reset password
Author Message
steveshim
Posts: 6
Posted 13:14 Feb 13, 2016 |

Is it possible to get the values for around what our cost function values should be before gradient_descent and after? I want to know if I'm on the right track.

msargent
Posts: 519
Posted 15:22 Feb 13, 2016 |

Probably the best way to find out if you are on track is to graph the learning curve, if you get the curve decreasing rapidly until it asymptotes at a value lower than your original value (the value with the thetas initialized to zero), things are looking good. 

Also, your cost value with thetas initialized to zero should be higher than with thetas trained by gradient descent. 

Last edited by msargent at 15:22 Feb 13, 2016.
lmann2
Posts: 156
Posted 15:57 Feb 13, 2016 |

So our final graph should be asymptotic (1/x where x > 0) ?  It's very hard to tell if our calculations are right...

msargent
Posts: 519
Posted 16:37 Feb 13, 2016 |

It should look like the upper right quadrant of 1/x. 

lmann2
Posts: 156
Posted 17:26 Feb 13, 2016 |

Relief! lol