Evaluator

The evaluation of a solution is done by a program, called evaluator. The evaluator is the most important part of a problem: it prepares the input sent to the solution and judges its output, producing a feedback.

To test a solution, the evaluator has to start a solution process, using the TuringArena library. Then, it can communicate with the process by calling directly the functions implemented by the solution. The evaluator can start a solution process as many times as it wants.

We'll see later how the communication is realized.

Evaluator of sum_of_two_numbers

In this example problem, the evaluator is written in Python, in the file evaluate.py.

It tests the solution on several pairs of numbers a and b (called test cases), checking if the result of sum(a,b) is actually a+b. It performs a fixed number of iterations, one per test case. In each iteration it does the following.

  • It generates a random pair of numbers a and b.
  • It starts a new solution process, and communicates with it.
    • It calls the sum function, passing the values of a and b just generated, and stores the result in a variable c.
  • If sum behaves correctly and returns the value c == a+b, the evaluator considers this test case passed.
  • If there is an error in the solution process, say, the solution attempts to open a file, executes a disallowed system call, or takes too long to answer, then the evaluator is notified by TuringArena, and considers the test case failed.
  • If sum returns a value which is not the sum of a and b, the test case is also considered failed.

The evaluator reports the outcome of each test case. At the end, the evaluator marks the goal correct as achieved if all the test cases are passed.

In general, a problem may have several goals, which are achieved depending on, say, how many functionalities are implemented, the quality of the outputs, and the computational efficiency of the solution.

Try it yourself!

Try to modify the file evaluator.py. Some suggestions:

  • Change the number of test cases.
  • Make the numbers a and b smaller/larger.
  • Make the evaluator stop as soon as a test case is failed.
  • Test the solution on negative values.