Output Artifacts

There are several types of artifacts produced by LogML:

  • Report is the primary one. This is a set of HTML files produced from Jupyter Notebooks in a batch mode. It includes visual results - tables, charts, etc.

  • Data artifacts are number results of analysis. They are usually in form of CSV files and are specific to a particular analysis.

There are also diagnostic or low level artifacts, which can be also used, but require pretty high level of proficiency with python and LogML:

  • Log files - usually required in case as any problems as a first tool of diagnostics.

  • Jupyter Notebooks - files with .ipynb extension, which can be executed in Jupyter Notebook service launched in the same conda environment as LogML version which produced those notebooks.

  • Serialized models - pickled model object. Can be used for diagnostic or debugging.

In the image below we present an example of how output folder of the LogML run is structured:

../../_images/report_files.png

Output Package

As a last step of pipeline LogML executes special action called release-artifacts. In carefully filters all the files, leaving only required data, and packs it into a zip file, named by the name of logml run.

In the image above it is named ‘gbsg2-saml4.zip’.

Report

Report is an ultimate output provided by LogML. General structure is represented below:

../../_images/report_overview.png