rio-mucho 0.0.1

Windowed multiprocessing wrapper for rasterio

Parallel processing wrapper for rasterio

Install

From pypi:

pip install rio-mucho --pre

From github (usually for a branch / dev):

pip install pip install git+ssh://[e-mail]/mapbox/rio-mucho.git@<branch>

Development:

git clone [e-mail]:mapbox/rio-mucho.git
cd rio-mucho
pip install -e .

Usage

with riomucho.RioMucho([{inputs}], {output}, {run function},
    windows={windows},
    global_args={global arguments},
    kwargs={kwargs to write}) as rios:

    rios.run({processes})

Arguments

inputs

An list of file paths to open and read.

output

What file to write to.

run_function

A function to be applied to each window chunk. This should have input arguments of:

1. A data input, which can be one of:

• A list of numpy arrays of shape (x,y,z), one for each file as specified in input file list mode="simple_read" [default]
• A numpy array of shape ({n input files x n band count}, {window rows}, {window cols}) mode=array_read"
• A list of open sources for reading mode="manual_read"

1. A rasterio window tuple
2. A rasterio window index (ij)
3. A global arguments object that you can use to pass in global arguments

This should return:

1. An output array of ({count}, {window rows}, {window cols}) shape, and of the correct data type for writing

def basic_run({data}, {window}, {ij}, {global args}):
    ## do something
    return {out}

Keyword arguments

windows={windows}

A list of rasterio (window, ij) tuples to operate on. [Default = src[0].block_windows()]

global_args={global arguments}

Since this is working in parallel, any other objects / values that you want to be accessible in the run_function. [Default = {}]

global_args = {
    'divide_value': 2
}

kwargs={keyword args}

The kwargs to pass to the output. [Default = srcs[0].kwargs

Example

import riomucho, rasterio, numpy

def basic_run(data, window, ij, g_args):
    ## do something
    out = np.array(
        [d[0] /= global_args['divide'] for d in data]
        )
    return out

# get windows from an input
with rasterio.open('/tmp/test_1.tif') as src:
    ## grabbing the windows as an example. Default behavior is identical.
    windows = [[window, ij] for ij, window in src.block_windows()]
    kwargs = src.meta
    # since we are only writing to 2 bands
    kwargs.update(count=2)

global_args = {
    'divide': 2
}

processes = 4

# run it
with riomucho.RioMucho(['input1.tif','input2.tif'], 'output.tif', basic_run,
    windows=windows,
    global_args=global_args,
    kwargs=kwargs) as rm:

    rm.run(processes)

Utility functions

`riomucho.utils.array_stack([array, array, array,…])

Given a list of ({depth}, {rows}, {cols}) numpy arrays, stack into a single (l{list length * each image depth}, {rows}, {cols}) array. This is useful for handling variation between rgb inputs of a single file, or separate files for each.

One RGB file

files = ['rgb.tif']
open_files = [rasterio.open(f) for f in files]
rgb = `riomucho.utils.array_stack([src.read() for src in open_files])

Separate RGB files

files = ['r.tif', 'g.tif', 'b.tif']
open_files = [rasterio.open(f) for f in files]
rgb = `riomucho.utils.array_stack([src.read() for src in open_files])