Issue with trainNetwork() function

I have imported a trainNetwork image into this question, and hopefully you can see to the right of it, where the simulation ends, as this is the most important part, if this is not visible after I post this question, I will resize and repost the image.

I seem to be getting an accuracy at the end of this simulation, that is performing low and is not following the 'darker blue' trend of this simulation.

Could anyone offer any clues as to what might be going on ?

I am trying to interpret what and how this darker blue field of the training plots means, and why my validation classification rate no longer seems to be tracking the darker blue training performance.

Any clues where I should look would be a great help.

Regards,

Eva Riherd

[Matt J]

if this is not visible after I post this question, I will resize and repost the image.

It is not necessary to repost an image if you just want to resize it. That can be done within the post editor.

Answer 1

[Matt J]

I am trying to interpret what and how this darker blue field of the training plots means,

The light blue line measures accuracy based on the current minibatch of training data that is being processed, and the dark blue is a time-smoothed version of that. The black line is the accuracy as measured from your validation data set.

I seem to be getting an accuracy at the end of this simulation, that is performing low and is not following the 'darker blue' trend of this simulation.

I don't think so. The training accuracy plot has some downward dips that, although infrequent, may be large enough to bring the average training accuracy downward, to a level consistent with the validation curve.

Possibly if you shuffle the training datastore, you might get a less spiky training accuracy plot.

[Eva]

Wow, you responded FAST!!

Much thanks for this response, I really appreciate it.

I <3 MATLAB !

[Matt J]

You're welcome, but if the advice worked, please Accept-click the answer.

[Eva]

Much thanks for taking the time to look into this question!