Release 1.1.0

ML Launchpad version 1.1.0 is our first feature update!

It includes a few new features and improvements that we found useful when working with ML Launchpad. We hope you like it, too! :)

As always, to enjoy the goodness of this release, update with

$ pip install --upgrade mllaunchpad

If you already pinned an earlier version of ML Launchpad, update your requirements to

mllaunchpad~=1.1.0

(We recommend using ~=, the version specifier for compatible semantically versioned releases, to receive fixes but not potentially breaking changes).

This version includes the following updates:

1.1.0 (2021-12-13)

• FEATURE Add training metadata reporting and querying, issue #142, by Andreas Schuderer.
• FEATURE Add support for typed CSVs (option dtypes_path of FileDataSource and FileDataSink, issue #127, by Elisa Partodikromo
• FEATURE Add Spark support (experimental), see examples/spark_datasource.py and issue #145, by Andreas Schuderer.
• ENHANCEMENT FileDataSink (csv, euro_csv, raw DataSink types) now attempts to create missing paths if required, issue #148, by Andreas Schuderer.
• FIXED Keep generated RAML files free of command line messages, issue #126, by Andreas Schuderer.
• FIXED Change default text file encoding to UTF-8 for config files, text_file Data Sources/Sinks and JSON model metadata.
  NOTE: If you have been using non-ASCII characters in any of the above, you will need to check that the encoding of the relevant existing file(s) is UTF-8.
  Issue #128, by Andreas Schuderer.
• FIXED Explicitly support repeated query parameters (and array types like number[]) through RAML definition.
  Remove IP from training metadata due to problem on macOS Big Sur.
  Fix doc build by pinning Sphinx dependency docutils to version 0.17.1.
  Issue #147, by Andreas Schuderer.

Release 1.0.0

We are happy to release ML Launchpad version 1.0.0!

This being the first major release, we had this version in the works for a long time -- thoroughly tested, tweaked, and tested again for the better part of a year. We hope you like it! :)

As always, to enjoy the goodness of this release, update with

$ pip install --upgrade mllaunchpad

If you already pinned an earlier version of ML Launchpad, update your requirements to

mllaunchpad~=1.0.0

(We recommend using ~=, the version specifier for compatible semantically versioned releases, to receive fixes but not potentially breaking changes).

This version includes the following updates:

1.0.0 (2020-06-08)

• FIXED mllaunchpad --verbose now correctly logs DEBUG information, issue #119, by Andreas Schuderer.
• FIXED Fixed an issue where builtin DataSources could not be found when configured, issue #118, by Andreas Schuderer.
• FIXED Readthedocs now shows the up-to-date API docs, issue #110, by Andreas Schuderer.
• ENHANCEMENT Added chunksize parameter for piecemeal data reading to builtin DataSources, issue #120, by Andreas Schuderer.
• FEATURE Added functionality to better support unit testing in model development (added optional parameters to mllaunchpad.train_model(), mllaunchpad.retest() and mllaunchpad.predict(), added mllaunchpad.get_validated_config_str()), issue #116, by Andreas Schuderer.
• FEATURE Added generic SqlDataSource for RedShift, Postgres, MySQL, SQLite, Oracle, Microsoft SQL (ODBC), and their dialects, issue #121, by Andreas Schuderer.

The following changes have already been published in the previous pre-release (version 1.0.0rc1). They are also included in this final version of 1.0.0:

• ENHANCEMENT New command line interface (usage changes only slightly, see issue), issue #77, by Andreas Schuderer.
• ENHANCEMENT DataSource caching overhaul: data cached separately for different params, configurable cache_size, issue #97, by Andreas Schuderer.
• REMOVED Removed ‘api:version:’ (deprecated since 0.1.0) from configuration (‘model:version:’ is now the only location to specify both the model and the API version), issue #66, by Andreas Schuderer.

Release 1.0.0rc1

-- Note: this is a pre-release. To pip-install, specify either --pre or mllaunchpad==1.0.0rc1 --

Release 1.0.0 introduces some changes that might affect the way you work with ML Launchpad. Use this release candidate to test them (or use them - if you're adventurous ;) ):

• ML Launchpad will leave Beta status and go into Production status. This does not change anything for you, and ML Launchpad will be free and open source forever. If your company needs support, this will be handled through our OpenCollective. Please contact us there if you require support hours.
• The "version" key in the configuration file's "api" section, which has been deprecated since version 0.1.0, will not be accepted any more from version 1.0.0 on, and an error will be raised if you add it to your configuration (for the next few minor versions at least). The model's "version" key is used as the one, single version for your model-api from now on. The reason is that, in pretty much any conceivable scenarios, the two versions would change at the same time. The corresponding issue on GitHub goes into details.
• The command line interface will change slightly. The hyphen-prefixes ("--") of the commands "--train", "--predict", "--api" and "--generate-raml" will be removed (as will be the short versions "-t" and so on). Use just "mllaunchpad ... train" etc instead. You can still abbreviate commands using just their first letter(s). The options (e.g. "--config) will stay the same. More details in the GitHub issue.
• Caching (as configured by the "expires" key in the configuration) will change for OracleDataSource. Right now, the result of any query will be cached according to the "expires" setting, no matter what. Even SQL queries using varying parameters will just be cached if you configure "expires" to -1 of anything larger than 0, yielding the resulting data of the first query even if your new query's parameters (and result) would be different. While this is 99.9% a bug that should (and will) be fixed (and replaced by memoization), some existing implementations might implicitly rely on it, so we're mentioning it here just for good measure.

Release 0.1.2

• FIXED Oracle DataSource's get_dataframe now interprets Null as nan, issue #86, by Bob Platte.
• ENHANCEMENT Add a truckload of unit tests, issue #46, by Andreas Schuderer.

Release 0.1.1

Fix PyPI Classifiers

Release 0.1.0

• DEPRECATED ‘api:version:’ to be removed from configuration (‘model:version:’ will be the only location to specify both the model and the API version), issue #66
• ENHANCEMENT Add Impala DataSource (in examples), issue #4
• ENHANCEMENT Document DataSources and DataSinks, issue #88
• ENHANCEMENT Document configuration, issue #67
• ENHANCEMENT Add utility user function order_columns for enforcing equal data column order between data sources and API parameters, issue #37
• FIXED Fix misleading error message at WSGI entry point if model could not be loaded, issue #61
• ENHANCEMENT Config file is now being checked for omitted required keys, PR #65
• FIXED Use correct reference to werkzeug’s FileStorage, issue #63

v0.0.5

Bump version: 0.0.4 → 0.0.5

v0.0.2

Bump version: 0.0.1 → 0.0.2