Why is this so much faster?

[Terry Reedy]

On 6/2/2011 6:07 PM, Keir Rice wrote:
> Hi All,
>
> The following function was showing up in my profiles as a large bottle neck:
>
> # Slow version
> def RMSBand(self, histogram):
> 	"""Calculates the root-mean-squared value for the given colour stream histogram."""
> 	intermediateResult = map(lambda (i, h): h*(i**2), zip(histogram, range(255)))
> 	totalSum = sum(intermediateResult)
> 	
> 	# calculate rms
> 	return math.sqrt(totalSum / self.Area())
>
> So after a bit of trial and error I came up with the following function which is a lot faster:
>
> # Fast version
> def RMSBand(self, histogram):
> 	"""Calculates the root-mean-squared value for the given colour stream histogram."""
> 	totalSum = 0
> 	for i, h in enumerate(histogram):
> 		totalSum += h*(i**2)

> 	# calculate rms
> 	return math.sqrt(totalSum / self.Area())
>
> My question is why is the second method so much faster?
> Is it the removal of the extra function calls?

Yes. Map is only 'fast' when one already has a function and is going to 
call it repeatedly regardless of the other code. When one has an 
expression, wrapping it as a function to use map is surely slower. Have 
you tried

   return math.sqrt(sum([h*i*i for i,h in enumerate(histogram)])
     / self.Area())

or same without [] brackets?

i*i should be faster than i**2 in any version.

> Is it skipping the creation of a list?

A bit.

See Tim's response.

-- 
Terry Jan Reedy