Skip to content

Changelog

2.6.0 - Better control of Session

  • A single Session object is now managed by the AzmlClient instance. This object is created by default, and possibly configured with the information from the ClientConfig. It is automatically closed when the client instance is garbaged out. A custom Session can be passed instead of the automatically-created one. In that case it is not automatically configured from the ClientConfig, but it is easy to do so using <config>.configure_session(session). Fixes #15 and #16.

  • create_session_for_proxy_from_strings is deleted and create_session_for_proxy is deprecated. Users should now use set_http_proxy(session, ...) as the preferred way to configure a new requests.Session() object.

  • Fixed a bug (Series has no items() attribute) on some versions of pandas.

  • Migration to Github Actions.

2.5.0 - New replace_NaN_with and replace_NaT_with options

  • New replace_NaN_with and replace_NaT_with options to control how NaN and NaT get converted. Fixes #11
  • packaging improvements: set the "universal wheel" flag to 1, and cleaned up the setup.py. In particular removed dependency to six for setup and added py.typed file. Added pyproject.toml too. Fixes #13

2.4.0 - new helper method + doc

New method create_response_body in RequestResponseClient. Updated doc to show how these goodies can be used.

2.3.2 - added __version__ attribute

Added __version__ at package level.

2.3.1 - Config files can now be templates

ClientConfig.load_config() now accepts keyword arguments that will be used as replacements for variables in the config files. Fixes #10.

New dependency: jinja2

2.2.0 - Request-Response: new default behaviour concerning outputs, and new parameter.

Low-level API:

  • RequestResponseClient.read_response_json_body now does not filter outputs anymore even if a non-None output_names is provided.

  • New parameter to execute_rr: only_keep_selected_output_names. If this is True the legacy behaviour of filtering outputs by names is preserved. Otherwise all outputs are preserved. Fixes #9

High-level API:

  • AzureMLClient.call_azureml: ws_output_names now appears as optional in the type hints. Improved docstring comments slightly.

2.1.0 - High-level clients: swagger, remote-only services + Bugfix

New:

  • Swagger format is now supported in both high-level (call modes) and low-level (RR and Batch clients) API. Fixes #5.

  • New argument remote_only to disable local usage of a service. Fixes #8.

Misc:

  • Fixed bug with decoding AzureML errors. Fixes #7.

  • call_local_service: renamed argument service_name to function_name to distinguish better between the service (azureml) and the function (local implementation's method).

2.0.0 - New tools for high level api creation

New features:

  • AzureMLClient helper class to create pythonic high-level APIs, supporting both local and remote call modes, and configurable from yaml and ini/cfg files.

Refactoring:

  • azure-storage dependency is now optional (only for batch mode)

  • improved documentation and type hints, and changed some method names and contents to be more pythonic. This was indeed a quite old project :).

  • removed use_new_ws from all methods - this concept is not meaningful anymore for AzureML usage.

  • renamed RR_Client into RequestResponseClient and Batch_Client into BatchClient. Old names will stay around for a version or two for compatibility reasons.

  • got rid of the useless Converters container classes for data binding.

1.2.0 - Support for "swagger" mode

  • Added swagger=True mode support for azureML calls

1.1.0 - Python 2 support, and a few bugfixes

  • Updated this old project (my first public python package :) ) to more recent continuous integration and tests standards.
  • Support for python 2
  • A few bugfixes
  • Added all correct dependencies in setup.py

1.0.1 - First version

Request-response and Batch modes support, python 3 only.