Scaling Code Pattern Inference with Interactive What-If Analysis
SURF is an approach for active learning for code pattern inference.
It reduces users’ labelling effort, incorporates direct feature-level feedback, and actively guides users in assessing the implication of having a particular feature choice in the constructed pattern. The key insight behind SURF is that users can effectively select appropriate features with the aid of impact/what-if analysis.
SURF offers the following features:
-
SURF summarizes the population of usages into a skeleton, useful for understanding a large population of how an API is used. All features (e.g., Cipher.getInstance()) are overlaid with one another under a single code skeleton view. The features are structurally grouped, with each group distinguished from another by a different background color.
-
SURF provides hints on the global distribution, including how each feature is consistent with already labelled positive and negative instances. Each feature has support (S) - how many instances include the feature and information gain (I) — how well the feature separates already labelled positive and negative instances. These scores guide a user to grasp the distribution of individual features in the entire population
-
Impact Analysis. Clicking on each feature shows how the feature is distributed among already labelled positive and negative instances, and would match additional instances in the population.
-
What-If Analysis. Users can contrast the impact of two feature choices to understand how the pattern can include (or exclude) more instances in the rest of the population.
The left side of the interface is where the inferred pattern will be displayed.
The right side shows the candidate features.
From the command line in the meteor_app directory, run either one of the following three commands:
sh run_cipher.sh
sh run_digest.sh
sh run_random.sh
Follow the tutorial in study/surf_tutorial.pdf,
Use the docker image codesurf/surf
Run docker run -p 3000:3000 -it codesurf/surf /bin/bash
From code/meteor_app, run sh run_digest.sh
Go to localhost:3000.
The left side of the interface is where the inferred pattern will be displayed. The right side shows the candidate features. Initially, both are empty.
Click on "predefined examples".
Scroll down and click on "Infer pattern". Now, a pattern will be inferred and be shown on the left side of the interface.
Both the inferred pattern and the candidate features are structurally grouped (with colors distinguishing the different groups). For example, code related to error and exception-handling are purple. Code related to constructors are in green. Each feature's support and information gain is displayed beside it.
To perform an impact analysis, click on a feature to expand it
.
From the impact analysis, you can inspect which instances will be included and excluded from the inclusion of a feature.
To perform a what-if analysis, check the checkboxes of a pair of features. Then, click on "Why?"
From the what-if analysis, you can contrast the impact of a pair of features.
Finally, select the feature that you believe should be included, and click "Reinfer pattern".
For the Cipher task, go to taskA.
For the task on generating keys, go to taskB.