CS698U (Computer Vision)
-
To load mnist data loadmnistImages.m and loadmnistLabes.m are used.
-
0 is mapped to 10 (for convenience as in octave there is no 0th index). And predictions from "mlp_predict" function maps 10 back to 0. Thus gives value between 0-9. To calculate cost on test set y is again mapped from 0 to 10.
-
To change the gradient method of change the last argument of "mlp_train" function in main.m in "Training MLP" section:
Select: "adaGrad" / "gd_momentum" / "standard_gd"
-
In "mlp_costAndGrad.m" file Feedforward and Backpropagation is implemented.
-
Run using command:
octave main.m