Performance "Tuning": running in 1/100th the time

Hypothetical Step 1: Profile

My first thought was to run the profiler and see what popped up as a possible root cause of slowness.

Generally, there are three kinds of performance problems.

• Wrong data structure.

  Or wrong algorithm. (They're simply opposite sides of the same coin.) This generally leads to dramatic, earth-shattering improvements.

• Piss-Poor Programming Practices.

  This kind of fine-tuning yields minuscule improvements. In some cases, no measurable change at all.

• Bugs.

  Every time I've been asked to improve "working" code, I find that it has bugs. Every time. These can exacerbate performance problems. Interestingly, most of these are readily apparent during an initial survey: i.e., while simply reading the code to see how it works. Trying to create unit tests for the purposes of refactoring often reveals additional bugs.

I didn't know what I was going to see in this 400-line program.

Hypothetically, profiling will help show what kind of problems we have.

Prior to profiling, of course, we need to run the darn thing. Thankfully, they provided some typical input files. For example, 1979 High School Football season results. 159 lines of patiently gathered teams and scores.

It Didn't Run

When we tried to run it with the profiler, we found that we had a bug. Tracking down the bug, revealed the essential performance problem, also.

The core problem was a failure to make use of Python data structures.

This manifested itself in a number of really bad design errors. Plus, it presented as the actual, show-stopping, serious bug.

In order to work out team rankings, the program kept a list of teams.

I'll repeat that for folks who are skimming.

In order to work out team rankings, the program kept a list of teams.

Not a dict mapping from team name to team details. But a simple list. Locating a team in the list meant iterating through the list, looking for the matching team. Really.

for t in team_list:
   if t['name'] == target_name:
       process( t )

This kind of thing was repeated in more than one place.

And one of those repeats had the bug in it.

What we have here is "wrong data structure". Replacing a list with a dict will have earth-shattering impact on performance.

The Bug

The bug, BTW, was a lookup loop which had the additional requirement of adding missing teams. It tried to use the for-else structure. This was the intended code (not the actual code).

for t in team_list:
   if t['name'] == target_name:
else:
   t['name']= init_new_team(target_name)

This is a first-class bit of Python. An else clause on a for statement is a real feature of the language. However, it's obscure enough that it's easy to get wrong.

However, it's also unique to Python, and the kind of thing that can lead to confusion. I discourage it's use.

Test-Driven Reverse Engineering

We're doing TDRE on a little piece of recreational programming. This means that we need to fabricate unit tests for code that is purported to work. Some folks like to call this "exploratory testing" (a phrase which is contradictory, and consequently meaningless.) We're "exploring". We're not "testing".

Once the core bug is fixed, we can run sample data through the application to get "big picture" results. We can extract bits from the sample data and exercise various functions in isolation to determine what they actually do now, bugs included.

Since this is so simple--and we're going to review it during a 2-hour meet up--and there's nothing billable going on--we can get by with a few really simple unit tests. We'll run the darn thing once to create some expected output. We save the output and use that single output log as the expected results from each refactoring step.

More complex applications require more unit tests. For a billable TDRE effort last year, I had to create 122 unit tests to successfully refactor a program of over 6,000 lines of code. It took several weeks.

Real Step 1: Fix The Data Structure

Before profiling (but after running to capture some output) we have to fix the essential data structure. Repeated scanning of a list is never going to perform well.

The whole point of the study of "Data Structures" is to prevent (or optimize) search.

In this case, we can prevent linear search of a list by using a dictionary. That, clearly, is job one.

It was a pervasive rewrite. Essentially, everything in this little program included a goofy loop to lookup a team in the list. The Elo algorithm itself, which is already O(n²), is dragged down by using the linear search for the opposing team four more times, making it O(n³).

Cyclomatic Complexity

One of the big "issues" is the use of if statements throughout the scoring algorithm. An if statement creates Cyclomatic Complexity and can lead to performance problems. Generally, if statements should be avoided.

This algorithm applies some normalization factors to reconcile scoring with win/loss numbers in different sports. Basketball, specifically, involves generally high scores. Since there are 2-point and 3-point scoring opportunities, a factor is used to normalize the points into "goals". Football, similarly, has numerous scoring opportunities with values of 1, 2, 3 and 6 points; the scores here are also normalized.

This normalization was done with an if statement that was evaluated deep inside the Elo algorithm. Repeatedly. Evaluated. The two functions that handled the normalizations, plus the normalization factors, are ideal candidates for OO design. There's a clear hierarchy of classes here. A superclass handles most sports, and two polymorphic subclasses handle football and basketball normalization.

The if statement is now "pushed up" to the very beginning of the program where an instance of the sports normalization object is created. This object's methods are then used by the Elo algorithm to normalize scores.

Icing on the Cake

Once we've fixed the bug and replaced a list with a dict, everything else is merely icing.

Some other OO changes.

1. The "Team" information should not be a flat, anonymous dictionary. It should be a proper class definition with proper attributes. There aren't many methods, so it's easy to create.
2. The "Game" information is read by csv.DictReader. However, it should not remain a simple, anonymous dict. As with a Team, a simple class can be created to handle Game.
3. The overall structure of the application needs to be broken into two sections. The command-line interface parses options, opens files, and generally gets everything set up. The actual ranking algorithm should be a function that is given an open file-like object plus the Sport object for normalization. This allows the ranking algorithm to be reused in other contexts than the command-line (i.e. a web service).

A more subtle OO design point is the question of "mutability". A Team in this application is little more than a name. There are also numerous "stateful" values that are part of the Elo algorithm. A Game, similarly, is an immutable pair of teams and scores. However, it has some mutable values that are part of the Elo algorithm.

Really, we have immutable Team and GameHistory objects, plus a few values that are used as part of the Elo calculation. I'm a big fan of disentangling these mutable and immutable objects from each other.

I suspect that the Elo algorithm doesn't really need to update the "state" of an object. I suspect that it actually creates (and discards) a number of immutable candidate ranking objects. The iteration that leads to convergence might be a matter of object creation rather than update. I didn't make this change, since it required real work and we were out of time.

Bottom Line

The more times I do TDRE to improve performance, the more I realize that it's all about bugs and data structures. This recreational application took 45-60 seconds to process one year's record of games for a given league. It now takes less than 0.2 seconds to do the same volume of work. Two test cases involving a complete run of 159 records runs in 0.411 seconds. That's 1/100th the time simply from switching data structures.

The idea of "tweaking" a working program to improve performance is generally misleading. It might happen, but the impact is minuscule at best.

Here's the checklist for getting 100:1 improvements.

• Remove searches.
• Remove deeply-nested if statements.

Generally, reduce Cyclomatic Complexity.

This also falls nicely into the “anti if campaign”...

Robert Lucente<noreply@blogger.com>

2012-08-24 19:52:30.407000-04:00

This also falls nicely into the “anti if campaign” genra http://www.antiifcampaign.com/

"The Anti-IF Campaign"