Accepted Solutions
Lesson 2... *tif file in … info out.
First the verbose code that produces the outcome, followed by explanations
fn = "c:/Temp/InRaster/InRaster.tif"
import arcpy
a = arcpy.RasterToNumPyArray(fn)
uniq, cnts = np.unique(a, return_counts=True)
per = cnts[:-1]/np.sum(cnts[:-1]) *100.
dt = [('Value', '<i4'), ('Count', '<i4'), ('Percent', '<f8')]
sze = cnts[:-1].size
out = np.zeros(sze, dtype=dt)
out['Value'] = uniq[:-1]
out['Count'] = cnts[:-1]
out['Percent'] = perWe don't need to know the coordinates of the lower left, the number of rows and columns or the nodata value, since we can get that when we read in the raster.
RasterToNumPyArray—ArcPy Functions | ArcGIS Desktop
RasterToNumPyArray (in_raster, {lower_left_corner}, {ncols}, {nrows}, {nodata_to_value})
Basic information on the array shape and data type
a.shape # ---- number of rows and columns
(84, 55)
a.dtype # ---- data type of the array, unsigned 16bit integer
dtype('uint16')
# ---- the nodata value is 65535, which is the maximum integer value for uint16
# ---- get this infor from `uniq`
uniq
array([ 1, 2, 3, 4, 65535], dtype=uint16)
# ---- counts return a 64bit integer, the 2098 is the nodata cell count
cnts
array([1057, 1186, 274, 5, 2098], dtype=int64)
# ---- At this stage we want to remove our nodata values from the counts and % calculationsLine 5 in the first code block
per = cnts[:-1]/np.sum(cnts[:-1]) *100.basically means " use all the cnts up to but no including the last value, divide it by their sum and multiply by 100%.
cnts[:-1] ... is a slice, the : means everything up to but excluding... Read up on array slicing... This is only a 2D array, but it can be head melting for arrays beyond 4D.
You will notice that I did slicing in lines 10 and 11 to exclude the no data values, but I didn't need to do if for the 'per' line because I sliced already.
Of course, the above can be put into a 'def' (aka function) if you need to do it all the time, and the code can be simplified greatly... Just to show you how working with basic arrays is easy.
When you want to work with things like area... then it becomes important to know the cell size of each array element. That is where 'human-ware' comes into play (ie, take notes) and sometimes exporting rasters to old-school ASCII rasters is useful since the cell size information goes out with it. RasterToNumPyArray, doesn't... you have to remember..
So you want to save stuff? I use numpy as well.
out_fn = "C:/temp/InRaster/in_raster.npy"
np.save(out_fn, a)Hmmm maybe I should look at it again, just in case
a_gain = np.load("C:/temp/InRaster/in_raster.npy")
np.all(a == a_gain)
TrueAll is good! 
I can check when I get back, there are a few tricks when moving from structured to non-structured not covered in my example.... lesson 2 coming soon 
Lesson 2... *tif file in … info out.
First the verbose code that produces the outcome, followed by explanations
fn = "c:/Temp/InRaster/InRaster.tif"
import arcpy
a = arcpy.RasterToNumPyArray(fn)
uniq, cnts = np.unique(a, return_counts=True)
per = cnts[:-1]/np.sum(cnts[:-1]) *100.
dt = [('Value', '<i4'), ('Count', '<i4'), ('Percent', '<f8')]
sze = cnts[:-1].size
out = np.zeros(sze, dtype=dt)
out['Value'] = uniq[:-1]
out['Count'] = cnts[:-1]
out['Percent'] = perWe don't need to know the coordinates of the lower left, the number of rows and columns or the nodata value, since we can get that when we read in the raster.
RasterToNumPyArray—ArcPy Functions | ArcGIS Desktop
RasterToNumPyArray (in_raster, {lower_left_corner}, {ncols}, {nrows}, {nodata_to_value})
Basic information on the array shape and data type
a.shape # ---- number of rows and columns
(84, 55)
a.dtype # ---- data type of the array, unsigned 16bit integer
dtype('uint16')
# ---- the nodata value is 65535, which is the maximum integer value for uint16
# ---- get this infor from `uniq`
uniq
array([ 1, 2, 3, 4, 65535], dtype=uint16)
# ---- counts return a 64bit integer, the 2098 is the nodata cell count
cnts
array([1057, 1186, 274, 5, 2098], dtype=int64)
# ---- At this stage we want to remove our nodata values from the counts and % calculationsLine 5 in the first code block
per = cnts[:-1]/np.sum(cnts[:-1]) *100.basically means " use all the cnts up to but no including the last value, divide it by their sum and multiply by 100%.
cnts[:-1] ... is a slice, the : means everything up to but excluding... Read up on array slicing... This is only a 2D array, but it can be head melting for arrays beyond 4D.
You will notice that I did slicing in lines 10 and 11 to exclude the no data values, but I didn't need to do if for the 'per' line because I sliced already.
Of course, the above can be put into a 'def' (aka function) if you need to do it all the time, and the code can be simplified greatly... Just to show you how working with basic arrays is easy.
When you want to work with things like area... then it becomes important to know the cell size of each array element. That is where 'human-ware' comes into play (ie, take notes) and sometimes exporting rasters to old-school ASCII rasters is useful since the cell size information goes out with it. RasterToNumPyArray, doesn't... you have to remember..
So you want to save stuff? I use numpy as well.
out_fn = "C:/temp/InRaster/in_raster.npy"
np.save(out_fn, a)Hmmm maybe I should look at it again, just in case
a_gain = np.load("C:/temp/InRaster/in_raster.npy")
np.all(a == a_gain)
TrueAll is good!
Let me know if you are having any troubles or if there is something you use often in your workflow...there is always lesson 3
