Metrics

Metrics

Method

Decription

confusion_matrix

Visualize classifier performance  via  a contingency table visualization

plot

Visualize classifier performance via ROC/PR values over a spread of probability threasholds

pstat

Report the execution summary of a given snippet of code using the cProfile run method

from intel_ai_safety.explainer import metrics

Several base metrics are provided for ML/DL classification models. These metrics cover model execution and performance and orient the data scientist to where there is potential for classification bias.

Algorithms

Provided with a classfication model’s predictions and their corresponding ground truths, staple performance metrics can be calculated to determine prediction behaviors in the real world. These functions leverage scikit-learn and plotly (eventually) to calculate and visualize said metrics, respectively.

Environment

  • Jupyter Notebooks

Metrics

  • Performance metrics

    • Confusion Matrix

    • Performance Plots

  • Execution metrics

    • Python profiler

Toolkits

  • Scikit-learn

  • Plotly

  • Python Profilers

References

Scikit-learn
Plotly
Python Profiler

API Refrence