View Forum Topic

Forums
CS461
Make sure you have the latest Gist for Neural Nets

Author	Message
msargent Posts: 519	Posted 13:13 Mar 04, 2014 \| In particular, early versions had the right cost function value with the trained weights with regularization only. Now it includes what the value should be if the function isn't regularized: "Cost at parameters loaded from ex4weights.mat. (This value should be about 0.383770 with regularization, 0.287629 without):" Also, the second argument used in the first randomization function call for the thetas was originally 15, it should be .12. Just make sure you have the latest gist and you will be okay.

Author

Message

msargent

Posts: 519

Posted 13:13 Mar 04, 2014 |

In particular, early versions had the right cost function value with the trained weights with regularization only. Now it includes what the value should be if the function isn't regularized:

"Cost at parameters loaded from ex4weights.mat. (This value should be about 0.383770 with regularization, 0.287629

without):"

Also, the second argument used in the first randomization function call for the thetas was originally 15, it should be .12. Just make sure you have the latest gist and you will be okay.