I haven't tested Jo solution, but just wanted to mentioned that with Alok's one looping a hundred time on this method is about the same speed of treating it once (around 6secs on my machine.)
On 27 July 2012 18:06, Alok Gandhi <[email protected]> wrote: > Hi Jo, > > > *"This is the purpose of the DOD paradigm to think about how the data > should be input and how it should be treated. This would be too > pretentious from me to say it has been used here, but trying to treat the > data the more linearly possible is a good start, it saves memory and > speed. I can bet easily, speed increases massively on yours or Jeremie's > if lot more than 2 or 3 parameters are treated compared to mine. One of > the effects of using DOD is you're ending by processing the data linearly > which avoid any query next and make your code really fast and reactive. "* > > The DOD is a nice pattern , I agree. > > *"About the function calls thing, I would say, that making one function > call instead of many would not influence so much speed results. A function > is a simple object as many other stored on the heap once created and > called with arguments (other objects). There is no special initialisation > nor building done each time. But even by considering this, and making my > function called with xsi parameters passed one by one, this is still faster > (20perc on my machine) and this for many different other reasons."* > > I would like to differ here, please look at the simple logic or semantics. > If want to grab data from XSI, it takes some time to query the data. If i > query once and store it in a hash, do not have to query it again, I simply > look at my stored hash which is way faster. If you wrap a function which is > trying to query data inside of itself, every time you make a call to the > function you adding the overhead by the data query. As simple as that. You > can see the this has already worked for Jeremy from his reply above. > > *"*If I didn't chose any dictionaries based implementations as you did, > this is because the python hash table implementation is generic and use lot > of memory with lost of blocks where py list are implemented as linked list > and permit really fast appending, O(1).*"* > * > * > In this case here, the reason to use dictionary was for faster look ups. > For look ups specially dictionary are a way to go for me always. I would > rather use a dictionary then to store in a list/tuple and then have to do > dirty labmdas or other stuff to retreive by list.index() and what not.You > really can't beat the python hash table. They may have more memory > requirements, but these days, it really is not a problem. For tasks such as > these, I suspect it really would not make impact even on a machine with 8 > gigs. The hash table are good and really good for one thing, fast look ups. > No wonder Google uses python as their main scripting languages and the > dictionary is one good reason for it. You can easily verify this on > Official Google Developers Videos on You Tube > http://www.youtube.com/watch?v=haycL41dAhg. > > Cheers ! >
