Skip to content

Releases: ServiceNow/BrowserGym

v0.7.0: changes in experiments

20 Sep 19:58

Choose a tag to compare

browsergym-experiments

  • Breaking change

    • save package versions to a separate file package_versions.txt instead of the logs #152
  • Bugfixes

v0.6.4: bugfix

19 Sep 13:50

Choose a tag to compare

browsergym-core

  • Bugfix
    • fixed bug in Frame has been detached catching #153

v0.6.3: more robust IFrame handling

17 Sep 18:50
2a171c4

Choose a tag to compare

browsergym-core

  • Improvements
    • ignoring Frame has been detached errors during unmarking #148 #147
    • more robust handling of missing frames in extract_merged_axtree #148 #146

v0.6.2: minor update

17 Sep 14:52

Choose a tag to compare

browsergym-experiments

  • New features
    • new field html_page: str in AgentInfo to enable agents to send an HTML page for visualization #13

v0.6.1: more robust iFrames

17 Sep 14:30
71a452e

Choose a tag to compare

browsergym-core

  • Improvements
    • the 26 hard limit on max number of iFrames per page is removed #143
    • handling more Frame was detached errors during marking / unmarking #143

v0.6.0: `agent_name` field

12 Sep 15:22
ced1a6c

Choose a tag to compare

browsergym-experiments

  • Breaking change
    • AbstractAgentArgs.agent_name becomes a field (instead of a property) #138

v0.5.1: minor bugfix

11 Sep 16:20
23e7c14

Choose a tag to compare

browsergym-experiments

  • Bugfix
    • Fixed a bug due to an incompatibility with joblib when running pip list or pip freeze. Python distributions are now extracted using importlib and logged in the regular logfile #135

v0.5.0: improved page marking

10 Sep 20:34
ce4eb4b

Choose a tag to compare

browsergym-core

  • New features

    • new option tags_to_mark: Literal["all", "default_html"] for all BrowserGym environments #129
      When set to "all", all page elements get marked and receive a BID.
      When set to "default_html", only standard HTML tags get marked and receive a BID (default behaviour).
  • Improvements

    • the bid of div and span elements will now be available in the AXTree as well as the DOM #119
      The aria-description field is now used in addition to the aria-roledescription as a fallback. This allows us to recover the bid of elements with generic role (div, span) for which aria-roledescription is not supported in the AXTree. Thanks @MurtyShikhar for noticing #110

browsergym-experiments

  • New features
    • The output of pip list and pip freeze is now logged when an experiment is ran #130

v0.4.3: minor improvements

05 Sep 17:44
c088e18

Choose a tag to compare

browsergym-core

  • New features
    • HighLevelActionSet has a new option retry_with_force #118
      When activated, all Playwright actions will be retried with argument force=True if they fail. For example:
      def check(bid: str):
          ...
          if retry_with_force:
              try:
                  elem.check(timeout=500)
              except Exception as e:
                  elem.check(force=True, timeout=500)
          else:
              elem.check(timeout=500)

browsergym-experiments

  • Improvements

    • More robust experiments loop #120
  • New Features

    • Introducing experiment IDs (preparing for dependency graph between experiments) #121

v0.4.2: bgym namespace

01 Aug 14:56
418421a

Choose a tag to compare

browsergym-experiments

  • New convenience namespace bgym, available whenever the browsergym-experiments package is installed
    import bgym
    
    bgym.Agent
    bgym.AgentInfo
    bgym.EnvArgs
    bgym.ExpArgs
    bgym.AbstractAgentArgs
    bgym.ExpResult
    bgym.StepInfo
    bgym.StepTimestamps
    bgym.AbstractActionSet
    bgym.HighLevelActionSet
    bgym.PythonActionSet