Final validation accuracy is different from validation accuracy data collected from traininfo

Hi,

I trying to figure out how does matlab come up with validation accuracy when the training is finished in training progress. As you can see in the figure below, the validation accuracy was 92.30%.

I tried to calculate the validation accuracy based on the data collected in traininfo, i found out that the value does not match. i tried to see the maximum validation accuracy based on the excel attached above, it does not match. i also tried to get the average value by sum all the value and divide by 200 also does not match. Even inside the traininfo data, there is no validation accuracy with 92.30%

Appreciate if someone could help explaining the formula for calculating the training accuracy based on the result from the training plot. Thank you.

Answer 1

[T.Nikhil kumar]

Hello Teo,

As per my understanding, you want to know how the validation accuracy is calculated after training is completed and also the reason for difference in final and intermediate validation accuracies.

The validation accuracy shown in the plot is often calculated during the training process using a separate validation dataset to evaluate the performance of the model during training. It is usually observed that, the training accuracy is lesser than the validation accuracy. The final validation accuracy is calculated after the last iteration when the network's statistics are updated.

Possible causes for this anomaly and calculation processes involved are discussed in the following attached questions on the forum.

• https://www.mathworks.com/matlabcentral/answers/2015506-the-difference-of-best-validation-point-and-final-point-at-training-process-plot
• https://www.mathworks.com/matlabcentral/answers/1615045-validation-accuracy-for-cnn-showing-different-than-in-the-plot

Hope this helps!