Description
GOAL: This a short exercise to learn how to measure and fit performance to curves. Also to present these result in graphical form. On the EC504 GitHUB at Plotting and Fitting there is information on how to use gnuplot and even in LittleGnuplot.pdf instructions on how to install gnuplot in your laptop. This is convenient but all can be done easily using gnuplot on the SCC through the SCC OnDemand interface. Use Slack to exchange helpful hints on graphing!
1. The main file on GitHub has a range of sorting algorithms and the ability to construct random lists of varying sizes N. The exercise is to run them for a range of the sizes N = 16, 32, 64, 128, · · · , average over at an ensemble of cases (as much as 100). to get good statistics. Report the average behavior a function of N for each sorting methed as to determine the scaling empirically as function of the number the mean number of swap operations vs N.
The main code sorScaling.cpp runs the example:
The exercise it to make a table of average performance for all 4 algorithm and plot them to see how the scale with N.
For the standard O(N2) search algorithms involve local (nearest neighbor) exchanges of element of the given list
Alist = a[0], a[1], a[2], a[3], · · · , a[N − 1] (1) You should find for insertonSort you should verify empirically average the algorithm would
insertionSort mergeSort quickSort shellSort
have
For mergeSort it should be exactly Θ(NlogN) and for quickSort on average O(NlogN).
Number of Exchanges = N (N − 1) (2) 4
Finally see if you can find the value of the γ for shel sort O(Nγ).
The exercise it to modify the main file to build an out put data file to plot.
# N insertionSort mergeSort quickSort shellSort
16 xxxx 32 xxxx 64 xxxx 128 xxxx ....
xxxx xxxx xxxx xxxx xxxx xxxx xxxx xxxx
xxxxx xxxxx xxxxx xxxxx
1
where xxxx are the average values. This is convient for using gnuplot to plot and fit the curves.
This output file can be make by a hack by printing to the standard output. Just run the code in a terminal (aka shell) with $./sort > datafile.txt Then you take what you need using an editor. This is useful quick trick, however you should really set up a separate output file. This is necessary if you want submit you code in queue. To set up a output file see the example to do this on GitHub at HW1_codes/makeSortedList.cpp (Hey basic software technique. Steal method from other codes!)
The basic commands in this code even allow naming the file with a parameter!
// open file char FileName[80];
sprintf(FileName,"MySorted%d.txt",ListSize);
ofstream outfile(FileName);
// put stuff in this file
outfile << a[i] << endl;
// close file outfile.close();
Place your final source code, outfile and figures with fits in directoryHW2. Include the makefile so we can compile and test it.
Extra Credit: If you have time you could add error bars to the average (called σ) defined as mean square deviation a second column next to the tabulate averages xxxx. These are define for each algorithm and size N by
1 Ntrials − 1
Ntrials
(Swaps[i] − AvergeSwap)2
σ2 =
where above we suggested fixing Ntrials = 100. The average numbers of swaps in the 100
trials for each algorithm and size N in the table are:
1 Ntrials
AvergeSwap = Swaps[i] Ntrials i=1
You will want to have your code compute the standard error and put into another column in your output file. By the way all these analysis skill will likely come in handy for the team project.
For general background information the sorting.h files has a few more sorting algorithms to play with. We could add others like bucket and improve pivots for quicksort etc.
i=1
2
