Experiment Manager Regression task

I wanted to use the Experiment manager to train a network which takes in input an image of size 1xM and outputs 1 numerical value. My dataset is composed by N of such pairs and usually I feed the network using trainNetwork(TrainIn, TrainOut, layers, options), where TrainIn is a matrix [1 M 1 N] shaped and TrainOut is a numerical array with N elements, and it works. Also the validation set is shaped like this and I input it in trainingOptions using 'ValidationData',{VaIn,VaOut}.

I watched the "How to Set Up Your Own Deep Learning Experiments", and I thought it was enough to let the output of the function be

[ TrainIn, TrainOut, layers, output] = experimentFunction(params)

instead of just [ dataset, layers, output] = experimentFunction(params)

but when I run the Experiment Manager I get the error:

Error(s) occurred while validating the setup function: Caused by: Error using textscan

Invalid file identifier. Use fopen to generate a valid file identifier.

and this happens both if I explicit the validation set or not. I imagine that this happens because I am not using a datastore. If I manage to use a datastore for the TrainIn images, then, does it have to be a single datastore containing TrainIn and TrainOut or can I have a datastore input and an array of numbers output? If it must contain both the input and output set, how do I signal to the code which is which?

Thank you in advance.

Answer 1

[Anshika Chaurasia]

Hi Claudia,

You can refer to the following example:

https://www.mathworks.com/help/deeplearning/ref/experimentmanager-app.html#mw_8052d44f-fb6a-4472-92cd-da06404277e8

In the above example following function is used:

function [TrainIn,TrainOut,layers,options] = experimentFunction(params)

Moreover, I wasn't able to reproduce your error at my end as I don't have the datasets and script which you have used.

[Claudia Borredon]

I didn't try to run the experimentFunction on its own before using it in the Experiment Manager. I found out that the error was in the function, thank you again.

Answer 2

[ahmed shahin]

please could u help me to solve this error ?

Unable to determine if experiment is for classification or regression because setup function returned invalid outputs.

[JESÚS MARÍA]

I have that same problem. I need help too. Thank you.