To submit your solutions:
For Questions 1-2, 4, 5(b,c,d,e), 6, 7{The comparison part,b}, 8-9: submit in paper form to the course assignment drop box (on the first floor of CSC). Please ensure that your name, login, and
student ID are written on your assignment.
For Questions 3, 5(a), 7(a,c,d): First zip your codes and figures into a single file, then email that file to c466-submit@cs.ualberta.ca.
(Be sure to watch the newsgroup for possible further directions.)
You will need to use the code in
ZIP file.
Question 3
Datasets:
- The training data is randomly generated by the test_gp functions, so every time you make a run (a plot) you get different results.
You should hand in:
- The Matlab code (well commented!),
- The classification results. You can choose the experiment parameters on your own. Do it to get interesting plots for the conclusion (Please submit both .fig and .jpg files).
- plots for test_fn_1d_1.m function: fn_1d_1_plot_{1,2}.{fig,jpg}
- plots for test_fn_1d_2.m function: fn_1d_2_plot_{1,2}.{fig,jpg}
- plots for test_fn_2d_1.m function: fn_2d_1_plot_{1,2}.{fig,jpg}
- plots for test_fn_2d_2.m function: fn_2d_2_plot_{1,2}.{fig,jpg}
- the running times of the algorithms in seconds.
- a summary about the experiments with your conclusions in PDF format
Question 5
Dataset:
- The data set is the one used in Figure 2.1. of the textbook. It has 200 data points. The input is 200x2 dimensional (tr_p) and the output is 200x1 (label) which is either +1 or -1.
You should hand in:
- The Matlab file (well commented!)
- back_prop.m
The back_prop function will have two inputs P and T, in which P is the network inputs (nx2) and T (nx1) is the network outputs. Note that n is number of training points.
It will also have two outputs, IW and LW. IW is a 2x2 matrix for the wights that connect the input layer to
hidden layer and it should be in the form of [A,C;B,D]. LW is for the weights that connect hidden layer to output layer. It is 2x1 vector of the form [E;F].
Considering the bias nodes is not necessary here.
- The Results
- Q5a.{fig,jpg}
This should plot both the data points and the classification boundary. Use 'x' and red color for points with label +1 and 'o' and green color for the points with label -1.
[hint: you do not need to compute the boundary. The first dimension of inputs ranges from [-3,5] and the second dimension ranges from [-2,3]; therefore, you can just consider many points in the input space and plot a red dot for positive points and a green dot for negative points and a black dot for points near zero.]
Question 7
You should hand in:
- The Matlab files (well commented!)
- GradientDescent.m
Note that the specification of the inputs and outputs are available in the .pdf file of the assignment.
- Comparison of the behavior of three routines over the given initial set-ups.
- For this part, just add three columns to the given initial set-ups table and specify the number of iterations required for the algorithm to converge. Submit this table along with the value that was used for &kappa in the paper part of your submission.