Giskard

by Oct 10, 20220 comments

Genre : Business
Platform : Web
Developer : Giskard
Rating : 4.5
Positives :
Negatives :

Giskard is the first collaborative and open-source software platform for ensuring AI model quality. From your Python environment, upload your model. It is a simple upload for any Python model, including PyTorch, TensorFlow, Transformers, Scikit-learn, and others.

Key Features of Giskard

  • Collaboratively evaluate ML models.
  • Get immediate feedback on your ML models from business stakeholders.
  • A diagram of the exhaustive test suites
  • In seconds, you can put machine learning models to the test.
  • Extensive test suites supported by cutting-edge machine learning research
  • Install tests in your CI/CD pipeline.
  • Protect your machine-learning models from regressions, drift, and bias.
  • Generate synthetic data to remove bias

Gather Feedback on Your Ml Model

Select a model and a dataset to examine. You can inspect your model after uploading it by

  • Selecting the Models tab
  • You can select the model version you want to inspect by clicking the Inspect button.
  • Choosing the dataset to be used for the inspection
  • Picking the correct target variable from your dataset.
  • This is your target variable’s true value.
  • If your dataset contains no actual target variables, select the blank cell.

Experiment with the model.

  • Change the examples, and feature values, or generate local explanations to experiment with the model.

Automated tests should be used to eliminate regression and bias. You have faster collaboration with feedback from business stakeholders.

Important Benefits

  • Automated tests of ML models can be deployed quickly.
  • Obtain direct feedback from key stakeholders in the business.
  • Create more performant, robust, and ethical machine learning models.

Validate Your Machine Learning Model

Giskard allows you to create AI model test suites. It includes test presets to help you quickly design and execute your tests.

Filter and Organise the Feedback

You can filter feedback based on the following criteria to target it better:

  • The version of the model
  • Dataset
  • Feedback can be of three types: value perturbation, feature feedback, or general feedback.

Final Words

It is best suited toward a CI/CD platform for ML products. You have a good option of self-hosted &open-Source. They are also compatible with your preferred machine-learning tools.

Web App: Giskard