How to View the Visualization Graph of Machine Learning

Test training data from training data

Basically, this is a recognition, but please note that we do not re-test using training data after learning.
We always calculate errors during learning.
Because it is learning to modify the model to reduce errors.
The blue line represents the error per epoch (left graph: loss) and the percentage of correct answers (right graph: accuracies).

Tested with training data

I think there is no problem with my understanding.
The orange line represents the error or correct answer rate when tested with test data for each epoch.

Please note that the error calculation method is not only the mean square sum error.
The method of error calculation depends on the learning algorithm and the purpose of the model.
For example, cross-entropy errors are often used in neural network classification problems.
From this graph, we don't know which calculation method we used.