5.1.3

• Update snips-nlu and pychatl
• Some minor fixes in the documentation

Since snips-nlu is mostly discontinued, we are actually trying to write another interpreter implementation to keep this project going 😊

5.1.2

Minor enhancements and chore release (mostly documentation related stuff).

5.1.0

New features

• Added SettingsStore.write_to_file to write a settings store to a file #29
• Added SettingsStore.to_dict to flatten a SettingsStore into a dictionary ready to be used in agent metadata #32

Fixes

• Fix a bug where the interpreter cache directory may not exists and persisting failed as a result #34
• SettingsStore.getpath now resolve the path from the configuration file which has been read #33

5.0.0

And that's a big one 🍾 ! The 5.0.0 version represents a more robust and future proof approach.

Currently available skills will be updated soon.

Breaking changes

This version has several breaking changes documented here so if you use the library, you may have some modifications to do. If you still can't find how to migrate, feel free to open an issue.

Here is a quick summary of changes.

Code organization

Everything has been moved to a more domain centric approach, some core refactoring has made tests easier and isolated.

Utilities files have been splitted.

Command Line Utility

• Looks for a configuration file pytlas.ini in the current working directory by default
• Default skills directory is the current working directory (was /skills)

Settings handling

See https://pytlas.readthedocs.io/en/develop/migrating.html#access-to-settings

Testing

• pytlas.testing.mock.ModelMock.get_call defaut to the last call (was the first one)

New features

• SnipsInterpreter now automatically download needed resources at runtime
• SkillsManager does not allow to install a skill for an incompatible pytlas version (by looking at the skill requirements.txt), you should now use pytlas~=5.0.

Chore

• ⬆️ transitions 0.6.9 -> 0.7.0
• ⬆️ fuzzywuzzy 0.16.9 -> 0.17.0
• ⬆️ colorlog 3.1.4 -> 4.0.2
• ⬆️ pychatl 1.2.7 -> 2.0.1
• ⬆️ python-dateutil 2.7.3 -> 2.8.0
• ⬆️ Babel 2.6.0 -> 2.7.0
• ⬆️ Markdown 3.0.1 -> 3.1.1
• ⬆️ beautifusoup4 4.6.3 -> 4.8.0
• ⬆️ snips-nlu 0.19.7 -> 0.20.0

4.0.6

• ⬆️ Updated snips-nlu to 0.19.7
• ✨ Add settings.reset and settings.load

4.0.5

• ⬆️ Updated snips-nlu to 0.19.6
• ♻️ Added include_cancel_state to build_scopes because of snipsco/snips-nlu#788

4.0.4

• ⬆️ Updated snips-nlu to 0.19.5
• ⬆️ Updated pychatl to 1.2.7

4.0.3

• 🐛 Removed shell=True in pam which prevented cloning under Linux
• 🐛 Added sorting functions related to training data to ensure the same checksum will be generated (there is still an issue on my rasp I could not resolve related to this...) and prevent unneeded retraining

4.0.0

Major release since there are breaking changes on how SlotValue are represented.

• ✨ Added pytlas.testing namespace containing some utilities to easily unit test your skills, see the documentation for more informations
• 💥 SlotValue value now not always returns a str. The value property will now contains a more meaningful python representation of the slot (dateutil.relativedelta for durations, pytlas.interpreters.slot.UnitValue for AmountOfMoney and Temperature, float between 0 and 1 for percentages, datetime.datetime for InstantTime, tuple of datetime.datetime for date range and str for anything else). This make the developer stuff easier when retrieving slot with request.intent.slot('my_slot').first().value
• ♻️ SkillData is now called Meta

3.2.2

• ⬆️ Updated snips-nlu to 0.19.4
• ✨ Added agent.id property which contains a unique identifier for the agent instance