evaluator 0.4.0 2019-04-10

This release includes a greatly improved flow when starting directly from quantitative inputs. In addition to the spreadsheet importing flow, users have the option to start with quantitative inputs and leverage the new tidyrisk_scenario objects to move directly to simulation. Users with quantitative flows may also use collector to work with their quantitative data.

New Feature

  • read_qualitative_inputs() pulls in all qualitative inputs.
  • tidyrisk_scenario object for containing the elements of a risk scenario for simulation.
    • A RStudio Add-in for creating a skeleton tidyrisk_scenario object is available.
  • summarize_scenario() function creates summary for a single scenario’s results.
    • summarize_scenarios() is now a wrapper around summarize_scenario()
  • An alternative model openfair_tef_tc_diff_plm_sr() is now available for simulating scenarios with secondary loss risk.
  • loss_scatterplot() generates a scatterplot of total loss exposure vs. number of loss events for a single scenario.
  • exposure_histogram() generates a histogram of losses for a single scenario, optionally displaying the 95% value at risk level.
  • loss_exceedance_curve() generates a loss exceedance curve for one or more simulations summarized at the iteration level.

Improvements

  • Massive speed fix for a slow down that has been present since the v0.2.x series. Initial testing shows improvements of up to 700% when running simulations!
  • Summary statistics for mean TC and DIFF exceedance now properly handle extreme situations where 100% of threat events are either resisted or become loss events.
  • All percentages are consistently imported, stored, and processed as decimal values (from 0 - 1). This makes the TC and DIFF parameters use the same scale as VULN.
    • As a side benefit, the restrictions on the format of the mappings CSV are relaxed. Doubles (decimals) are now permitted in the low, most likely, and high columns.
    • The custom percent format function in the default Risk Report is removed and functionality is now provided via scales::percent().
  • Scenario Explorer application now can be used when skipping qualitative data and starting directly from quantitative data.
  • Risk Report can now be used when skipping qualitative data and starting directly from quantitative data.
  • The risk report and risk dashboard now use the same metric for minimum expected losses. The previous value reported on the dashboard was incorrect.
  • select_loss_events() now returns zeros when there are zero threat events in a given period. Previously this function returned NA which could potentially cause issues in reporting.
  • Updated the default sample mapping files to reference the current maximum OCR fine (SLE), from the 2015 Anthem breach.
  • All save and load functions consistently use rds formatted files instead of a mix of rda/rds.
  • Removed all remaining uses to SE forms of dplyr verbs.
  • Removed use of soft deprecated ggplot aes_() functions.
  • Clean up CSS for HTML reports, improving style consistency, particularly around font families.
  • Increase test coverage, including moving more shinytest tests to run on CI instances and package spelling tests.
  • The internal contract between run_simulation() and modeling functions such as openfair_tef_tc_diff_lm() is established. run_simulation() will always confirm a scenario object has all of the OpenFAIR components needed by the specified modeling function. This enables easier extension of Evaluator to new models.
  • openfair_example() uses (at last!) the same simulation engine as run_simulation(). This provides more consistent results, less code to maintain, and opens this demo app to displaying more metrics that are native to evaluator. A very initial pass to this last point has implemented.

Other Changes

  • generate_scatterplot() is deprecated in favor of the new loss_scatterplot() function.
  • Switched the base font for the Risk Report to Open Sans, retaining the Condensed version for headers.
  • Renamed the default sample dataset to the hypothetical MetroCare Hospital. The default sets now all consistently use the mc_ prefix to distinguish them from parameter names.
  • Deprecated load_data(). Use read_qualitative_inputs() or read_quantitative_inputs() as appropriate.
  • Re-export the pipe operator %>%
  • Rename simulation column to iteration to be more consistent with general MC uses.
  • Move from soft-deprecated purrr::invoke() to rlang::eval().

evaluator 0.3.2 2019-02-05

Improvements

  • Removed dependency on purrrlyr
  • Improve checks on summary function to avoid floating point comparisons generating false failures.

evaluator 0.3.1 2018-10-18

  • Correct ORCID identifier

New Feature

  • read_quantitative_inputs() added to allow easier skipping over qualitative inputs and going straight to quantitative inputs, such as generated by collector.

Bug fix

  • summarize_domains() was still referencing an ALE column which did not exist on the summary roll up (aggregating ALE is possible as a strict sum across scenarios).

evaluator 0.3.0 2018-07-02

  • Data structure changes
    • IDs for simulations and capabilities no longer need to be numeric. ID styles in the format of “FOO-123” and “MY_Scenario” are now supported.
    • Quantified scenarios now store parameters for TEF, TC, LM, and DIFF values as list columns. This allows non L/ML/H/CONF distributions to be more easily sampled. Qualitative scenario structure is unchanged so this should have no impact on most users.
    • capabilities table has renamed the id column to capability_id to be consistent with other ID columns throughout the schema. Survey (Excel) users are not impacted by this change.
  • Model interface change - With the unification on list columns for OpenFAIR parameters, the top level model objects no longer take a dedicated diff_estimates option. The run_simulations() function accounts for this change. Users using the standard flow will not be impacted.
  • summarize_domains() - Incorporates the now removed calculate_domain_impact() and calculate_weak_domains() functions. As part of this consolidation, the mean_tc_exceedance and mean_diff_exceedance calculations are improved by handling NAs in some simulations without zeroing out the entire calculation.
  • summarize_domains() - Properly calculates mean_diff_exceedance when there all threat events are successfully avoided.
  • generate_heatmap() - Takes a domain_summary input rather than the deprecated domain_impact structure.

Bug Fixes

  • Using distributions not in the base or stats namespaces was practically impossible. All atomic OpenFAIR functions have been refactored to take a fully qualified function (i.e. EnvStats::rnormTrunc).
  • explore_scenarios() was trying to assign a mappings variable to the global context, which rightly failed. Scaled back the assignment to the current scope.
  • select_loss_opportunities() properly returns an NA for the threat & difficulty exceedance calculations when there are no threat events in a given simulated period.
  • summarize_scenarios() - correctly handles scenarios in which no threat events occur in a given simulation. This bug was limited to mean_tc_exceedance. For previously run simulations, re-summarizing the scenario_results will generate corrected values.
  • calculate_max_losses() - no longer returns a duplicate set of results if not passed any outliers.

Improvements

  • run_simulations() - New ale_maximum parameter allows an absolute cap on per simulation annual losses to be set. This is an interim step in lieu of full hierarchical interaction modeling.
  • run_simulations() - Errors encountered during runs are now reported better.
  • run_simulations() - Implements parallel execution via the furrr package. To run simulations across all cores of a local machine, load furrr and run plan(multicore) before launching an analysis. For more information, see the furrr::future_map() documentation.
  • sample_lm() and sample_tc() check if they are asked to generate zero requested samples, bypassing calling the underlying generation function. This avoids problems with generating functions which do not gracefully handle being asked to sample a non positive number (zero) of events.
  • load_data() now fully specifies the expected CSV file formats, avoiding possible surprises and making invocations less noisy on the console.
  • Removed all deprecated standard-evaluation tidyverse verbs in favor of rlang::.data constructs, making CRAN checks much simpler.
  • Minor documentation cleanup.

evaluator 0.2.3 2018-04-09

Bug Fixes

  • Additional logic in tests to verify that supplemental packages are available, skipping the test if said packages are not installed.
  • Update tests to convert tibbles to data.frames prior to comparison as a work around for dplyr/2751.

evaluator 0.2.2 2018-03-30

Analysis Change

  • Previous versions sampled threat frequency (TEF) as a continuous distribution. Threat event counts are discrete (they either happen or don’t in a given simulated period), so this previous method was incorrect and inflated threat counts. The differences are small in the standard 10,000 evaluation range, but users should note the change.
  • Part of the discretization step for TEF has a positive side effect of making the threat_event column in data objects an integer rather than a double- long. As this is still a numeric type, there should be no impact on users for this change apart from a slightly tidier data output.

Bug Fixes

  • Update risk_dashboard to not cache any chunks. This keeps render from from trying to write to the package install directory.
  • Updated tests to use four significant digits where long reference values are passed, avoiding precision issues on machines without long doubles.

evaluator 0.2.1 2018-03-22

Bug Fixes

  • Document optional dependency on pandoc and add test skip logic to avoid Rmarkdown tests on systems where pandoc is not available.
    • Add testing to Travis matrix for Linux builds with and without pandoc.
  • run_analysis script was missing namespace calls for readr::cols.
  • All Rmarkdown calls now specify an intermediates_dir value of tempdir(). This can be overwritten on the function call if needed.
  • Report generation is now much quieter by default.

evaluator 0.2.0 2018-03-13

New Functionality

  • New sample dataset: mappings contains sample qualitative to quantitative parameters.
  • risk_dashboard expects a mandatory output parameter to the desired rendered output.
  • New pkgdown generated web documentation at https://evaluator.severski.net.
  • Refactored generate_report function.
  • Expose OpenFAIR model selection in run_simulation() call
    • Provide default TEF/TC/DIFF/LM OpenFAIR model
  • New create_templates() function for populating starter/sample files, making starting a fresh analysis easier than ever!
  • Experimental quick start script, run_analysis.R, supplied with create_templates().
  • All default directories normalized to ~/evaluator/[inputs|results]
  • New OpenFAIR primitives:
    • sample_tef
    • sample_lm
    • sample_tc
    • sample_diff
    • sample_vuln
  • New composition functions:
    • compare_tef_vuln
    • select_loss_opportunities
  • New difficulty composition functions:
    • get_mean_control_strength

Bug Fixes

  • Improved help documentation on many functions
  • Update of usage vignette
  • Auto loads extrafont database for better font detection
    • Falls back to standard sans family when none of the preferred options are available
  • Drop use of tcltk progress bar in favor of console-compatible dplyr::progress_estimated(). Also enables reduced package dependencies.
  • Tests and code coverage reporting added
    • Improve faulty capabilities validation
  • Removed dependency on purrrlyr

Miscellaneous Changes

  • generate_report defaults to creating a MS Word document as the output type

evaluator 0.1.1 2017-11-20

  • Replaced dependency on modeest with a slimmer statip dependency
  • Removed dependency on magrittr
  • Default (overridable) locations of input and results directories now consistently set to “~/data” and “~/results” respectively
  • generate_report() now takes an optional focus_scenario_ids parameter to override the scenarios on which special emphasis (usually executive interest) is desired.
  • Improve user experience for optional packages. User is now prompted to install optional dependencies (shiny, DT, flexdashboard, statip, rmarkdown, etc.) when running reporting functionality which requires them.
  • Substantial improvements in the sample analysis flow detailed in the usage vignette. You can now actually run all the commands as-is and have them work, which was previously “challenging”.
  • summarize_all() renamed to the more descriptive summarize_to_disk() to avoid dplyr conflict
  • Add requirement for at least pander v0.6.1 for tibble compatibility
  • Substantial refactoring on vignette
    • Added missing save steps
    • Corrected package name for Viewer to rstudioapi
    • Fixed a few incorrect placeholders
    • Properly committed compiled files to package for distribution and installation
  • Update all tidyverse calls to account for deprecations and split out of purrrlyr
  • Windows CI builds added via AppVeyor
  • Use annotate_logticks() over manual breaks on risk_dashboard

evaluator 0.1.0 2017-02-26

  • Initial submission to CRAN