Simple C++ question with floating-point tolerance This question demonstrates the use of a per-test-case template grader that checks the output from a C++ program to see if it matches the expected result to within a given absolute tolerance. It is just a simplistic proof of concept. In particular:

  1. It assumes the output from the student's program is a single line containing just one floating point number.
  2. It uses a per-test-case grader which results in the student's code being compiled and run separately for each test case. This is very inefficient - a full combinator template grader should really be used, but that adds significant extra complexity.

The question posed to the student for demonstration purposes is just the following:

Write a C++ program that reads a single floating point number from stdin and computes its square root using Newton's method to an accuracy of 0.0001. Print your computed result. You may assume the input value is a positive float. 

]]> 1 0 0 python3 0 1 10, 20, ... 0 0 0 18 100 0 #include #define EPSILON 0.0001 using namespace std; int main() { double n; cin >> n; // We'll assume this gives us n > 0. // Newton's method for square root. double x = n; // Initial guess. double last_x = -10; // Silly value to get things rolling. while (abs(last_x - x) > EPSILON) { last_x = x; x = 0.5 * (x + n / x); } cout << x << endl; return 0; }]]> 1 cpp TemplateGrader 1 1 None 0 {} 0 0 0 10240 1 0 4.0 2.0 SHOW 100.0 10.0 SHOW 2.0 1.4142135623730951 SHOW 27 5.196152422706632 SHOW 99.98 9.998999949995 SHOW