reset password
Author Message
jonathankroening
Posts: 39
Posted 21:11 Jan 29, 2014 |

I'm getting some different coefficients in scikit-learn's regression model and a pretty different y-intercept too than I have in my own model. Is that because scikit-learn's LinearRegression method uses batch gradient descent and my implementation performs step-wise updates on Theta (stochastic)?

msargent
Posts: 519
Posted 11:48 Jan 30, 2014 |

The stochastic algorithm shouldn't be really far off. Ng's later videos discuss stochastic gradient descent. But for this assignment you should implement batch gradient descent.