Raster processing in arcpy (performance issues)

SebastianSantibanez

Hi,
I have created a piece of code that allows me to create a cost distance raster per feature (point) per feature class in a GDB iterating firstly the FCs (applying 'for' in FC list) and then the rows (using cursor).  The script works fine but it takes an overwhelmingly long time to complete the process (40 hours for ~15,000 features). 
- Is there any way to speed up the script?
- Somebody suggested me that the cursor might be the cause of the performance I am getting. Is this possible? if so, how can I fix it?
- Another approach that I am considering is using another function to calculate the distances (not arcpy) does anybody know another python module that I could use instead?

Thanks so much.

NilsBabel

Things have changed a lot at 10.x for Spatial Analyst. Before you look for fancy alternatives, just refactor it for the new Python Grid Algebra.

The key improvement that stands out here is that the grid is held in memory, so you could iterate over the same grid without reloading it each time, and the output is also held in memory until you want to save it.

But if that is still to slow, could numpy be programmed to run the same processes?

Cup of Coffee Rule
If any process takes longer than a cup of coffee, interrupt it and fix it or find a better way.

I would never accept a model that takes 40 hours. What if you have to re-run it?

Heads up, your raster objects created in arcpy (v10) are not held in memory.  They are temporary rasters that are written to disk.  They disappear when your script completes but they are written to disk (either in your current workspace or scratchworkspace).  Check it out next time you're running a script.  Open up a file browser and refresh it while the script is working with raster objects, you'll see some strangely named rasters appear in your folder and then disappear.  Apparently, in 10.1 you can create rasters in in_memory.  That should be nice.

Good luck,

Nils