GRADE Reporting And Definition Environment

Inspired by the grading infrastructure developed by David Eckhardt at Carnegie Mellon University for http://www.cs.cmu.edu/~410/ , GRADE is a lightweight, file-based, directive-based approach to managing the work of grading students in a course. It uses a minimal markup format for both its machine-readable scoring rubric and for grader data files.

Student View

Students will get back tastefully formatted grade files. A very short example looks something like this:

Packaging: [10/10] (100.00%)

 Grader comments:

  Everything looks great here.  Thanks for the very informative README!

Functionality Tests: [30/40] (75.00%)

  (-6.0)
  The buffer-passing test seems to mangle bytes on occasion.

  (-4.0)
  There is a minor problem with the buffer-passing test output
  when given an unusually long input string.

 Grader comments:

  Both of these test failures occur because ...

TOTAL: [40/50] (80.00%)

Basic Grader Use

The instructor will have prepared a skeleton file which will look something like this:

# Basic features of the handin
#@packaging
 #:ftbfs_all
 #:missing_readme
 #:readme_no_commentary
 #:readme_no_instructions
 #:tarball_directory
 #:missing_make

$BEGIN_COMMENTS

$END_COMMENTS

# Automated test result section
# Un-comment the appropriate directive for each test failed.
#@tests
 #:ftbfs_all
 #:simple_test
 #:more_interesting_test
 #:more_interesting_test_minor
 #:test_everything

$BEGIN_COMMENTS

$END_COMMENTS

etc. The graders should copy this skeleton and mutate it appropriately for each assignment they grade. The example grade output above might have come from a grading data file like this:

@packaging
$BEGIN_COMMENTS

Everything looks great here.  Thanks for the very informative README!

$END_COMMENTS

@tests
:simple_test
:more_interesting_test_minor

$BEGIN_COMMENTS

Both of these test failures occur because ...

$END_COMMENTS

In more detail:

• Sections are separated by lines beginning with @.
• Blocks surrounded by $BEGIN_COMMENTS/$END_COMMENTS will be copied into the appropriate section of the student report under the heading of Grader comments.
• outside the comment blocks, all characters on a line after a # mark will be ignored. Inside comment blocks, all bytes will be copied over verbatim (modulo an indentation prefix for pretty-printing).
• Text and scores associated with each un-commented :-line (which are defined by the instructor) will also be emitted to the correct section.

Instructor Use

Constructing a Rubric

A rubric file, typically called defines.conf, consists of a number of sections. Each section has a name as well as some other parameters, and contains the definition of the flags seen in the grader data.

Sections are introduced with @-lines, like in the grading data, except that here, they take arguments:

@section-name type [extra] - friendly-name

where

• section-name is the short name as used in the grade data. It may not contain whitespace.

• type indicates to the grade program how to interpret this section. If type begins with !, the section will be omitted from the generated skeleton and this initial ! will be stripped from the type before consulting the following choices.

  There are a few base-case scoring engines:

  • The word simple defines a section of define flags whose invoked scores are simply summed. extra here should be the section's maximum value.
  • The word equal defines a section of equally-weighted flags; again, extra should be the section's maximum value.
  • The word seconly defines a section with no flags (unless possibly combined with one of the modifiers below) but instead just takes a literal score after the @section line in the data file. (The parser also accepts a literal ! in this position, in which case it will raise an error unless some modifier below overrides the score; this facilitates common rationale text even in this kind of section.)

  There are, additionally, some modifiers available:

  • The word bounding followed by (whitespace and) another type will behave as that type except that the score will be between zero and that type's derived maximum. That is, this section will behave as if it had that type but will yield no scores below zero and no extra credit.
  • The word nonneg behaves like bounding but bounds the section's score from below at 0; that is, it permits extra credit but not extra loss.
  • The word commenting followed by (whitespace and) another type will permit the definitions of flags with argument !C which will not influence the score at all and will not print out a score modifier before the flag text in generated reports.
  • The word zeroing followed by (whitespace and) another type will permit the definitions of flags with argument !0 which will set the section score to zero.

  Some shorthands are defined:

  • The word 0 is a shorthand for zeroing bounding commenting simple.

• friendly-name is the section heading as presented to students. It may contain spaces, and is in fact the remainder of the @ line.

Within sections, each flag definition takes the form

:flag-name score-modifier
Commentary paragraph 1
paragraph 1 line 2

paragraph 2
.

flag-name is the name of the flag used in the grading data files. score-modifier is defined by the section type. While most flags will be introduced with :, it is also possible to use ;; flags defined this later way are OK for multi-use (with the commensurate impact on the section's score), whereas :-defined flags will trigger an error if used more than once.

• For simple sections, the score-modifier may be
  • a number, which adjusts the score of this section by that many absolute points. As such, this number is almost always negative (i.e., beginning with a -), but positive numbers are understood for some form of extra credit. (Note that the script will refuse to set a score higher than the section maximum.)
  • a number followed by a % character, which will adjust the section score by that percentage of the maximum number of points available in the section.
  • The literal string !0. Engaging any whole number of flags so defined will set the section's score to zero.
• For equal sections, the only permitted non-empty score-modifier is !0, which is interpreted as in simple sections. All other flags in this type of section should have an empty score-modifier.

Text between the line beginning with : (or ;) and the dot on a line by itself will be copied into student grade reports whenever the flag is given in a grade data file. In many cases, there are many conditions that may merit the use of the same flag, and students will benefit from additional feedback about exactly what offense has been committed; historically, rather than introduce many flags, a simple "see the note below" in the prose has sufficed. Of course, this is up to judgement and taste.

Lines that begin with # and not #! will be copied into the skeleton. Lines beginning with #! will be ignored entirely, except for some additional advanced handling:

• #!\n (yes, a literal backslash) will cause an empty line to be emitted into the skeleton if the containing section is not being skipped.
• #!noskip will cause subsequent comment lines in a skipped section to to be emitted. In a non-skipped section, it has no effect.
• #!reskip will cause subsequent comment lines to be skipped if the containing section is skipped. It has no effect otherwise.

Continuing the example above, the corresponding defines.conf contains, among other defines

#! This line will be ignored
# Basic features of the handin
@packaging 0 10 - Packaging
:ftbfs_all -10
The submission failed to compile.
.

#! ...

# Automated test result section
# Un-comment the appropriate directive for each test failed.
@tests 0 40 - Functionality Tests

#! ...

:simple_test -6
The buffer-passing test seems to mangle bytes on occasion.
.

#! ...

:more_interesting_test_minor -4
There is a minor problem with the buffer-passing test output
when given an unusually long input string.
.

#! ...

Generating a Skeleton

Given a rubric, typically called defines.conf, one can produce a skeletal grading file by

grade make-skeleton < defines.conf > skeleton

Producing grade results

Given a rubric and a grader data file, student.data, one runs

grade grade-one defines.conf < student.data

to obtain the pretty-printed report and numeric score result.

It is easy to adjust the weights of different flags and sections by simply altering the values in defines.conf file and re-running the grade program.