Luminaire Versions Save

Add a random_state argument to the HyperparameterOptimization class to ensure the reproducibility of results.

MR's involved: https://github.com/zillow/luminaire/pull/125 https://github.com/zillow/luminaire/pull/131

Full Changelog: https://github.com/zillow/luminaire/compare/v0.4.2...v0.4.3

Authored by: @papaemman Review changes authored by: @sayanchk

Fix: lad_filtering applying appropriate diff order instead of hardcoded value

Changes:

• Leveraging https://pypi.org/project/bayescd/ instead of explicitly installing that same repos code for ci/cd since its now available on PyPI
• Adding bayescd to dependency so users don't have to install them manually
• Support for weekly frequency for data exploration

Closes issues:

• https://github.com/zillow/luminaire/issues/112
• https://github.com/zillow/luminaire/issues/115
• https://github.com/zillow/luminaire/pull/118

• Support up to Python 3.10
• Support of latest versions of scipy, statsmodels and bayesian-changepoin-detection
• Minor bug fixes and improvements in data exploration
• Minor bug fixes and improvements in structural model
• Ability to perform model validation due to under-fit added in structural model
• Holiday list updated

Note: We had to remove bayesian-changepoint-detection package from requirements due to deployment issues in pypi (the latest version of scipy is not supported by bayesian-changepoint-detection 0.2.dev1 available in PyPI). If you are planning to use this luminaire v0.4.0, you have to manually install a compatible version of bayesian-changepoint-detection from github provided by the community but not yet made available on PyPI using the following script:

pip install git+https://github.com/hildensia/bayesian_changepoint_detection@2dd95f5c1d028116899a842ccb3baa173f9d5be9#egg=bayesian-changepoint-detection

Luminaire cd fixes from dev2

Release notes from dev1:

• Support up to Python 3.10
• Support of latest versions of Scipy, Statsmodels and bayesian-changepoin-detection
• Minor bug fixes and improvements in data exploration
• Minor bug fixes and improvements in structural model
• Ability to perform model validation due to underfit added in structural model
• Holiday list updated

Please read: We had to remove bayesian-changepoint-detection package from requirements due to deployment issues in pypi (the latest version of scipy is not supported by bayesian-changepoint-detection 0.2.dev1). If you are planning to use this dev release of luminaire, you have to manually install a compatible version of bayesian-changepoint-detection from github using the following script:

pip install git+https://github.com/hildensia/bayesian_changepoint_detection@2dd95f5c1d028116899a842ccb3baa173f9d5be9#egg=bayesian-changepoint-detection

Luminaire cd fixes from dev1

Release notes from dev1:

• Support up to Python 3.10
• Support of latest versions of Scipy, Statsmodels and bayesian-changepoin-detection
• Minor bug fixes and improvements in data exploration
• Minor bug fixes and improvements in structural model
• Ability to perform model validation due to underfit added in structural model
• Holiday list updated

• Support up to Python 3.10
• Support of latest versions of Scipy, Statsmodels and bayesian-changepoin-detection
• Minor bug fixes and improvements in data exploration
• Minor bug fixes and improvements in structural model
• Ability to perform model validation due to underfit added in structural model
• Holiday list updated

Update the requirements files list:

• The existing requirements file specifies hard version requirements which is not helpful to the user
• We ran the existing test cases to see what all recent version of dependent packages can be supported and based on that update the requirements files list

Support for python 3.7 for build and deploy

• Major dependent packages are kept the same
• Just making the code compatible with Python 3.7 as Python 3.6 is reaching End of Life

Note: Major package upgrade is planned for Q1-2022.

Bug fixes:

• Data aggregation fixed in the presence of duplicate index.