Replication Package for the Paper "How are MLOps Frameworks Used in Open Source Projects? An Empirical Characterization"

Replication Package for the Paper "How are MLOps Frameworks Used in Open Source Projects? An Empirical Characterization" RQ1 & RQ2/ Results_Overview.csv: for each project: - Repo: dependent repository - Framework: framework being used - UseinCI: whether the framework is used in CI/CD - Mappings: phases being used (comma-separated) Mapping_Phases_On_Repos.csv: Similar to Results_Overview.csv, but in this case the mapping is related to the specific ML steps RQ1 & RQ2/Mappings/ For each framework, the file contains the mapping between invoked APIs and MLOps pipeline phases RQ1 & RQ2/Scripts/ Scripts to address RQ1 and RQ2: download_dependents.py: to download dependents and apply filters over them usage_analysis.py: given a framework and its list of dependents, produces the output stored in the Usages/ folder RQ1 & RQ2/Projects/ For each framework, the list of dependents RQ1 & RQ2/Usages/ For each framework, the invoked APIs in each file RQ3/ Taxonomy - Consolidation.csv: Taxonomy merger task performed to consolidate the taxonomy Taxonomy with Descriptions and Issues.csv: For each category (before consolidation), description and examples Human-coding-output.csv: Results of the first-level manual coding on the sample Human-LLM-conflict-resolution.csv: Results of the second-level coding human/LLM with conflict resolution Complete dataset LLM_classification.zip: Raw output of the LLM classification on the whole dataset Complete dataset LLM_classification.xlsx: The output above in a spreadsheet Prompt_template.txt: Prompt used by the LLM to classify feature requests MLOps Tools Raw Dataset.xlsx: All the downloaded issues RQ3/Scripts/ async_openrouter_pipeline.py: script for the automated, LLM-based classification RefinedCodebookJSON.json: Category Codebook used to drive the LLM-based classification