GRASS Python Scripting Library: Difference between revisions

From GRASS-Wiki
Jump to navigation Jump to search
mNo edit summary
mNo edit summary
Line 2: Line 2:
See also [[GRASS and Python]] for more general info.
See also [[GRASS and Python]] for more general info.


The code in {{src|lib/python/|lib/python}} provides <tt>grass.script</tt> in order to support GRASS scripts written in Python. The {{src|scripts}} directory of GRASS 7 contains a series of examples actually provided to the end users (while the script in GRASS 6 are shell scripts).
The code in {{src|lib/python/|lib/python}} provides <tt>grass.script</tt> in order to support GRASS scripts written in Python. The {{src|scripts}} directory of GRASS 7 contains a series of examples actually provided to the end users (while the script in GRASS 6 are shell scripts). See also [[Converting Bash scripts to Python]].


Python Scripting Library code details:
Python Scripting Library code details:
Line 8: Line 8:
* [http://grass.osgeo.org/programming6/pythonlib.html for GRASS 6]: core.py, db.py, raster.py, vector.py, setup.py, array.py task.py
* [http://grass.osgeo.org/programming6/pythonlib.html for GRASS 6]: core.py, db.py, raster.py, vector.py, setup.py, array.py task.py
* [http://grass.osgeo.org/programming7/pythonlib.html for GRASS 7]: core.py, db.py, raster.py, vector.py, setup.py, array.py task.py
* [http://grass.osgeo.org/programming7/pythonlib.html for GRASS 7]: core.py, db.py, raster.py, vector.py, setup.py, array.py task.py
== Uses for read, feed and pipe, start and exec commands ==
All of the <tt>*_command</tt> functions use <tt>make_command()</tt> to construct a command
line for a program which uses the GRASS parser. Most of them then pass
that command line to <tt>subprocess.Popen()</tt> via <tt>start_command()</tt>, except
for <tt>exec_command()</tt> which uses <tt>os.execvpe()</tt>.
[To be precise, they use <tt>grass.Popen()</tt>, which just calls
<tt>subprocess.Popen()</tt> with 'shell=True' on Windows and 'shell=False'
otherwise. On Windows, you need to use 'shell=True' to be able to
execute scripts (including batch files); 'shell=False' only works with
binary executables.]
<tt>start_command()</tt> separates the arguments into those which
<tt>subprocess.Popen()</tt> understands and the rest. The rest are passed to
<tt>make_command()</tt> to construct a command line which is passed as the
"args" parameter to <tt>subprocess.Popen()</tt>.
In other words, <tt>start_command()</tt> is a GRASS-oriented interface to
<tt>subprocess.Popen()</tt>. It should be suitable for any situation where you
would use <tt>subprocess.Popen()</tt> to execute a normal GRASS command (one
which uses the GRASS parser, which is almost all of them; the main
exception is {{cmd|r.mapcalc}} in 6.x).
Most of the others are convenience wrappers around <tt>start_command()</tt>, for common use cases.
* <tt>run_command()</tt> calls the wait() method on the process, so it doesn't return until the command has finished, and returns the command's exit code. Similar to <tt>system()</tt>.
* <tt>pipe_command()</tt> calls <tt>start_command()</tt> with 'stdout=PIPE' and returns the process object. You can use the process' .stdout member to read the command's stdout. Similar to popen(..., "r").
* <tt>feed_command()</tt> calls <tt>start_command()</tt> with stdin=PIPE and returns the process object. You can use the process' .stdin member to write to the command's stdout. Similar to popen(..., "w")
* <tt>read_command()</tt> calls <tt>pipe_command()</tt>, reads the data from the command's stdout, and returns it as a string. Similar to `backticks` in the shell.
* <tt>write_command()</tt> calls <tt>feed_command()</tt>, sends the string specified by the "stdin" argument to the command's stdin, waits for the command to finish and returns its exit code. Similar to "echo ... | command".
* <tt>parse_command()</tt> calls <tt>read_command()</tt> and parses its output as key-value pairs. Useful for obtaining information from {{cmd|g.region}}, {{cmd|g.proj}}, {{cmd|r.info}}, etc.
* <tt>exec_command()</tt> doesn't use <tt>start_command()</tt> but <tt>os.execvpe()</tt>. This causes the specified command to replace the current program (i.e. the Python script), so <tt>exec_command()</tt> never returns. Similar to bash's "exec" command. This can be useful if the script is a "wrapper" around a single command, where you construct the command line and execute the command as the final step.
If you have any other questions, you might want to look at the code ({{src|lib/python/core.py}}). Most of these functions are only a few lines long.
=== Interfacing with NumPy ===
The {{api|pythonlib.html#pythonArray|grass.script.array}} module defines a {{api|classpython_1_1array_1_1array.html|class array}} which is a subclass of [http://docs.scipy.org/doc/numpy/reference/generated/numpy.memmap.html numpy.memmap] with <code>.read()</code> and <code>.write()</code> methods to read/write the underlying file via {{cmd|r.out.bin}}/{{cmd|r.in.bin}}. Metadata can be read with {{api|namespacepython_1_1raster.html#a69e4a61eb59a31608410b733d174b8a7|raster::raster_info}}():
Example:
<source lang="python">
import grass.script as grass
import grass.script.array as garray
def main():
    map = "elevation.dem"
    # read map
    a = garray.array()
    a.read(map)
    # get raster map info
    print grass.raster_info(map)['datatype']
    i = grass.raster_info(map)
   
    # get computational region info
    c = grass.region()
    print "rows: %d" % c['rows']
    print "cols: %d" % c['cols']
    # new array for result
    b = garray.array()
    # calculate new map from input map and store as GRASS raster map
    b[...] = (a / 50).astype(int) * 50
    b.write("elev.50m")
</source>
The size of the array is taken from the current region ([[computational region]]).
The main drawback of using numpy is that you're limited by available
memory. Using a subclass of <code>numpy.memmap</code> lets you use files which may
be much larger, but processing the entire array in one go is likely to
produce in-memory results of a similar size.
=== Interfacing with NumPy and SciPy  ===
[http://docs.scipy.org/doc/scipy/reference/index.html SciPy] offers simple access to complex calculations. Example:
<source lang="python">
from scipy import stats
import grass.script as grass
import grass.script.array as garray
def main():
    map = "elevation.dem"
    x = garray.array()
    x.read(map)
    # Descriptive Statistics:
    print "max, min, mean, var:"
    print x.max(), x.min(), x.mean(), x.var()
    print "Skewness test: z-score and 2-sided p-value:"
    print stats.skewtest(stats.skew(x))
</source>
=== Interfacing with NumPy, SciPy and Matlab ===
One may also use the SciPy - Matlab interface:
   
    ### SH: in GRASS ###
    r.out.mat input=elevation output=elev.mat
<source lang="python">
    ### PY ###
    import scipy.io as sio
    # load data
    elev = sio.loadmat('elev.mat')
    # retrive the actual array. the data set contains also the spatial reference
    elev.get('map_data')
    data = elev.get('map_data')
    # a first simple plot
    import pylab
    pylab.plot(data)
    pylab.show()
    # the contour plot
    pylab.contour(data)
    # obviously data needs to ne reversed
    import numpy as np
    data_rev = data[::-1]
    pylab.contour(data_rev)
    # => this is a quick plot. basemap mapping may provide a nicer map!
    #######
</source>


== Examples ==
== Examples ==

Revision as of 18:48, 14 March 2012

See also GRASS and Python for more general info.

The code in lib/python/ provides grass.script in order to support GRASS scripts written in Python. The scripts directory of GRASS 7 contains a series of examples actually provided to the end users (while the script in GRASS 6 are shell scripts). See also Converting Bash scripts to Python.

Python Scripting Library code details:

  • for GRASS 6: core.py, db.py, raster.py, vector.py, setup.py, array.py task.py
  • for GRASS 7: core.py, db.py, raster.py, vector.py, setup.py, array.py task.py

Uses for read, feed and pipe, start and exec commands

All of the *_command functions use make_command() to construct a command line for a program which uses the GRASS parser. Most of them then pass that command line to subprocess.Popen() via start_command(), except for exec_command() which uses os.execvpe().

[To be precise, they use grass.Popen(), which just calls subprocess.Popen() with 'shell=True' on Windows and 'shell=False' otherwise. On Windows, you need to use 'shell=True' to be able to execute scripts (including batch files); 'shell=False' only works with binary executables.]

start_command() separates the arguments into those which subprocess.Popen() understands and the rest. The rest are passed to make_command() to construct a command line which is passed as the "args" parameter to subprocess.Popen().

In other words, start_command() is a GRASS-oriented interface to subprocess.Popen(). It should be suitable for any situation where you would use subprocess.Popen() to execute a normal GRASS command (one which uses the GRASS parser, which is almost all of them; the main exception is r.mapcalc in 6.x).

Most of the others are convenience wrappers around start_command(), for common use cases.

  • run_command() calls the wait() method on the process, so it doesn't return until the command has finished, and returns the command's exit code. Similar to system().
  • pipe_command() calls start_command() with 'stdout=PIPE' and returns the process object. You can use the process' .stdout member to read the command's stdout. Similar to popen(..., "r").
  • feed_command() calls start_command() with stdin=PIPE and returns the process object. You can use the process' .stdin member to write to the command's stdout. Similar to popen(..., "w")
  • read_command() calls pipe_command(), reads the data from the command's stdout, and returns it as a string. Similar to `backticks` in the shell.
  • write_command() calls feed_command(), sends the string specified by the "stdin" argument to the command's stdin, waits for the command to finish and returns its exit code. Similar to "echo ... | command".
  • parse_command() calls read_command() and parses its output as key-value pairs. Useful for obtaining information from g.region, g.proj, r.info, etc.
  • exec_command() doesn't use start_command() but os.execvpe(). This causes the specified command to replace the current program (i.e. the Python script), so exec_command() never returns. Similar to bash's "exec" command. This can be useful if the script is a "wrapper" around a single command, where you construct the command line and execute the command as the final step.

If you have any other questions, you might want to look at the code (lib/python/core.py). Most of these functions are only a few lines long.

Interfacing with NumPy

The grass.script.array module defines a class array which is a subclass of numpy.memmap with .read() and .write() methods to read/write the underlying file via r.out.bin/r.in.bin. Metadata can be read with raster::raster_info():

Example:

import grass.script as grass
import grass.script.array as garray

def main():
    map = "elevation.dem"

    # read map
    a = garray.array()
    a.read(map)

    # get raster map info
    print grass.raster_info(map)['datatype']
    i = grass.raster_info(map)
    
    # get computational region info
    c = grass.region()
    print "rows: %d" % c['rows']
    print "cols: %d" % c['cols']

    # new array for result
    b = garray.array()
    # calculate new map from input map and store as GRASS raster map
    b[...] = (a / 50).astype(int) * 50
    b.write("elev.50m")

The size of the array is taken from the current region (computational region).

The main drawback of using numpy is that you're limited by available memory. Using a subclass of numpy.memmap lets you use files which may be much larger, but processing the entire array in one go is likely to produce in-memory results of a similar size.

Interfacing with NumPy and SciPy

SciPy offers simple access to complex calculations. Example:

from scipy import stats
import grass.script as grass
import grass.script.array as garray

def main():
    map = "elevation.dem"

    x = garray.array()
    x.read(map)

    # Descriptive Statistics:
    print "max, min, mean, var:"
    print x.max(), x.min(), x.mean(), x.var()
    print "Skewness test: z-score and 2-sided p-value:"
    print stats.skewtest(stats.skew(x))

Interfacing with NumPy, SciPy and Matlab

One may also use the SciPy - Matlab interface:

   ### SH: in GRASS ###
   r.out.mat input=elevation output=elev.mat
    ### PY ###
    import scipy.io as sio
    # load data
    elev = sio.loadmat('elev.mat')
    # retrive the actual array. the data set contains also the spatial reference
    elev.get('map_data')
    data = elev.get('map_data')
    # a first simple plot
    import pylab
    pylab.plot(data)
    pylab.show()
    # the contour plot
    pylab.contour(data)
    # obviously data needs to ne reversed
    import numpy as np
    data_rev = data[::-1]
    pylab.contour(data_rev)
    # => this is a quick plot. basemap mapping may provide a nicer map!
    #######

Examples

Display example

Example of Python script, which is processed by g.parser:

#!/usr/bin/env python
#
############################################################################
#
# MODULE:      d.shadedmap
# AUTHOR(S):   Unknown; updated to GRASS 5.7 by Michael Barton
#              Converted to Python by Glynn Clements
# PURPOSE:     Uses d.his to drape a color raster over a shaded relief map
# COPYRIGHT:   (C) 2004,2008,2009 by the GRASS Development Team
#
#              This program is free software under the GNU General Public
#              License (>=v2). Read the file COPYING that comes with GRASS
#              for details.
#
#############################################################################

#%Module
#% description: Drapes a color raster over a shaded relief map using d.his
#%End
#%option
#% key: reliefmap
#% type: string
#% gisprompt: old,cell,raster
#% description: Name of shaded relief or aspect map
#% required : yes
#%end
#%option
#% key: drapemap
#% type: string
#% gisprompt: old,cell,raster
#% description: Name of raster to drape over relief map
#% required : yes
#%end
#%option
#% key: brighten
#% type: integer
#% description: Percent to brighten
#% options: -99-99
#% answer: 0
#%end

import sys
from grass.script import core as grass

def main():
    drape_map = options['drapemap']
    relief_map = options['reliefmap']
    brighten = options['brighten']
    ret = grass.run_command("d.his", h_map = drape_map,  i_map = relief_map, brighten = brighten)
    sys.exit(ret)

if __name__ == "__main__":
    options, flags = grass.parser()
    main()

Parsing the options and flags

grass.parser() is an interface to g.parser, and allows to parse the options and flags passed to your script on the command line. It is to be called at the top-level:

if __name__ == "__main__":
    options, flags = grass.parser()
    main()

Global variables "options" and "flags" are Python dictionaries containing the options/flags values, keyed by lower-case option/flag names. The values in "options" are strings, those in "flags" are Python booleans. All those variables have to be previously declared in the header of your script.

>>> options, flags = grass.parser()
>>> options
{'input': 'my_map', 'output': 'map_out', 'option1': '21.472', 'option2': ''}
>>> flags
{'c': True, 'm': False}

Example for embedding r.mapcalc (map algebra)

grass.mapcalc() accepts a template string followed by keyword arguments for the substitutions, e.g. (code snippets):

grass.mapcalc("${out} = ${rast1} + ${rast2}",
              out = options['output'],
              rast1 = options['raster1'],
              rast2 = options['raster2'])

Best practice: first copy all of the options[] into separate variables at the beginning of main(), i.e.:

def main():
    output = options['output']
    raster1 = options['raster1']
    raster2 = options['raster2']
 
    ...
 
    grass.mapcalc("${out} = ${rast1} + ${rast2}",
                  out = output,
                  rast1 = raster1,
                  rast2 = raster2)

Example for parsing raster category labels

How to obtain the text labels

    # dump cats to file to avoid "too many argument" problem:
    p = grass.pipe_command('r.category', map = rastertmp, fs = ';', quiet = True)
    cats = []
    for line in p.stdout:
        cats.append(line.rstrip('\r\n').split(';')[0])
    p.wait()

    number = len(cats)
    if number < 1:
        grass.fatal(_("No categories found in raster map"))

Example for parsing category numbers

Q: How to obtain the number of cells of a certain category?

A: It is recommended to use pipe_command() and parse the output, e.g.:

       p = grass.pipe_command('r.stats',flags='c',input='map')
       result = {}
       for line in p.stdout:
           val,count = line.strip().split()
           result[int(val)] = int(count)
       p.wait()

Example for getting the region's number of rows and columns

Q: How to obtain the number of rows and columns of the current region?

A: It is recommended to use the "grass.region()" function which will create a dictionary with values for extents and resolution, e.g.:

#!/usr/bin/env python
#-*- coding:utf-8 -*-
#
############################################################################
#
# MODULE:       g.region.resolution
# AUTHOR(S):    based on a post at GRASS-USER mailing list [1]               
# PURPOSE:	Parses "g.region -g", prints out number of rows, cols
# COPYLEFT:     ;-)
# COMMENT:      ...a lot of comments to be easy-to-read for/by beginners
#
#############################################################################
#
#%Module
#% description: Print number of rows, cols of current geographic region
#% keywords: region
#%end

# importing required modules
import sys # the sys module [2]
from grass.script import core as grass # the core module [3]

# information about imported modules can be obtained using the dir() function
# e.g.: dir(sys)

# define the "main" function: get number of rows, cols of region
def main():
    
    # #######################################################################
    # the following commented code works but is kept only for learning purposes
     
    ## assigning the output of the command "g.region -g" in a string called "return_rows_x_cols"
    # return_rows_x_cols = grass.read_command('g.region', flags = 'g')
    
    ## parsing arguments of interest (rows, cols) in a dictionary named "rows_x_cols"
    # rows_x_cols = grass.parse_key_val(return_rows_x_cols)
    
    ## selectively print rows, cols from the dictionary "rows_x_cols"
    # print 'rows=%d \ncols=%d' % (int(rows_x_cols['rows']), int(rows_x_cols['cols']))
    
    # #######################################################################
    
    # faster/ easier way: use of the "grass.region()" function
    gregion = grass.region()
    rows = gregion['rows']
    cols = gregion['cols']
    
    # print rows, cols properly formated 
    print 'rows=%d \ncols=%d' % (rows, cols)

# this "if" condition instructs execution of code contained in this script, *only* if the script is being executed directly 
if __name__ == "__main__": # this allows the script to be used as a module in other scripts or as a standalone script
    options, flags = grass.parser() #
    sys.exit(main()) #

# Links
# [1] http://n2.nabble.com/Getting-rows-cols-of-a-region-in-a-script-tp2787474p2787509.html
# [2] http://www.python.org/doc/2.5.2/lib/module-sys.html
# [3] http://download.osgeo.org/grass/grass6_progman/pythonlib.html#pythonCore

Managing mapsets

To check if a certain mapset exists in the active location, use:

       grass.script.mapsets(False)

... returns a list of mapsets in the current location.

r.mapcalc example

Example of Python script, which is processed by g.parser:

The shell script line:

  r.mapcalc "MASK = if(($cloudResampName < 0.01000),1,null())"

would be written like this:

       import grass.script as grass

       ...

       grass.mapcalc("MASK=if(($cloudResampName < 0.01000),1,null())",
                     cloudResampName = cloudResampName)

The first argument to the mapcalc function is a template (see the Python library documentation for string.Template). Any keyword arguments (other than quiet, verbose or overwrite) specify substitutions.

Using output from GRASS modules in the script

Sometimes you need to use the output of a module for the next step. There are dedicated functions to obtain the result of, for example, a statistical analysis.

Example: get the range of a raster map and use it in r.mapcalc. Here you can use grass.script.raster_info(), e.g.:

       import grass.script as grass

       max = grass.raster_info(inmap)['max']
       grass.mapcalc("$outmap = $inmap / $max",
                     inmap = inmap, outmap = outmap, max = max)

Calling a GRASS module in Python

Imagine, you wanted to execute this command in Python:

  r.profile -g input=mymap output=newfile profile=12244.256,-295112.597,12128.012,-295293.77

All arguments except the first (which is a flag) are keyword arguments, i.e. arg = val. For the flag, use flags = 'g' (note that "-g" would be the negative of a Python variable named "g"!). So:

       grass.run_command(
               'r.profile',
               input = input_map,
               output = output_file,
               profile = [12244.256,-295112.597,12128.012,-295293.77]

or:

               profile = [(12244.256,-295112.597),(12128.012,-295293.77)]

i.e. you need to provide the keyword, and the argument must be a valid Python expression. Function run_command() etc accept lists and tuples.

Differences between run_command() and read_command():

  • run_command() executes the command and waits for it to terminate; it doesn't redirect any of the standard streams.
  • read_command() executes the command with stdout redirected to a pipe, and reads everything written to it. Once the command terminates, it returns the data written to stdout as a string.

How to retrieve error messages from read_command():

None of the existing *_command functions redirect stderr. You can do so with e.g.:

def read2_command(*args, **kwargs):
   kwargs['stdout'] = grass.PIPE
   kwargs['stderr'] = grass.PIPE
   ps = grass.start_command(*args, **kwargs)
   return ps.communicate()

This behaves like read_command() except that it returns a tuple of (stdout,stderr) rather than just stdout.

Percentage output for progress of computation

A) Within a Python script, the grass.script.core.percent() module method wraps the g.message -p command.

B) If you call a GRASS command within the Python code, you have to parse the output by setting GRASS_MESSAGE_FORMAT=gui in the environment when running the command and read from the command's stderr; e.g.

       import grass.script as grass
       env = os.environ.copy()
       env['GRASS_MESSAGE_FORMAT'] = 'gui'
       p = grass.start_command(..., stderr = grass.PIPE, env = env)
       # read from p.stderr
       p.wait()

If you need to capture both stdout and stderr, you need to use threads, select, or non-blocking I/O to consume data from both streams as it is generated in order to avoid deadlock.

ALTERNATIVE:

Redirect both stdout and stderr to the same pipe (and hope that the normal output doesn't include anything which will be mistaken for progress/error/etc messages):

       p = grass.start_command(..., stdout = grass.PIPE, stderr = grass.STDOUT, env = env)

NULL data management

How to analyse if there are only NULL cells in a map:

If a map contains only null cells, its minimum and maximum will be "NULL":

       $ r.mapcalc 'foo = null()'
       $ r.info -r foo
       min=NULL
       max=NULL

Using the Python API, The 'min' and 'max' values in the result of the raster_info() function will be None.

Counting cells

Counting cells is far more expensive than simply determining whether there are any non-null cells. Counting cells requires reading the entire map, while the r.info approach only needs to read the metadata files.

If you do need to count cells, r.stats is likely to be more efficient than {{cmd|r.univar}.

A count loop:

       while grass.raster_info(inmap)['max'] is not None:
           ...

Path to GISDBASE

In order to a avoid hardcoded paths to GRASS mapset files like the SQLite DB file, you can get the GISDBASE variable from the environment:

       import grass.script as grass
       import os.path

       env = grass.gisenv()

       gisdbase = env['GISDBASE']
       location = env['LOCATION_NAME']
       mapset = env['MAPSET']

       path = os.path.join(gisdbase, location, mapset, 'sqlite.db')

Use Python reserved keyword

Question: r.resamp.bspline uses 'lambda' as a command line parameter name, but when you try to use it with grass.run_command() you get an error as lambda is a python reserved keyword. How to work around that?

Answer: Prepend an underscore to the name, i.e.:

       grass.run_command('r.resamp.bspline', _lambda = ...)

Controlling the PNG display driver

Code fragment to control the pngdriver in Python:

import os
import sys
from grass.script import core as grass
def main():
       os.environ['GRASS_PNGFILE'] = filename
       os.environ['GRASS_WIDTH'] = str(width)
       os.environ['GRASS_HEIGHT'] = str(height)
       grass.run_command('d.his', i='elevation_shade', h='elevation')

Sophisticated cleanup procedure

Scripts which create several temporary files need a more sophisticated cleanup procedure which deletes all the tmp maps which have been created. This procedure should also work if the script stops (e.g due to an error).

Solution: Define a list of map names which starts out empty and has names appended to it as the names are generated. Code fragment:

       tmp_rast = []

       def cleanup():
           for rast in tmp_rast:
               grass.run_command("g.remove",
                                 rast = rast,
                                 quiet = True)

       def main():
           ...
           while ...:
               next_rast = ...
               tmp_rast.append(next_rast)
               ...

       if __name__ == "__main__":
           options, flags = grass.parser()
           atexit.register(cleanup)
           sys.exit(main())