v7.5 (2023-10-21)


  • Provide support for HTCondor clusters in a third-party repository and add link to docs (Martín Pozo).

  • Add documentation for AI Basel’s infai_3 partition (Silvan Sievers).

  • Don’t rely on the existence of the ‘runs-00001-00100’ dir when fetching results (Jendrik Seipp).

v7.4 (2023-08-18)


  • Require revision_cache parameter in CachedRevision constructor (Jendrik Seipp).

  • Add subdir option for CachedRevision to support solvers at deeper levels of a repo (Jendrik Seipp).

  • Add CachedRevision.get_relative_exp_path() method to query where cache artefacts will land in the experiment directory (Jendrik Seipp).

  • Document CachedRevision class and stabilize its API (Jendrik Seipp).

  • Only use documented classes and functions in example experiments (Jendrik Seipp).

Downward Lab

v7.3 (2023-03-03)


  • Transparently handle xz-compressed properties files (Jendrik Seipp).

  • Add CI tests for Python 3.11 (Jendrik Seipp).

Downward Lab

  • Adapt code for Matplotlib version 3.7 (Jendrik Seipp).

v7.2 (2022-10-09)


  • Raise minimum supported Python version to 3.7 (Jendrik Seipp).

  • Add support for Python 3.10 (Jendrik Seipp).

  • Apply parsing functions in the order in which they were added (Jendrik Seipp).

  • For contributors: document pre-commit hook in file (Jendrik Seipp).

Downward Lab

  • Parse peak memory in anytime search parser (Jendrik Seipp).

  • Only store “planner_memory” and “planner_time” attributes for successful planner runs (Jendrik Seipp).

  • Add fully customizable example planner experiment without FastDownwardExperiment class (Jendrik Seipp).

  • Show how to group domain directories in example Fast Downward experiment (Jendrik Seipp).

v7.1 (2022-06-20)


  • Revamp Singularity example experiment: use runsolver to limit resource usage (Silvan Sievers and Jendrik Seipp).

Downward Lab

  • Fix header sizes in HTML reports (Jendrik Seipp).

  • Include domains in attribute overview tables even if none of their tasks has an attribute value for all algorithms (Jendrik Seipp).

  • Compute “score_planner_time” and “score_planner_memory” attributes in planner parser (Jendrik Seipp).

  • Only consider files ending with “.pddl” and “.sas” when building suites (Jendrik Seipp).

  • Explicitly left-align non-numeric cells to avoid \multicolumn entries in Latex output (Jendrik Seipp).

v7.0 (2021-10-24)


  • Remove support for Mercurial repositories (Jendrik Seipp).

Downward Lab

  • Fix rules for finding domain files for airport and psr-small domains (Silvan Sievers).

  • Add more ticks on y axis in relative plots (Jendrik Seipp).

v6.5 (2021-09-27)


  • Allow rerunning experiments. This is useful if some runs were never started, for example, due to grid node failures. All runs that have already been started are skipped. For more information see the corresponding FAQ (Jendrik Seipp).

Downward Lab

  • Slightly generalize rules for finding domain files, adapted from Fast Downward (Silvan Sievers).

v6.4 (2021-07-06)


  • Add TetralithEnvironment for the NSC cluster in Linköping (Jendrik Seipp).

  • Automatically group multiple runs into one Slurm task when the number of runs exceeds the maximum number of Slurm tasks (Jendrik Seipp).

  • Add time_limit_per_task parameter to SlurmEnvironment (Jendrik Seipp).

  • Add cpus_per_task parameter to SlurmEnvironment (#98, Lucas Galery Käser).

  • Catch OverflowError when casting large ints to floats (#95, Silvan Sievers).

Downward Lab

  • None.

v6.3 (2021-02-14)


  • Use long Git revision hashes for revision cache. The short ones differ in length between Git versions (Jendrik Seipp).

  • Run continuous integration tests for Python 3.9 (Jendrik Seipp).

Downward Lab

  • Remove “revision_summary” column from info table (Jendrik Seipp).

v6.2 (2020-10-20)


  • Reports: round values to desired precision before determining colors (Jendrik Seipp).

  • Restructure and extend documentation (Jendrik Seipp).

  • For developers: run CI tests on Ubuntu 20.04 in addition to 18.04 (Jendrik Seipp).

Downward Lab

  • Allow adding SAS+ files with FastDownwardExperiment.add_suite() (Jendrik Seipp).

v6.1 (2020-09-15)


  • Take float precision into account when highlighting table cells (Jendrik Seipp).

  • Allow serializing pathlib.Path objects into JSON files (Jendrik Seipp).

  • For developers: add .github/ file (Jendrik Seipp).

  • For developers: separate tests for Singularity and FF example experiments from other tests (Jendrik Seipp).

  • For developers: skip cached_revision doctests if DOWNWARD_REVISION_CACHE variable is not set (Jendrik Seipp).

  • For developers: use f-strings in code (Jendrik Seipp).

Downward Lab

  • Print number of tasks above and below separator lines in scatter plots (Jendrik Seipp).

  • Ignore tasks for which runs have been filtered out in aggregate reports (Jendrik Seipp).

  • Fix order of bracketed task counts per domain in table reports (Jendrik Seipp).

  • Gracefully handle empty scatter plots (Jendrik Seipp).

  • Make score_* attributes absolute, i.e., include tasks for which not all algorithms have a value in aggregations (Jendrik Seipp).

v6.0 (2020-04-05)


  • Bump minimum Python version to 3.6.

  • Move CachedRevision from downward to lab package (Thomas Keller). Please note that the interface to the class is experimental and may change in the future. Feedback is welcome!

  • Let tests fail if any example experiment produces unexplained errors.

Downward Lab

  • No changes.

v5.5 (2020-03-13)


  • Sort numbers with suffixes (5K, 2M, 8G) and “infinity” correctly in tables.

  • Gracefully handle missing “info” or “summary” tables in HTML reports.

  • Abort if a function is passed to a filter_* kwarg.

  • Abort if a filter checks missing attribute names (e.g., when passing filter_algorithms instead of filter_algorithm).

Downward Lab

  • Add example experiment for running Singularity planner images.

v5.4 (2020-03-01)


  • Use newer txt2tags version and remove bundled copy.

  • Call parsers with active Python interpreter.

  • Don’t call deprecated time.clock() (removed in Python 3.8).

  • Don’t add Lab to PYTHONPATH in BaselSlurmEnvironment.

Downward Lab

  • Revision cache: only delete “misc” and “experiments” dirs if they exist (Maximilian Fickert).

v5.3 (2020-02-03)


Downward Lab

  • Remove call to rm -f Newer Fast Downward versions remove the temporary file automatically. If you want to keep the file, add "--keep-sas-file" to the driver_options.

  • Fix ScatterPlotReport: skip None values in max() computation (Silvan Sievers).

  • Fix ScatterPlotReport: place diagonal line correctly even if axis scales differ.

v5.2 (2020-01-07)


  • Use line buffering for run.err files.

Downward Lab

  • Preserve line breaks for error logs in tables.

  • If an error log in a table has more than 100 lines, omit surplus lines from the middle of the log.

  • Always print the number of runs with unexplained errors when generating any type of report.

v5.1 (2019-12-10)


  • Test Lab on Python 3.8.

  • Use active Python version to call run files in local experiments.

Downward Lab

  • Support Fast Downward Git repos (Patrick Ferber).

v5.0 (2019-12-04)


  • Deprecate support for Python versions 2.7 to 3.5.

  • Allow only a single aggregation function for Attribute objects.

  • If there is only a single HTML table, show it when the page loads.

  • Remove broken --log-level command line parameter. You can call tools.configure_logging(logging.DEBUG) to enable debug messages instead.

  • Pass old hard memory limit when setting soft memory limit.

Downward Lab

  • Scatter plots:

    • Add relative parameter for drawing relative scatter plots.

    • Draw points for algorithm pairs with missing values on axis boundaries.

    • Allow drawing negative values on linear and symlog axes.

    • Remove xscale and yscale parameters in favor of a new scale parameter.

    • Fold PlotReport class into ScatterPlotReport.

    • Simplify code by letting Matplotlib compute axis limits automatically.

v4.2 (2019-09-27)


  • Upload to PyPI. Install Lab and Downward Lab with pip install lab.

  • Add support for running Lab in Python virtual environments (Guillem).

  • Parser scripts don’t have to be executable anymore, but they must be Python scripts.

Downward Lab

  • Abort if two algorithms are identical, i.e., use the same revision, build config and commandline options.

  • Scatter plot report: include tasks for which both algorithms have no data if show_missing=True.

v4.1 (2019-06-03)

  • Add support for Python 3. Lab now supports Python 2.7 and Python >= 3.5.

v4.0 (2019-02-19)


  • Parser: don’t try to parse missing files. Print message to stdout instead.

  • Add soft memory limit of “memory_per_cpu * 0.98” for Slurm runs to safeguard against cgroup failures.

  • Abort if report contains duplicate attribute names.

  • Make reports even if fetcher detects unexplained errors.

  • Use flags='' for lab.parser.Parser.add_pattern() by default again.

  • Include node names in standard reports and warn if report mixes runs from different partitions.

  • Add new example experiment using a simple vertex cover solver.

  • BaselSlurmEnvironment: don’t load Python 2.7.11 since it might conflict with an already loaded module.

  • Raise default nice value to 5000.

Downward Lab

  • Support new Fast Downward exitcodes (Silvan).

  • Parse “planner_wall_clock_time” attribute in planner parser.

  • Include “planner_wall_clock_time” and “raw_memory” attributes in unexplained errors table.

  • Make PlanningReport more generic by letting derived classes override the new PREDEFINED_ATTRIBUTES, INFO_ATTRIBUTES and ERROR_ATTRIBUTES class members (Augusto).

  • Don’t compute the “quality” attribute automatically. The docs and show how to add the two filters that together add the IPC quality score to each run.

v3.0 (2018-07-10)


  • Add exp.add_parser() method. See also Parser (Silvan).

  • Add exp.add_parse_again_step() method for running parsers again (Silvan).

  • Require that the build, start_runs and fetch steps are added explicitly (see Experiment).

  • Remove required argument from add_resource(). All resources are now required.

  • Use stricter naming rules for commands and resources. See respective add_* methods for details.

  • Use required=False and flags='M' by default for lab.parser.Parser.add_pattern().

  • Only support custom command line arguments for locally executed steps.

  • Log errors to stderr.

  • Log exit codes and wall-clock times of commands to driver.log.

  • Add unexplained error if driver.log is empty.

  • Let fetcher fetch properties and static-properties files.

  • Remove deprecated possibility of passing Step objects to add_step().

  • Remove deprecated exp.__call__() method.

Downward Lab

  • Add “planner_timer” and “planner_memory” attributes.

  • Reorganize parsers and don’t add any parser implicitly. See Bundled parsers.

  • Add anytime-search parser that parses only “cost”, “cost:all” and “coverage”.

  • Revise and simplify single-search parser.

  • Parse new Fast Downward exit codes (

  • Don’t exclude (obsolete) “benchmarks” directory when caching revisions.

  • Only copy “raw_memory” value to “memory” when “total_time” is present.

  • Rename “fast-downward” command to “planner”.

  • Make “error” attribute optional for reports.

v2.3 (2018-04-12)


  • BaselSlurmEnvironment: Use infai_1 and normal as default Slurm partition and QOS.

  • Remove OracleGridEngineEnvironment.

Downward Lab

  • Use --overall-time-limit=30m and --overall-memory-limit=3584M for all Fast Downward runs by default.

  • Don’t add -j option to build options ( now uses all CPUs automatically).

v2.2 (2018-03-16)


  • Print run and task IDs during local experiments.

  • Make warnings and error messages more informative.

  • Abort after fetch step if fetcher finds unexplained errors.

  • Improve examples and docs.

Downward Lab

  • Don’t parse preprocessor logs anymore.

  • Make regular expressions stricter in parsers.

  • Don’t complain if SAS file is missing.

v2.1 (2017-11-27)


  • Add BaselSlurmEnvironment (Florian).

  • Support running experiments in virtualenv (Shuwa).

  • Redirect output to driver.log and driver.err as soon as possible.

  • Store all observed unexplained errors instead of a single one (Silvan).

  • Report unexplained error if run.err or driver.err contain output.

  • Report unexplained error if “error” attribute is missing.

  • Add configurable soft and hard limits for output to run.log and run.err.

  • Record grid node for each run and add it to warnings table.

  • Omit toprule and bottomrule in LaTeX tables.

  • Add lab.reports.Table.set_row_order() method.

  • Only escape text in table cells if it doesn’t contain LaTeX or HTML markup.

  • Allow run filters to change a run’s ID (needed for renaming algorithms).

  • Add merge kwarg to add_fetcher() (Silvan).

  • Exit with returncode 1 if fetcher finds unexplained errors.

  • Let fetcher show warning if slurm.err is not empty.

  • Include content of slurm.err in reports if it contains text.

  • Add continuous integration testing.

  • Add --skip-experiments option for tests/run-tests script.

  • Clean up code.

  • Polish documentation.

Downward Lab

  • For each error outcome show number of runs with that outcome in summary table and dedicated tables.

  • Add standalone exit code parser. Allow removing translate and search parsers (Silvan).

  • Allow passing Problem instances to FastDownwardExperiment.add_suite() (Florian).

  • Don’t filter duplicate coordinates in scatter plots.

  • Don’t round scatter plot coordinates.

  • Remove instead of compressing it.

  • Fix scatter plots for multiple categories and the default None category (Silvan).

v2.0 (2017-01-09)


  • Show warning and ask for action when evaluation dir already exists.

  • Add scale parameter to Attribute. It is used by the plot reports.

  • Add digits parameter to Attribute for specifying the number of digits after the decimal point.

  • Pass name, function, args and kwargs to exp.add_step(). Deprecate passing Step objects.

  • After calling add_resource("mynick", ...), use resource in commands with “{mynick}”.

  • Call: make name parameter mandatory, rename mem_limit kwarg to memory_limit.

  • Store grid job files in <exp-dir>-grid-steps.

  • Use common run-dispatcher script for local and remote experiments.

  • LocalEnvironment: support randomizing task order (enabled by default).

  • Make path parameter optional for all experiments.

  • Warn if steps are listed explicitly and --all is used.

  • Change main experiment step name from “start” to “run”.

  • Deprecate exp(). Use exp.run_steps() instead.

  • Don’t filter None values in lab.reports helper functions.

  • Make logging clearer.

  • Add example FF experiment.

  • Remove deprecated code (e.g. predefined Step objects, tools.sendmail()).

  • Remove Run.require_resource(). All resources have always been available for all runs.

  • Fetcher: remove write_combined_props parameter.

  • Remove Sequence class.

  • Parser: remove key_value_patterns parameter. A better solution is in the works.

  • Remove tools.overwrite_dir() and tools.get_command_output().

  • Remove lab.reports.minimum(), lab.reports.maximum(), lab.reports.stddev().

  • Move to

  • Rename to lab.reports.geometric_mean() and lab.reports.avg() to lab.reports.arithmetic_mean().

  • Many speed improvements and better error messages.

  • Rewrite docs.

Downward Lab

  • Always validate plans. Previous Lab versions don’t add --validate since older Fast Downward versions don’t support it.

  • HTML reports: hide tables by default, add buttons for toggling visibility.

  • Unify “score_*”, “quality” and “coverage” attributes: assign values in range [0, 1] and compute only sum and no average.

  • Don’t print tables on commandline.

  • Remove DownwardExperiment and other deprecated code.

  • Move FastDownwardExperiment into downward/

  • Rename config attribute to algorithm. Remove config_nick attribute.

  • Change call name from “search” to “fast-downward”.

  • Remove “memory_capped”, and “id_string” attributes.

  • Report raw memory in “unexplained errors” table.

  • Parser: remove group argument from add_pattern(), and always use group 1.

  • Remove cache_dir parameter. Add revision_cache parameter to FastDownwardExperiment.

  • Fetcher: remove copy_all option.

  • Remove predefined benchmark suites.

  • Remove IpcReport, ProblemPlotReport, RelativeReport, SuiteReport and TimeoutReport.

  • Rename CompareConfigsReport to ComparativeReport.

  • Remove possibility to add _relative to an attribute to obtain relative results.

  • Apply filters sequentially instead of interleaved.

  • PlanningReport: remove derived_properties parameter. Use two filters instead: one for caching results, the other for adding new properties (see QualityFilters in downward/reports/

  • PlotReport: use fixed legend location, remove category_styles option.

  • AbsoluteReport: remove colored parameter and always color HTML reports.

  • Don’t use domain links in Latex reports.

  • AbsoluteReport: Remove resolution parameter and always use combined resolution.

  • Rewrite docs.

v1.12 (2017-01-09)

Downward Lab

  • Only compress “output” file if it exists.

  • Preprocess parser: make legacy preprocessor output optional.

v1.11 (2016-12-15)


  • Add bitbucket-pipelines.yml for continuous integration testing.

Downward Lab

  • Add IPC 2014 benchmark suites (Silvan).

  • Set min_wins=False for dead_ends attribute.

  • Fit coordinates better into plots.

  • Add finite_sum() function and use it for initial_h_value (Silvan).

  • Update example scripts for repos without benchmarks.

  • Update docs.

v1.10 (2015-12-11)


  • Add permissions parameter to lab.experiment.Experiment.add_new_file().

  • Add default parser which checks that log files are not bigger than 100 MB. Maybe we’ll make this configurable in the future.

  • Ensure that resource names are not shared between runs and experiment.

  • Show error message if resource names are not unique.

  • Table: don’t format list items. This allows us to keep the quotes for configuration lists.

Downward Lab

  • Cleanup downward.suites: update suite names, add STRIPS and ADL versions of all IPCs. We recommend selecting a subset of domains manually to only run your code on “interesting” benchmarks. As a starting point you can use the suites suite_optimal_strips or suite_satisficing.

v1.9.1 (2015-11-12)

Downward Lab

  • Always prepend build options with -j<num_cpus>.

  • Fix: Use correct revisions in FastDownwardExperiment.

  • Don’t abort parser if resource limits can’t be found (support old planner versions).

v1.9 (2015-11-07)


Downward Lab

v1.8 (2015-10-02)


  • Deprecate predefined experiment steps (remove_exp_dir, zip_exp_dir, unzip_exp_dir).

  • Docs: add FAQs, update docs.

  • Add more regression and style tests.

Downward Lab

  • Parse both evaluated states (evaluated) and evaluations (evaluations).

  • Add example experiment showing how to make reports for data obtained without Lab.

  • Add suite_sat_strips().

  • Parse negative initial h values.

  • Support CMake builds.

v1.7 (2015-08-19)


  • Automatically determine whether to queue steps sequentially on the grid.

  • Reports: right-align headers (except the left-most one).

  • Reports: let return 0 if any of the numbers is 0.

  • Add test that checks for dead code with vulture.

  • Remove Step.remove_exp_dir step.

  • Remove default time and memory limits for commands. You can now pass mem_limit=None and time_limit=None to disable limits for a command.

  • Pass extra_options kwarg to lab.environments.OracleGridEngineEnvironment to set additional options like parallel environments.

  • Sort properties files by keys.

Downward Lab

  • Add support for new python driver script

  • Use booktabs package for latex tables.

  • Remove vertical lines from Latex tables (recommended by booktabs docs).

  • Capitalize attribute names and remove underscores for Latex reports.

  • Allow fractional plan costs.

  • Set search_time and total_time to 0.01 instead of 0.1 if they are 0.0 in the log.

  • Parse initial h-value for aborted searches (Florian).

  • Use EXIT_UNSOLVABLE instead of logs to determine unsolvability. Currently, this exit code is only returned by EHC.

  • Exit with warning if search parser is not executable.

  • Deprecate downward/ module.

  • Deprecate examples/ module.

  • Remove script.

  • By default, use all CPUs for compiling Fast Downward.



  • Restore earlier default behavior for grid jobs by passing all environment variables (e.g. PYTHONPATH) to the job environments.

Downward Lab

  • Use write-once revision cache: instead of cloning the full FD repo into the revision cache only copy the src directory. This greatly reduces the time and space needed to cache revisions. As a consequence you cannot specify the destination for the clone anymore (the dest keyword argument is removed from the Translator, Preprocessor and Planner classes) and only local FD repositories are supported (see downward.checkouts.HgCheckout). After the files have been copied into the cache and FD has been compiled, a special file (build_successful) is written in the cache directory. When the cached revision is requested later an error is shown if this file is missing.

  • Only use exit codes to reason about error reasons. Merge from FD main if your FD version does not produce meaningful exit codes.

  • Preprocess parser: only parse logs and never output files.

  • Never copy all.groups and test.groups files. Old Fast Downward branches need to merge from main.

  • Always compress (also for compact=False). Use xz for compressing.



  • Add Experiment.add_fetcher() method.

  • If all columns have the same value in an uncolored table row, make all values bold, not grey.

  • In Experiment.add_resource() and Run.add_resource() set dest=None if you don’t want to copy or link the resource, but only need an alias to reference it in a command.

  • Write and keep all logfiles only if they actually have content.

  • Don’t log time and memory consumption of process groups. It is still an unexplained error if too much wall-clock time is used.

  • Randomize task order for grid experiments by default. Use randomize_task_order=False to disable this.

  • Save wall-clock times in properties file.

  • Do not replace underscores by dashes in table headers. Instead allow browsers to break lines after underscores.

  • Left-justify string and list values in tables.

Downward Lab

  • Add optional nick parameter to Translator, Preprocessor and Planner classes. It defaults to the revision name rev.

  • Save hg id output for each checkout and include it in reports.

  • Add timeout parameter to DownwardExperiment.add_config().

  • Count malformed-logs as unexplained errors.

  • Pass legend_location=None if you don’t need a legend in your plot.

  • Pass custom benchmark directories in DownwardExperiment.add_suite() by using the benchmarks_dir keyword argument.

  • Do not copy logs from preprocess runs into search runs.

  • Reference preprocessed files in run scripts instead of creating links if compact=True is given in the experiment constructor (default).

  • Remove unexplained_error attribute. Errors are unexplained if run['error'] starts with ‘unexplained’.

  • Remove *_error attributes. It is preferrable to inspect *_returncode attributes instead (e.g. search_returncode).

  • Make report generation faster (10-fold speedup for big reports).

  • Add DownwardExperiment.add_search_parser() method.

  • Run make clean in revision-cache after compiling preprocessor and search code.

  • Strip executables after compilation in revision-cache.

  • Do not copy Lab into experiment directories and grid-steps. Use the global Lab version instead.



  • Add exp.add_report() method to simplify adding reports.

  • Use simplejson when available to make loading properties more than twice as fast.

  • Raise default check-interval in Calls to 5s. This should reduce Lab’s overhead.

  • Send mail when grid experiment finishes. Usage: MaiaEnvironment(email='').

  • Remove steps.Step.publish_reports() method.

  • Allow creating nested new files in experiment directory (e.g. exp.add_new_file('path/to/file.txt')).

  • Remove duplicate attributes from reports.

  • Make commandline parser available globally as lab.experiment.ARGPARSER so users can add custom arguments.

  • Add cache_dir parameter in Experiment for specifying where Lab stores temporary data.

Downward Lab

  • Move downward.experiment.DownwardExperiment to downward.experiments.DownwardExperiment, but keep both import locations indefinitely.

  • Flag invalid plans in absolute reports.

  • PlanningReport: When you append ‘_relative’ to an attribute, you will get a table containing the attribute’s values of each configuration relative to the leftmost column.

  • Use bzip2 for compressing files instead of tar+gzip to save space and make opening the files easier.

  • Use bzip2 instead of gzip for compressing experiment directories to save space.

  • Color absolute reports by default.

  • Use log-scale instead of symlog-scale for plots. This produces equidistant grid lines.

  • By default place legend right of scatter plots.

  • Remove --dereference option from tar command.

  • Copy (instead of linking) PDDL files into preprocessed-tasks dir.

  • Add table with Fast Downward commandline strings and revisions to AbsoluteReport.



  • For Latex tables only keep the first two and last two hlines.

Downward Lab

  • Plots: Make category_styles a dictionary mapping from names to dictionaries of matplotlib plotting parameters to allow for more and simpler customization. This means e.g. that you can now change the line style in plots.

  • Produce a combined domain- and problem-wise AbsoluteReport if resolution=combined.

  • Include info in AbsoluteReport if a table has no entries.

  • Plots: Add params argument for specifying matplotlib parameters like font-family, label sizes, line width, etc.

  • AbsoluteReport: If a non-numerical attribute is included in a domain-wise report, include some info in the table instead of aborting.

  • Add Attribute class for wrapping custom attributes that need non-default report options and aggregation functions.

  • Parse expansions_until_last_jump attribute.

  • Tex reports: Add number of tasks per domain with new \numtasks{x} command that can be cutomized in the exported texts.

  • Add pgfplots backend for plots.



  • Fetcher: Only copy the link not the content for symbolic links.

  • Make properties files more compact by using an indent of 2 instead of 4.

  • Nicer format for commandline help for experiments.

  • Reports: Only print available attributes if none have been set.

  • Fetcher: Pass custom parsers to fetcher to parse values from a finished experiment.

  • For geometric mean calculation substitute 0.1 for values <= 0.

  • Only show warning if not all attributes for the report are found in the evaluation dir, don’t abort if at least one attribute is found.

  • If an attribute is None for all runs, do not conclude it is not numeric.

  • Abort if experiment path contains a colon.

  • Abort with warning if all runs have been filtered for a report.

  • Reports: Allow specifying a single attribute as a string instead of a list of one string (e.g. attributes=’coverage’).

Downward Lab

  • If compact=True for a DownwardExperiment, link to the benchmarks instead of copying them.

  • Do not call ./build-all script, but build components only if needed.

  • Fetch and compile sources only when needed: Only prepare translator and preprocessor for preprocessing experiments and only prepare planners for search experiments. Do it in a grid job if possible.

  • Save space by deleting the benchmarks directories and omitting the search directory and validator for preprocess experiments.

  • Only support using ‘src’ directory, not the old ‘downward’ dir.

  • Use downward script regardless of other binaries found in the search directory.

  • Do not try to set parent-revision property. It cannot be determined without fetching the code first.

  • Make ProblemPlotReport class more general by allowing the get_points() method to return an arbitrary number of points and categories.

  • Specify xscale and yscale (linear, log, symlog) in PlotReports.

  • Fix removing downward.tmp.* files (use bash for globbing). This greatly reduces the needed space for an experiment.

  • Label axes in ProblemPlots with xlabel and ylabel.

  • If a grid environment is selected, use all CPUs for compiling Fast Downward.

  • Do not use the same plot style again if it has already been assigned by the user.

  • Only write plot if valid points have been added.

  • DownwardExperiment: Add member include_preprocess_results_in_search_runs.

  • Colored reports: If all configs have the same value in a row and some are None, highlight the values in green instead of making them grey.

  • Never set ‘error’ to ‘none’ if ‘search_error’ is true.

  • PlotReport: Add legend_location parameter.

  • Plots: Sort coordinates by x-value for correct connections between points.

  • Plots: Filter duplicate coordinates for nicer drawing.

  • Use less padding for linear scatterplots.

  • Scatterplots: Add show_missing parameter.

  • Absolute reports: For absolute attributes (e.g. coverage) print a list of numbers of problems behind the domain name if not all configs have a value for the same number of problems.

  • Make ‘unsolvable’ an absolute attribute, i.e. one where we consider problem runs for which not all configs have a value.

  • If a non-numeric attribute is present in a domain-wise report, state its type in the error message.

  • Let plots use the format parameter given in constructor.

  • Allow generation of pgf plot files (only available in matplotlib 1.2).

  • Allow generation of pdf and eps plots.

  • DownwardReport: Allow passing a single function for derived_properties.

  • Plots: Remove code that sets parameters explicitly, sort items in legend.

  • Add parameters to PlotReport that set the axes’ limits.

  • Add more items to Downward Lab FAQ.



  • Add filter shortcuts: filter_config_nick=['lama', 'hcea'], filter_domain=['depot'] (see Report) (Florian)

  • Ability to use more than one filter function (Florian)

  • Pass an optional filter to Fetcher to fetch only a subset of results (Florian)

  • Better handling of timeouts and memory-outs (Florian)

  • Try to guess error reason when run was killed because of resource limits (Florian)

  • Do not abort after failed commands by default

  • Grid: When –all is passed only run all steps if none are supplied

  • Environments: Support Uni Basel maia cluster (Malte)

  • Add “pi” example

  • Add example showing how to parse custom attributes

  • Do not add resources and files again if they are already added to the experiment

  • Abort if no runs have been added to the experiment

  • Round all float values for the tables

  • Add function for sending e-mails

  • Many bugfixes

  • Added more tests

  • Improved documentation

Downward Lab

  • Make the files, domain.pddl and problem.pddl optional for search experiments

  • Use more compact table of contents for AbsoluteReports

  • Use named anchors in AbsoluteReport (report.html#expansions, report.html#expansions-gripper)

  • Add colored absolute tables (see AbsoluteReport)

  • Do not add summary functions in problem-wise reports

  • New report class ProblemPlotReport

  • Save more properties about experiments in the experiments’s properties file for easy lookup (suite, configs, portfolios, etc.)

  • Use separate table for each domain in problem-wise reports

  • Sort table columns based on given config filters if given (Florian)

  • Do not add VAL source files to experiment

  • Parse number of reopened states

  • Remove temporary Fast Downward files even if planner was killed

  • Divide scatter-plot points into categories and lable them (see ScatterPlotReport) (Florian)

  • Only add a highlighting and summary functions for numeric attributes in AbsoluteReports

  • Compile validator if it isn’t compiled already

  • Downward suites: Allow writing SUITE_NAME_FIRST to run the first instance of all domains in SUITE_NAME

  • LocalEnvironment: If processes is given, use as many jobs to compile the planner in parallel

  • Check python version before creating preprocess experiment

  • Add avg, min, max and stddev rows to relative reports

  • Add RelativeReport

  • Add DownwardExperiment.set_path_to_python()

  • Many bugfixes

  • Improved documentation