However in this case the network will be to

Info iconThis preview shows page 1. Sign up to view the full content.

View Full Document Right Arrow Icon
This is the end of the preview. Sign up to access the rest of the document.

Unformatted text preview: e of the network when the training has converged. If using only the learning set, the network error will eventually converge toward zero. However, in this case the network will be to speed to the training set, and not able to generalise to point it has not seen before. In order to avoid that, the performance of the network is evaluated on the validation set after each iteration. When the network performance is still increasing on the test dataset, but stalled or decreasing on the validation dataset, then it means the network is starting to loose generalisation capabilities. It is then time to stop the training. Stopping after 4 epochs would not have decreased the network performance as the validation results are no longer improving after the 4th epoch. (b) Change the dataset definition so that the crab sex is represented as a binary value (1 = male, 2 = female). Does this change the classification performances? This should not change the classification performances. Differences observed in matlab are the results of random variations. 4 ÑÐ ×ØÖÒ ÑÔ ´ ß ¸ ³ Ñ Ð ³¸ ½µ ÑÐ ×ØÖÒ ÑÔ ´ ß ¸ ³Å Ð ³¸ ½µ ± Ò ÖÝ Ø Ö Ø× ÓÖ Ò ÙÖ Ð Ò ØÛÓÖ ×Ü ÓÙ Ð ´Ñ Ð µ³ Ô Ý× Ö× Ô Ý× Ö׳ ±ÖØ ÒÛ ÓÖÛ Ö Ò ØÛÓÖ ÒØ Ò Û ´Ô Ý× Ö×¸× Ü¸¾¼µ Ò Ø¸ØÖ℄ ØÖ Ò´Ò Ø¸Ô Ý× Ö×¸× Üµ Ø ×ØÁÒÔÙØ× Ø ×ØÌ Ö Ø× Ô Ý× Ö×...
View Full Document

This note was uploaded on 03/21/2010 for the course IT ANN taught by Professor Mario during the Spring '10 term at École Normale Supérieure.

Ask a homework question - tutors are online