Sacred v0.7.3 Release Notes
Release Date: 2018-05-06 // about 6 years ago-
🛠 Major bugfix release that fixes several critical issues including: experiments that sometimes didn't exit, racing conditions in the FileStorage and MongoObservers and several stdout-capturing problems.
- 🔋 Feature: support custom experiment base directory (thanks @anibali)
- 🔋 Feature: added option to pass existing MongoClient to MongoObserver (thanks @rueberger)
- 🔋 Feature: allow setting the config docstring from named configs
- 🔋 Feature: added py-cpuinfo as fallback for gathering CPU information (thanks @serv-inc)
- 🔋 Feature: added support for _log argument in config function
- 🛠 Bugfix: stacktrace filtering now correctly handles chained exceptions (thanks @kamo-naoyuki)
- 🛠 Bugfix: resolved issue with stdout capturing sometimes loosing the last few lines
- 🛠 Bugfix: fixed the overwrite option of MongoObserver
- 🛠 Bugfix: fixed a problem with the heartbeat sometimes not ending
- 🛠 Bugfix: fixed an error with running in interactive mode
- 🛠 Bugfix: added a check for non-unique ingredient paths (thanks @boeddeker)
- 🛠 Bugfix: fixed several problems with UTF-8 decoding (thanks @LukasDrude, @wjp)
- 🛠 Bugfix: fixed nesting structure of _config (thanks @boeddeker)
- 🛠 Bugfix: fixed crash when using git integration with empty repository (thanks @ramon-oliveira)
- 🛠 Bugfix: fixed a crash with first run using sqlite backend
- 🛠 Bugfix: fixed several problem with the tests (thanks @thomasjpfan)
- 🛠 Bugfix: fixed racing condition in FileStorageObserver (thanks @boeddeker)
- 🛠 Bugfix: fixed problem with overwriting named configs of ingredients (thanks @pimdh)
- 🛠 Bugfix: removed deprecated call to inspect.getargspec()
- 🛠 Bugfix: fixed problem with empty dictionaries disappearing from config updates and named configs (thanks @TomVeniat)
- 🛠 Bugfix: fixed problem with commandline parsing when program name contained spaces
- 🛠 Bugfix: loglevel option is now taken into account for config related warnings
- 🛠 Bugfix: properly handle numpy types in metrics logging