2. FixOut Web Dashboard¶

2.1. Intro¶

Warning

FixOut is a tool centered on sensitive features and their proxies. Not providing this information correctly beforehand will compromise the obtained results.

2.2. How to activate it¶

If you want to activate the web interface that generates the dashboard with the fairness assessment of your model and other functionalities, you must execute the run function with the show parameter activated (from fixout.runner.FixOutRunner).

>>> fxo.run(fxa, show=True)

Once execution is complete, you will be able to see a web page like the following by accessing http://localhost:5000, if you use the default parameters.

2.3. Main functionalities¶

The dashboard displays FixOut’s functionalities organized in tabs and centered on sensitive features. The two first tabs are briefly described bellow: Data and Metrics.

Note

Other features related to explainability and model-specific inspection are not available in this version of FixOut.

2.3.1. Data¶

In this tab, FixOut displays the results obtained after running pipelines maily related to data.

Data distribution (see Data distribution section)
Correlation analysis (see Correlation analysis section)
Reverse engineering (see Reverse engineering section and fixout.helper.ReverseFairness)
Discriminatory models (see Discriminatory models section and fixout.helper.UnfairModel)

2.3.2. Metrics¶

In this tab, the plots are obtained after calculating the fairness metrics. See also Fairness assessment section and fixout.fairness.