Learning Toolbox, we can use imageInputLayer() and imageDatastore() for image-type input.

How about the simplest type of input: the dataframe (say an array)?

Which input layer should we use? Should we use datastore() for this type of input?

I don't see many tutorial about this type of input and I got error for the below code. Each of my data point contains 2 features (i.e. size 2x1x1), but when Matlab read the data store, it can only read 1 feature (i.e. 1x1x1).

Failed code

Error message: The training images are of size 1x1x1 but the input layer expects images of size 2x1x1.

data=[ ...
    -0.4    -0.8; ...
    -1.4    -1.0; ...
    -1.5    -1.7; ...
    -2.3    -2.0; ...
    -1.2    -1.1; ...
];
csvwrite('data.csv',data);
ds_features = datastore('data.csv');


layers = [
    imageInputLayer([2 1 1],'Name','in')
    fullyConnectedLayer(10,'Name','fc1')
    softmaxLayer('Name','sm1')
    classificationLayer('Name','cf1')];

lgraph = layerGraph(layers);

options = trainingOptions('sgdm', ...
    'MaxEpochs',4, ...
    'Verbose',false, ...
    'Plots','training-progress');

net = trainNetwork(ds_features,layers,options);

Answer 1

[Srivardhan Gadila]

In the above question I see that you haven't provided any target data(classification labels) for training the network.

To use an image datastore as a source of training data, use the imds argument of trainNetwork. To use all other types of datastore as a source of training data, use the ds argument of trainNetwork. To be a valid input for training or validation, the read function of a datastore (with the exception of ImageDatastore) must return data as either a cell array or a table.

For networks with a single input, the table or cell array returned by the datastore must have two columns. The first column of data represents inputs to the network and the second column of data represents responses. Each row of data represents a separate observation.

In case of input size 2x1 & a table, it should be something like below:

>>>data = read(ds)
data =

  4×2 table

        input         response
    ______________    ________

    {2×1 double}       7    
    {2×1 double}       7    
    {2×1 double}       9    
    {2×1 double}       9  

Refer to Datastores for Deep Learning, trainNetwork(ds,layers,options), fileDatastore

[SC]

Thanks for your reply. How do you create the ds as you mentioned? I tried the following, but I can only construct a structure, not a datastore. I want a datastore that can be input into my model.

ds_table.response{1}=5;
ds_table.response{2}=9;
ds_table.response{3}=-1;
ds_table.response{4}=6;

ds_table.input{1}=[2 3];
ds_table.input{2}=[-2 8];
ds_table.input{3}=[9 7];
ds_table.input{4}=[1 5];

ds_table

Answer 2

[Sean de Wolski]

I don't think deep learning is the right approach if your input data has two points. Consider using a standard machine learning technique and the classification learner app.

https://www.mathworks.com/help/stats/classificationlearner-app.html

[SC]

Thanks for your reply.

My intention is to use 2 features and several points (i.e. 4 or 5) in my example.

I used this data as a toy model, and I want to extend it into an input of n features and m data points.

Since I am exploring something with dataframe input and the effect of different deep learning architectures, I want to use the Deep Learning toolbox to achieve this.

Answer 3

[Yomna Genina]

Starting R2020b, you could use featureInputLayer for a dataset containing numeric features.

See this example for more details: https://www.mathworks.com/help/deeplearning/ug/train-network-on-data-set-of-numeric-features.html