######################################################## Wednesday A: Adding QC to JEDI ######################################################## Introduction ---------------- This session is on the JEDI Unified Forward Operator (UFO) code and quality control (QC) filters. This tutorial has two main parts. First, you will review how to clone, build and customize a bundle. Second, you will add in a new QC filter to JEDI. Filters are an essential component in a data assimilation workflow. Filters can change quality control flags (i.e., to reject or retain observations) and observation error variances (e.g., one might wish to increase observation error variances to decrease the observation weight in the analysis instead of rejecting observations altogether). In JEDI, filters are customizable and generic. This means that you can use the same code (written in C++) to accomplish different tasks (specified by you in a YAML file). This tutorial introduces the YAML file format, explains how to specify a QC filter in YAML, and explains the backend code used to actually perform the filtering. We will implement a relatively simple filter in C++. This activity assumes that you have completed the previous activities and still have access to a JupyterLab or SSH session. Step 1: Access your AWS instance and enter the Singularity container ---------------------------------------------------------------------- Connect to your assigned compute node. You will use the same method as yesterday. You already have the singularity container that contains the JEDI dependencies. Enter the container using: .. code:: bash cd ~/ singularity shell -e jedi-gnu-openmpi-dev_latest.sif Once in the container be sure also to remove limits the stack memory to prevent spurious failures as noted :doc:`yesterday `: .. code-block:: bash ulimit -s unlimited ulimit -v unlimited Step 2: Customize the fv3-bundle ------------------------------------------ JEDI bundles are convenient ways for users to build all of the JEDI components that are needed for a particular application. In previous sessions, you cloned and built the ``fv3-bundle``. Now, we will further explore how bundles are structured and will make modifications to the Unified Forward Operator (UFO) code. Ordinarily, whenever you modify the JEDI source code, it would be a good idea to create a fresh clone of JEDI. This way, you could ensure that you do not make accidental changes to unrelated files. However, cloning and building a new copy of ``fv3-bundle`` takes a substantial amount of time, particularly on AWS nodes with fewer computational resources than an HPC environment. So, we will modify the existing ``fv3-bundle`` instead. Open a new terminal and navigate to the ``~/jedi/fv3-bundle`` directory. Change into this directory and look around. When you first cloned this bundle on Monday, it was very small and contained only a ``CMakeLists.txt`` file and a few supplemental configuration and documentation files. When ecbuild was executed on the bundle, it populated the bundle directory with many subdirectories. These each contain cloned codes for the JEDI repositories. Any file named ``CMakeLists.txt`` represents a set of instructions for the build system (i.e. CMake and ecbuild). The ``CMakeLists.txt`` file at the bundle root directory describes the components that are incorporated into this particular JEDI software bundle. The "Getting Started" activity provided a brief introduction to this file. If you modify JEDI, you will undoubtedly modify this file extensively, so some review is helpful. Open the ``CMakeLists.txt`` file in a text editor and examine it. It can be divided into three parts. - The first part (lines 1-35) is a preamble that you normally never have to change. The code here tells ecbuild that we are declaring a bundle, that we require a certain minimum version of the ``cmake`` program, and that we require C, C++ and Fortran compilers. - The second part (lines 36-69) has several ``ecbuild_bundle`` declarations that tell ecbuild about which JEDI components we want and where to find them. By default, the bundle depends on the ``develop`` branch of several repositories. You can easily change the branches that your code targets. This is quite helpful when you are adding a new feature or are testing existing code across multiple systems. - The final part (lines 70-73) is again mostly fixed. It calls a macro function in ecbuild to execute the bundle's instructions. Step 3: Make a new UFO feature branch --------------------------------------------------------------------------- You invoked ecbuild on Monday's Getting Started activity. Ecbuild cloned the stable branches of several repositories. However, in this tutorial we want to make modifications to the UFO code. In JEDI, we aim to follow the "git flow" paradigm when developing, and we will discuss this in depth in a later lecture on Friday. In short summary, the ``develop`` branch contains the development version of each repository. This version of the code should always build and test successfully. Whenever you want to add a new feature to the code, you should do your work in another branch of the repository. Once the work is done, you can issue a "Pull Request" to have other JEDI users review your code and merge in your changes into the ``develop`` branch. Every month or so, we aim to release a stable, consistent snapshot of the JEDI repositories. We copy the development branch to a git "tag" (an immutable branch). The meanings of "master", "develop", and "tags" will be discussed in the git-flow lectures and in later practical exercises. Because of the ordering of the lectures, and because we want stable, reproducible academy exercises, we have a special copy of the ufo repository in the jcsda-academy on GitHub. In this repository, both the 1.0.0 tag and develop branch are identical. This is **not** the case in actual development. Ordinarily, the top-level ``CMakeLists.txt`` file would not reference ``TAG 1.0.0`` when describing each package and instead would reference ``BRANCH develop``. Open the top-level CMakeLists.txt file (``~/jedi/fv3-bundle/CMakeLists.txt``). Change line 59 from: .. code:: bash ecbuild_bundle( PROJECT ufo GIT "https://github.com/jcsda-academy/ufo.git" TAG 1.0.0 ) to: .. code:: bash ecbuild_bundle( PROJECT ufo GIT "https://github.com/jcsda-academy/ufo.git" BRANCH feature/new_qc_filter_example ) Then, enter the source code's ``ufo`` subdirectory (``cd ~/jedi/fv3-bundle/ufo``; NOTE: There is also a ``ufo`` directory in your current directory at ``~/jedi/build/ufo`` <-- This is not the directory that you want.) Checkout the ``develop`` branch and then create a new branch as follows: .. code:: bash git checkout develop git checkout -b feature/new_qc_filter_example The ``-b`` option to ``git checkout`` creates the branch by effectively making a copy of the develop branch. Now, **very important**, we need to tell JEDI what test data to use. By default, when you create a new feature branch, it will use the same test data as the ``develop`` branch on the main JCSDA repository, ``https://github.com/JCSDA/ufo``. This makes sense because, following the git-flow paradigm as we will discuss on Friday, all feature branches should be created from the ``develop`` branch. But, the ``develop`` branch we just checked out is the develop branch from the fork of ufo that we created for the academy, namely ``https://github.com/jcsda-academy/ufo``. As noted above, this develop branch matches the release code in master. So, the develop branch of the academy fork is more than a month behind the actual JCSDA develop branch. This was done to create a stable environment for the Academy. So, the bottom line is that we need to tell JEDI to use the local copy of the test data that we created on :doc:`Monday (Step 9) `. And, we do so by linking the test data for our feature branch to the test data that's already there for ufo and ioda: .. code-block:: bash export LOCAL_PATH_JEDI_TESTFILES=$HOME/jedi/test_data cd $LOCAL_PATH_JEDI_TESTFILES/ioda ln -s 1.0.0 develop cd $LOCAL_PATH_JEDI_TESTFILES/ufo ln -s 1.0.0 develop cd $LOCAL_PATH_JEDI_TESTFILES/crtm mkdir -p develop cd develop tar xvf ../crtm_coefficients.tar.gz cd $LOCAL_PATH_JEDI_TESTFILES/fv3-jedi mkdir -p develop cd develop tar xvf ../fv3jedi_testdatainput_tier_1.tar.gz Without this, JEDI would download the newer test data from `AWS S3 `_, which would be out of synch with our code, possibly leading to some test failures. Finally, return to the build directory (``~/jedi/build``) and re-run ecbuild: .. code:: bash cd ~/jedi/build ecbuild ../fv3-bundle We want to re-run ecbuild because we updated ``CMakeLists.txt``. Ecbuild checks this file and ensures that we have checked out the correct repositories. Once ecbuild completes, verify that it reports that configuration has succeeded. Step 4: Review of YAML files ---------------------------- Programmers and computers typically store data as complex "objects” (`structures and classes`_). In a computer's memory, these objects may have very complicated storage involving pointers, references, dictionaries, and similar constructs. However, when we need to store these complex structures to a disk or send them across a network, we have to translate these complex structures into a series of bytes (a.k.a. we `serialize`_ an object into `a byte stream`_). There are lots of ways of doing this. However, JEDI wanted to employ a consistent, well-documented format that is easy for people to edit and for machines to read. So, we chose to use the YAML Ain't Markup Language (YAML) format to store the configuration data for the JEDI project. `YAML`_ was developed in 2001 and has been implemented for use with `several`_ programming languages. Let's take a look at a YAML file for a brief overview. .. code:: yaml --- # Comments are indicated with the '#' symbol. name: "Your name here" # A string a-boolean-value: true an-integer-value: 3 pi: 3.14159 list-of-some-jedi-components: - saber - oops - ioda - ufo dictionary-of-places-to-explore-in-a-staycation: - local-park: scenic: true features: - "Running trails" - Trees - "Duck pond" - aquarium: types-of-animals: - jellyfish - turtles - fish free: false mask: true # TODO: Explore this area and add more details. The file starts with three dashes. These dashes indicate the start of a new YAML document. YAML supports multiple documents, and compliant parsers will recognize each set of dashes as the beginning of a new one. Comments are started with a space and a hashtag (" #") and extend to the end of the line. Next, we see the construct that makes up most of a typical YAML document: a key-value pair. "name” is a key that points to a string value: "Your name here”. YAML allows for several types of values: strings, integers, floating-point numbers, boolean values and dates are all acceptable. Strings can optionally be enclosed in quotes. Quotes include both single and double quotes. You can also add in arrays / lists. Each element in a list is denoted by an opening dash. YAML elements can also be nested. This lets you emulate a group / folder structure. Nesting is accomplished by adding levels of spaces (no tabs allowed). See `this link`_ for more examples. Step 5: How do we invoke filters using a YAML configuration file? ----------------------------------------------------------------- Example YAML code for filters can be found in the UFO repository in ufo/test/testinput. The ``qc*.yaml`` files provide many examples of how to use QC filters. Let's look at the DifferenceCheck filter to see how a relatively basic filter works. See qc_differencecheck.yaml: .. code:: yaml window begin: 2018-01-01T00:00:00Z window end: 2019-01-01T00:00:00Z observations: - obs space: name: test data obsdatain: obsfile: Data/ufo/testinput_tier_1/filters_testdata.nc4 simulated variables: [variable1] obs filters: - filter: Difference Check # test minvalue with one var (compare var3-var4 with min) value: var3@MetaData # var3@MetaData = 1, 1, 1, 1, 1, 0, 0, 0, 0, 0 reference: var4@MetaData # var4@MetaData = 0, 0, 0, 0, 0, 1, 2, 3, 4, 5 minvalue: 0.0 passedBenchmark: 5 - obs space: name: test data obsdatain: obsfile: Data/ufo/testinput_tier_1/filters_testdata.nc4 simulated variables: [variable1, variable2, variable3] obs filters: - filter: Difference Check # test same minvalue with three vars (compare var3-var4 with min) value: var3@MetaData # var3@MetaData = 1, 1, 1, 1, 1, 0, 0, 0, 0, 0 reference: var4@MetaData # var4@MetaData = 0, 0, 0, 0, 0, 1, 2, 3, 4, 5 minvalue: 1.0 passedBenchmark: 15 - obs space: name: test data obsdatain: obsfile: Data/ufo/testinput_tier_1/filters_testdata.nc4 simulated variables: [variable1] obs filters: - filter: Difference Check # test maxvalue (compare var3-var4 with max) filter variables: - name: variable1 value: var3@MetaData # var3@MetaData = 1, 1, 1, 1, 1, 0, 0, 0, 0, 0 reference: var4@MetaData # var4@MetaData = 0, 0, 0, 0, 0, 1, 2, 3, 4, 5 maxvalue: -3.0 passedBenchmark: 3 - obs space: name: test data obsdatain: obsfile: Data/ufo/testinput_tier_1/filters_testdata.nc4 simulated variables: [variable1] obs filters: - filter: Difference Check # test min and maxvalue (compare var3-var4 with min and max) filter variables: - name: variable1 value: variable2@ObsValue # variable2@ObsValue = 10, 12, 14, 16, 18, 20, 22, 24, 26, 28 reference: variable1@ObsValue # variable1@ObsValue = 10, 11, 12, 13, 14, 15, 16, 17, 18, 19 minvalue: 2.0 maxvalue: 6.0 passedBenchmark: 5 - obs space: name: test data obsdatain: obsfile: Data/ufo/testinput_tier_1/filters_testdata.nc4 simulated variables: [variable1] obs filters: - filter: Difference Check # test threshold (compare abs(variable2 - variable1) with threshold) filter variables: - name: variable1 value: variable2@ObsValue # variable2@ObsValue = 10, 12, 14, 16, 18, 20, 22, 24, 26, 28 reference: variable1@ObsValue # variable1@ObsValue = 10, 11, 12, 13, 14, 15, 16, 17, 18, 19 threshold: 3 passedBenchmark: 4 - obs space: name: test data obsdatain: obsfile: Data/ufo/testinput_tier_1/filters_testdata.nc4 simulated variables: [variable1] obs filters: - filter: Difference Check # test min and maxvalue (equal), equivalent to previous test filter variables: - name: variable1 value: variable2@ObsValue # variable2@ObsValue = 10, 12, 14, 16, 18, 20, 22, 24, 26, 28 reference: variable1@ObsValue # variable1@ObsValue = 10, 11, 12, 13, 14, 15, 16, 17, 18, 19 minvalue: -3 maxvalue: 3 passedBenchmark: 4 UFO accesses observation data via functions and subroutines in the ObsSpace (Observation Space) class. The above YAML files specifies several instances of ObsSpace. Each space has a name, a path to the input data and a list of variables to be simulated. Paired with the ObsSpaces are the filters (ObsFilters) that act on each space. When specifying a filter, you must provide its name and any other configuration information that it requires. The above YAML file invokes the Difference Check filter. Its options are described in `[the ReadTheDocs Documentation Site] `__. For reference, a segment of the documentation is reproduced here: +-----------+---------------------------------------------------------+ | Parameter | Description | +===========+=========================================================+ | value | The variable that we are comparing | +-----------+---------------------------------------------------------+ | reference | The variable that we are comparing against | +-----------+---------------------------------------------------------+ | minvalue | The minimum difference of (value - reference) for a | | | valid datum | +-----------+---------------------------------------------------------+ | maxvalue | The maximum difference of (value - reference) for a | | | valid datum | +-----------+---------------------------------------------------------+ | threshold | A shortcut for expressing minvalue = -threshold, | | | maxvalue = threshold | +-----------+---------------------------------------------------------+ Since we are using this YAML file in a test, we also encode the expected number number of passed locations using the ``passedBenchmark`` option. Step 6: How do we implement a filter? ------------------------------------- The C++ code for all filters can be found in the UFO repository in `src/ufo/filters `_. The header file for the DifferenceCheck filter is available `here `_ and the `full source code is here `_. Defining the filter - the header file ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ Annotated excerpt from DifferenceCheck.h: .. code:: cpp namespace ufo { /// DifferenceCheck filter class DifferenceCheck : public FilterBase, private util::ObjectCounter { public: static const std::string classname() {return "ufo::DifferenceCheck";} /// !!! This Constructor function initializes an instance of the /// !!! filter based on options specified in the YAML configuration file. DifferenceCheck(ioda::ObsSpace &, const eckit::Configuration &, std::shared_ptr >, std::shared_ptr >); ~DifferenceCheck(); private: void print(std::ostream &) const override; /// !!! This is the function that does all of the work in the filter. !!! void applyFilter(const std::vector &, const Variables &, std::vector> &) const override; int qcFlag() const override {return QCflags::diffref;} const Variable ref_; const Variable val_; }; } // namespace ufo Implementing the filter - the source code file ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ Excerpt from DifferenceCheck.cc: .. code:: cpp // ----------------------------------------------------------------------------- /** !!! This is the constructor function. * * When we instantiate a new DifferenceFilter object, we read in configuration * from the YAML files (stored in the _config_ variable). * * We look for two keys: * - refererence: the name of the variable used for the reference. * - value: the name of the variable that contains our data. **/ DifferenceCheck::DifferenceCheck(ioda::ObsSpace & obsdb, const eckit::Configuration & config, std::shared_ptr > flags, std::shared_ptr > obserr) : FilterBase(obsdb, config, flags, obserr), ref_(config_.getString("reference")), val_(config_.getString("value")) { oops::Log::trace() << "DifferenceCheck contructor starting" << std::endl; /// Here we tell OOPS and IODA that our filter requires these variables to work. /// I.e. these variables have to be available in memory. allvars_ += ref_; allvars_ += val_; } // ----------------------------------------------------------------------------- /** !!! This function does the actual work. * * We read in three keys from the YAML configuration: minvalue, maxvalue, and threshold. * * When applying this filter, we loop over all possible locations. * For each location, we check the difference between the two variables (reference and value). * If the difference is outside of the bounds specified by minvalue, maxvalue and threshold, * then we flag that location. This flag gets passed back to the calling function, which then * sets the appropriate QC flag. **/ void DifferenceCheck::applyFilter(const std::vector & apply, const Variables & filtervars, std::vector> & flagged) const { oops::Log::trace() << "DifferenceCheck priorFilter" << std::endl; const float missing = util::missingValue(missing); const size_t nlocs = obsdb_.nlocs(); // min/max value setup float vmin = config_.getFloat("minvalue", missing); float vmax = config_.getFloat("maxvalue", missing); // check for threshold and if exists, set vmin and vmax appropriately const float thresh = config_.getFloat("threshold", missing); if (thresh != missing) { vmin = -thresh; vmax = thresh; } // Get reference values and values to compare (as floats) std::vector ref, val; data_.get(ref_, ref); data_.get(val_, val); ASSERT(ref.size() == val.size()); // Loop over all obs for (size_t jobs = 0; jobs < nlocs; ++jobs) { if (apply[jobs]) { // check to see if one of the reference or value is missing if (val[jobs] == missing || ref[jobs] == missing) { for (size_t jv = 0; jv < filtervars.nvars(); ++jv) { flagged[jv][jobs] = true; } } else { // Check if difference is within min/max value range and set flag float diff = val[jobs] - ref[jobs]; for (size_t jv = 0; jv < filtervars.nvars(); ++jv) { if (vmin != missing && diff < vmin) flagged[jv][jobs] = true; if (vmax != missing && diff > vmax) flagged[jv][jobs] = true; } } } } } } // namespace ufo Step 7: Try to add a new filter ------------------------------------ We are going to re-implement a **simplified** version of the Bounds Check filter. This filter checks that observation data are within certain user-specified bounds. Step 7a: The backend logic ~~~~~~~~~~~~~~~~~~~~~~~~~~ Navigate into the ``~/jedi/fv3-bundle/ufo/src/ufo/filters`` directory. Copy the ``DifferenceCheck.cc`` and ``DifferenceCheck.h`` files to ``PracticalBoundsCheck.cc`` and ``PracticalBoundsCheck.h``, respectively. Open these files in your editor of choice. In ``PracticalBoundsCheck.h``: - Rename all references of ``DifferenceCheck`` to ``PracticalBoundsCheck``. Search for all possible capitalizations. Don't forget the capitalized text on lines 8, 9 and 55! - Change the line ``int qcFlag() const override {return QCflags::diffref;}`` to return a different flag: ``QCflags::bounds``. This QC flag is conveniently already defined in ``ufo/filters/QCflags.h``. - Remove the lines defining ``const Variable ref_;`` and ``const Variable val_;``. In ``PracticalBoundsCheck.cc``: - Rename all references of ``DifferenceCheck`` to ``PracticalBoundsCheck``. - In ``PracticalBoundsCheck::PracticalBoundsCheck(...)``, remove references to ``ref_`` and ``val_``. - In ``PracticalBoundsCheck::applyFilter(...)``, replace the function body with something like this: .. code:: cpp const float missing = util::missingValue(missing); ufo::Variables testvars; testvars += ufo::Variables(filtervars, "ObsValue"); const float vmin = config_.getFloat("minvalue", missing); const float vmax = config_.getFloat("maxvalue", missing); // Sanity checks if (filtervars.nvars() == 0) { oops::Log::error() << "No variables will be filtered out in filter " << config_ << std::endl; ABORT("No variables specified to be filtered out in filter"); } // Loop over all variables to filter for (size_t jv = 0; jv < testvars.nvars(); ++jv) { // get test data for this variable std::vector testdata; data_.get(testvars.variable(jv), testdata); // apply the filter for (size_t jobs = 0; jobs < obsdb_.nlocs(); ++jobs) { if (apply[jobs]) { ASSERT(testdata[jobs] != missing); if (vmin != missing && testdata[jobs] < vmin) flagged[jv][jobs] = true; if (vmax != missing && testdata[jobs] > vmax) flagged[jv][jobs] = true; } } } - Feel free to customize the function further. Step 7b: Add your new filter to the build system ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ - Edit ``src/ufo/filters/CMakeLists.txt`` and add in ``PracticalBoundsCheck.cc`` and ``PracticalBoundsCheck.h``. - UFO needs to be told that another filter is available. The list of known filters is located in ``ufo/src/ufo/instantiateObsFilterFactory.h``. To add in the new filter, first add ``#include "ufo/filters/PracticalBoundsCheck.h"`` to the top of ``instantiateObsFilterFactory.h``. At the end of ``instantiateObsFilterFactory.h``, follow the pattern and add in: .. code:: cpp static oops::FilterMaker > practicalBoundsCheckMaker("Practical Bounds Check"); - The filter is added! Step 8: Compile your code ------------------------------ Finally, return to the build directory (``~/jedi/build``) and run ecbuild again: .. code:: bash cd ~/jedi/build ecbuild ../fv3-bundle We want to re-run ecbuild because we added source code files to UFO. Once ecbuild completes, verify that it reports that configuration has succeeded. Now that you have modified the ufo source code, recompile it. To save a little time, you can go directly to the ufo directory and just compile that: .. code:: bash cd ~/jedi/build/ufo make -j4 If an error is reported, review the console to see what went wrong. If you do not know what to fix, please ask for help. Once the build succeeds, congratulations! Depending on how much time you have left, you may run the unit tests with (if you run this from the ``~/jedi/build/ufo`` subdirectory, it will only run the ufo tests): .. code:: bash ctest CTest will then validate the compiled programs. A note on the ufo_coding_norms test ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ This test runs ``cpplint``, which is a command-line tool to check C/C++ files for style issues following `Google's C++ style guide `_. We use several rules in this style guide to ensure that code that we write is readable by other people. If you see an error in the ``ufo_coding_norms`` test, this indicates that the style checker has detected an issue. To view the output of a failed ``ufo_coding_norms`` test, run: .. code:: bash ctest -V -R ufo_coding_norms Then, apply any fixes to your code and try again. Step 9: Add in the YAML that describes this filter to a test ------------------------------------------------------------- This is covered in the later tutorial on unit testing. .. _structures and classes: http://www.cplusplus.com/doc/tutorial/structures/ .. _serialize: https://en.wikipedia.org/wiki/Serialization .. _a byte stream: https://en.wikipedia.org/wiki/Bitstream .. _YAML: https://yaml.org/about.html .. _several: https://yaml.org/ .. _this link: https://docs.ansible.com/ansible/latest/reference_appendices/YAMLSyntax.html .. _[link]: https://github.com/JCSDA/ufo/tree/develop/src/ufo/filters .. _[Google]: http://google.github.io/styleguide/cppguide.html