Regenerating training verbose table from saved traininfo

hi,

i trying to regenerate verbose table as shown below from the traininfo that i saved earlier after finish running the model.

However, when i treid to tabulate the data from traininfo using struct2table i get the following table which is completely different from original verbose table that we got after finish training the model in matlab.

Hope someone could share some help on contructing the same verbose table as the shown in the first figure. Thank you very much.

Answer 1

[yanqi liu]

yes，sir，if you want save the command information，may be use

diary(log);

diary on;

run your code

diary off;

finally，you will get log file，it save the command information

[Teo]

Hi @yanqi liu,

do you mean the run the following code before training of the model begin? Currently i already train the model and save the traininfo only. not sure how to tabulate back the data same the the one in first figure.

[yanqi liu]

yes，sir

if already run over，may be use traininfo to plot and view data

but its element is use for per training iteration，so i think it display not the same as the command

[Teo]

yea @yanqi liu, it is ok per iteration but the problem is the tabulated data is just in array format instead of the data itself as in the figure 2 above.

[yanqi liu]

s = fields(traininfo);
for i = 1:length(s)
    eval(sprintf('traininfo.%s=traininfo.%s(:);', strtrim(s{i}),strtrim(s{i})));
end
T = struct2table(traininfo);

[Teo]

Thanks @yanqi liu, your code really help solve my the issue i facing. Just curious when i tabulate the data may i know why some iteration having a 'NaN' value for both validation loss and validation accuracy?

[yanqi liu]

yes，sir，the loss NaN is common during train steps

may be the input data、learn rate and so on

for more information，may be check

https://stackoverflow.com/questions/56218256/nan-values-in-loss-in-keras-model#:~:text=If%20you%20are%20getting%20NaN%20values%20in%20loss%2C,function%20domain%2C%20then%20determine%20what%20those%20inputs%20are.

[Teo]

Thank you very much @yanqi liu for your explaination. It helps a lot.