Fix bug caused by refactoring, add comment on Octave vs. MATLAB

Author: danluu

README.md

@@ -2,17 +2,20 @@ These are solutions to the exercises up at the [Stanford OpenClassroom Deep Lear
 
 Apologies for the mess. I'll clean this up when I have some spare time. Pull requests welcome, of course.
 
+I've attempted to make this Octave compatible, so that you can run this with free software. I've done this through the self-taught learning exercise, and it seems to work, but the results are slightly different. One side effect of this is that I'm using [fminlbfgs instead of minFunc](http://ufldl.stanford.edu/wiki/index.php/Fminlbfgs_Details)
+
 Here's the order of the exercises:
 #### [Stanford OpenClassroom Deep Learning class](http://openclassroom.stanford.edu/MainFolder/CoursePage.php?course=DeepLearning)
-1. linear.m
-2. multiple.m
-3. logistic.m
+1. `linear.m`
+2. `multiple.m`
+3. `logistic.m`
 
 #### [Unsupervised Feature Learning and Deep Learning Tutorial](http://ufldl.stanford.edu/wiki/index.php/UFLDL_Tutorial)
-1. Sparse Autoencoder: sparseae_exercise/train.m
+1. Sparse Autoencoder: `sparseae_exercise/train.m`
 2. Vectorized Implementation: sparseae_exercise/train.m (`1` is already vectorized)
-3.1. PCA in 2d: pca_2d/pca_2d.m
-3.2. PCA: pca_gen/pca_gen.m
-4. Softmax Regression: softmax_exercise/softmaxExercise.m
-5. Self-Taught Learning: stl_exercise/stlExercise.m
+
+3.1. PCA in 2d: `pca_2d/pca_2d.m`
+3.2. PCA: `pca_gen/pca_gen.m`
+4. Softmax Regression: `softmax_exercise/softmaxExercise.m`
+5. Self-Taught Learning: `stl_exercise/stlExercise.m`
 

stl_exercise/stlExercise.m

@@ -86,9 +86,9 @@
 
 
 [opttheta, cost] = fminlbfgs( @(p) sparseAutoencoderCost(p, ...
-                                  visibleSize, hiddenSize, ...
+                                  inputSize, hiddenSize, ...
                                   lambda, sparsityParam, ...
-                                  beta, patches), ...
+                                  beta, unlabeledData), ...
                              theta, options);