Just followed a machine learning example and programmed a gradient decent maximizaation to read in a 64X64X3 image vector to predict whether it was a cat or not. The model had a parameter for every pixel*color so 12288 parameters were optimized. It worked, a vector of parameter estimates were obtained and predictions were obtained and worked pretty well (prediction accuracy good). How did this work, when I only had 200 samples, so p >>n. I thought we could not estimate parameters in regression when p
Asked
Active
Viewed 133 times
0
-
prediction accuracy for the training set or the test set? – Him Jan 20 '18 at 20:49
-
1n=200 training samples... m=50 test set. Accuracy was decent on the test set. – Melanie Jan 20 '18 at 20:56
-
2Rather depends what was being maximized (or descended - gotta make your mind up :-) – conjugateprior Jan 20 '18 at 21:16
-
The cost function for a logistic regression. – Melanie Jan 20 '18 at 21:46