trellixvulnteam / modelgarden_ekaw

Characterizing Deprecated Deep Learning Python APIs: An Empirical Study on TensorFlow

Nian Liu

This respository consists of experiment data, script, result for this study. Below is the introduction of files in each folder.

Experiment Data

• models-r1.10.0-models-r2.3.0: Deep Learning projects used in our study (Models). Models range from 1.10.0 to 2.3.0, there is no release for TensorFlow 1.14 and 1.15

RQ1

• ChangeDetect.py: script used to find deprecated APIs, removed APIs, newly deprecated APIs, etc. in TensorFlow 1.0-2.3.
• output/tensorflow-r1.0.csv-tensorflow-r2.3.csv: deprecated APIs in TensorFlow 1.0-2.3.
• output/removed_apis.csv: removed APIs in TensorFlow.
• output/age.csv: data of the number of versions deprecated APIs get deprecated since introduced into TensorFlow.
• output/api_survival_time.csv: data of the survival time of deprecated APIs get removed from TensorFlow.
• output/diff.csv: data of the newly deprecated APIs in each TensorFlow Version.

RQ2

• extract_deprecation_message.py: script used to extract deprecation message of deprecated APIs.
• Reason for Method Deprecation.md: deprecated APIs for each method deprecation reason in RQ2.
• Reason for Parameter Deprecation.md: deprecated APIs for each parameter deprecation reason in RQ2.

RQ3

• Compare.py: script used to find deprecated APIs in Models projects.
• output/models-r1.10.0(official).csv-models-r2.3.0(official).csv: data of deprecated APIs usage in each Model verison.
• output/models_all(official).csv: data of deprecated APIs usage in Models from version 1.10.0 to 2.3.0 without false positive.

RQ4

• p-value.py: script used to calculate p-value and cohen'd value.
• commit_history.csv: commit history of each commit that updated deprecated APIs.
• log: original log which contains model training and testing result for the 13 tests in RQ4.

Other files

• utils: other files.