ANN using Deep Network Designer with excel file as input and output values.

Hi. I am trying to create an Artificial Neural Network in Deep Network Designer where there will be input layer, hidden layers, output hidden layers and an output layer. My input values are a 8C x 368R matrix and my output value is a 1C x 368R matrix. I would like to train my values with a 70% training data, 15% validation data, and 15% testing data just like ANN. And also, I would like to train it using Levenberg - Marquardt algorithm. But, I won't use the >>nnstart tool. Because I will use it just for the validation of my data. I would like to use Deep Network Designer to simulate an >>nnstart like feature. Is there any way that I can import excel data file in Deep Network Designer to run an ANN feature? Thank you.

Answer 1

[Prateek Rai]

To my understanding, you want to import data from excel file and create an artificial neural network in Deep Network Designer. According to the information provided, you preferably need 1D convolution layer to build your ANN.

As of MATLAB R2021a, Deep Learning Toolbox does not support 1D-convolutional layer.

As a potential workaround you can try using "imageInputLayer" as described here.

To use "imageInputLayer" with 1D data you will need to structure your data in one of these two ways: [1 x T x C x N] or [T x 1 x C x N].

where:

• T is the sequence dimension
• C is the channel or variable dimension
• N is the observation dimension

Following link might also be useful for further reference. Moreover, you can refer to build-networks-with-deep-network-designer MathWorks documentation page to find more on building networks with deep network designer. You can also refer to import-data-into-deep-network-designer MathWorks documentation page to learn more on how to import data into deep network designer.