Robustness Tests

Robustness Toobox generated a perturbed dataset using boundary attack algorithm on the test dataset. To execute this test against your model, you will need to upload data and upload model onto the portal.

Run using Command Line (CLI) interface

Install the test using pip:

pip install aiverify-robustiness-toolbox

Run the bash script to execute the plugin:

#!/bin/bash

root_path="<PATH_TO_FOLDER>/aiverify/stock-plugins/user_defined_files"
python -m aiverify_robustness_toolbox \
  --data_path $root_path/data/sample_bc_credit_data.sav \
  --model_path $root_path/model/sample_bc_credit_sklearn_linear.LogisticRegression.sav \
  --ground_truth_path $root_path/data/sample_bc_credit_data.sav\
  --ground_truth default \
  --model_type CLASSIFICATION

Refer to Python notebook that walks through the steps with sample data.

Once the algorithm runs successfully, the results will be saved in an output folder. Use the generated result to create a report.

Generate report

To generate report using the test results, select

HomePage > Create New Project > Create New Report Template.

Select "Robustness Toolbox" on the left pane and pick the 3 widgets and load them onto the canvas and click "Next".

Choose the model and test results from previous run and click "Next".

Save the report or export as PDF.

Info

Report generation is optimised for Chrome browser. Users are recommended to use Chrome browser for best experience compared to other browsers.