METplus Practical Session Guide (Version 5.0) | MET Tool: Plot-Data-Plane > Python Embedding

Python Embedding

While the MET tools can read data from a few input gridded data file types, its ability to read data in memory from python greatly enhances its utility. Support for python embedding is optional, and must be enabled at compilation time as described in Appendix F of the MET User's Guide. MET supports three types of python embedding:

  1. Reading a field of gridded data values.
  2. Passing a list of point observations.
  3. Passing a list of matched forecast/observation pair values.

On this page, we'll demonstrate only the first type, reading a field of gridded data values. When getting started with a new dataset via python, validating the logic by running Plot-Data-Plane is crucial. When MET reads data from flat files, important information, like the location and orientation of the data, is defined in the metadata. That is not the case for python embedding, and the responsibility for confirming those details falls to the user. So while python embedding provide extra flexibility, it also requires additional diligence.

Let's run the simplest of examples using sample data included with the MET release. The first step is to confirm that python script runs by itself, outside of MET.

python3 ${METPLUS_DATA}/met_test/scripts/python/ \
${METPLUS_DATA}/met_test/data/python/fcst.txt Forecast

This sample script reads data from the input fcst.txt ASCII file and gives it a name, Forecast. Always run new python scripts on the command line first to confirm there aren't any syntax errors in the script itself. The required conventions for the python script are details in the Python Embedding for 2D Data section of the MET User's Guide.

Next, let's run Plot-Data-Plane using this python script to define the input data. As described in The Field String, this is done with the name configuration string and the level string does not apply.

plot_data_plane \
${METPLUS_TUTORIAL_DIR}/output/met_output/plot_data_plane/ \
'name = "${METPLUS_DATA}/met_test/scripts/python/ ${METPLUS_DATA}/met_test/data/python/fcst.txt Forecast";'

Since there is no input_filename to be specified as the first required argument for Plot-Data-Plane, we provide the constant string PYTHON_NUMPY in that spot. This triggers Plot-Data-Plane to interpret the field string as a python embedding script to be run. Specifying PYTHON_XARRAY also works but requires slightly different conventions in the python embedding script.

When MET is compiled, it links to python libraries that it uses to instantiate a python interpreter at runtime. That compile time instance does have a few required packages, but will likely not include all packages that every user may want to load. You may find that your python script runs fine on the command line, but Plot-Data-Plane's call to python can't load a requested module. In that case, set the ${MET_PYTHON_EXE} environment variable to tell MET which instance of python you'd like to run.

${MET_PYTHON_EXE} defines a specific instance of python to be run.

The following command just uses that version of python that is already present in your path:

export MET_PYTHON_EXE=`which python3`

Rerunning the command from above should produce the same result, but if you look closely at the log messages, you'll see that your custom python version writes a temporary file, and MET's compile time python version reads data from it.

plot_data_plane \
${METPLUS_TUTORIAL_DIR}/output/met_output/plot_data_plane/ \
'name = "${METPLUS_DATA}/met_test/scripts/python/ ${METPLUS_DATA}/met_test/data/python/fcst.txt Forecast";'

You can find several python embedding examples on the Sample Analysis Scripts page of the MET website. Each example includes both a python script and sample input data file. Please also see METplus Python Embedding use case examples.