Using .mat files to train convolutional NN - overfitting issue?
2 vues (au cours des 30 derniers jours)
Afficher commentaires plus anciens
Hi all,
I create .mat files out of my 3D dicom images. This is one step of my pipeline to work with a convolutional NN.
In the begining, I noticed that creating the .mat files once saves me time to train the NN with different kernel sizes.
Ex: for a 3x3 kernel size, I first create the imds with brand new .mat files.
Then, I clear all the variables and make the correspondent changes in my network to approach the imds with 5x5 kernel.
I only have to indicate the path to the imds (with the already created .mat files). Here, I noticed that the training step, even by clear all variables, it is faster than for 3x3.
Would it be the case MATLAB 'learned' the previous created .mat files, leading to an overfitting issue?
Thanks in advance.
0 commentaires
Réponses (1)
Rishik Ramena
le 13 Jan 2021
If you used a 'clear all' command, there's no way the previous created files and their contents(variables) are still in memory. The reason for a faster run in case of the 5x5 kernel could be different.
0 commentaires
Voir également
Catégories
En savoir plus sur Image Data Workflows dans Help Center et File Exchange
Produits
Community Treasure Hunt
Find the treasures in MATLAB Central and discover how the community can help you!
Start Hunting!