Skip to content

ahderman/planet-test

Repository files navigation

Calibration Vector Application

This repository contains code that provides a solution to a recruitment test provided by Planet.
The test description can be found here.

Comments

Difficulties

While completing this assignment, the biggest difficulty was finding documentation for the python gdal module, for example regarding how to pass the COMPRESS=LZW option, or about whether there was an equivalent to GDALClose() that needed to be called after writing to the image file.
In the end, StackOverflow was my most useful resource, as is so often the case.

In addition, I do find it surprising that in the formula Γ² = γ * α², the calibrated matrix is squared while the uncalibrated one isn't. I'd like to understand this better.

Documentation

For docstrings, I have copied the format used by pandas, but I'm not sure it's very standard and I haven't tried parsing it with a tool to ensure the output is what I expect. I might try reStructuredText next time.

Conda create package

I wanted to create a conda module using conda-build, because I could then specify conda packages gdal and numpy as dependencies of my module. However, I ran into the error described in (https://github.com/conda/conda-build/issues/1502)[https://github.com/conda/conda-build/issues/1502], due to the fact that I have an encrypted HOME. I did not take the time to fix this by installing miniconda on a non-encrypted part of my drive, opting instead to provide a setup.py file, and asking for it to be run in a conda environment that already contains the dependencies.

Assumptions

I have made a few assumptions regarding the input data:

  • Calibration spacing is an integer.
  • The image matrix contains only positive numbers.
  • I have understood the maths correctly (see above, in the Difficulties section).

Future work

I see several ways of improving this code:

  • Provide instructions to package the code using conda
  • Add exception handling in case bad input is provided
  • Allow image transformations, for example using larger pixels if the image is too small
  • In general, hard-coding less in the script to generate the image, to make it more flexible
  • Provide a Dockerfile, for people who prefer having docker as their single dependency rather than conda.

Prerequisites

Conda environment

Running the code requires numpy and gdal, and running the tests also requires nose. An environment.yml file is provided for installing the dependencies with conda.
Create the planet-test conda environment from the provided environment.yml file:

conda env create -f environment.yml

Install module

In the created conda environment, install the image_matrix_calibration module:

source activate planet-test
python setup.py install

Command-line tools

Image matrix calibration

A command-line wrapper around the image_matrix_calibration module is provided. It can be used as follows to generate a calibrated matrix and store it in a json file; example files are provided in the example_data folder:

python calibrate_image_matrix.py example_data/calibration_params.json example_data/image_matrix.json > calibrated_image_matrix.json

TIFF Image generation

To visualize the output matrix, the generate_image_from_matrix command-line tool is provided. It first scales all the pixel values to fill the range [0; 255], then uses the gdal module to create an image with 3 bands, compressed with LZW and whose values are unsigned ints. The transformations done to obtain the various bands are arbitrary. The tool can be used as follows:

python generate_image_from_matrix.py calibrated_image_matrix.json image.tif

This is the (zoomed in) resulting image:

Tests

The tests can be run with nose:

nosetests

or:

python -m nose

Linting

To ensure that code is PEP8-compliant, the provided conda environment includes Flake8, which can be run with the following command:

flake8

or

python -m flake8

About

My solution to the problem sent to me by Planet as part of my application.

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Languages