Tutorials

• Debugging scikit-learn text classification pipeline
  • 1. Baseline model
  • 2. Baseline model, improved data
  • 3. Pipeline improvements
  • 4. Char-based pipeline
  • 5. Debugging HashingVectorizer
• TextExplainer: debugging black-box text classifiers
  • Example problem: LSA+SVM for 20 Newsgroups dataset
  • TextExplainer
  • Why it works
  • Should we trust the explanation?
  • Let’s make it fail
  • Let’s make it fail, again
  • Customizing TextExplainer: sampling
  • Customizing TextExplainer: classifier
• Explaining XGBoost predictions on the Titanic dataset
  • 1. Training data
  • 2. Simple XGBoost classifier
  • 3. Explaining weights
  • 4. Explaining predictions
  • 5. Adding text features
• Named Entity Recognition using sklearn-crfsuite
  • 1. Training data
  • 2. Feature extraction
  • 3. Train a CRF model
  • 4. Inspect model weights
  • 5. Customization
  • 6. Formatting in console
• Visualize Token Probabilities and Model Confidence in LLM Predictions
  • 1. OpenAI models
  • 2. Limitations
  • 3. Open Source and other models
• Explaining Keras image classifier predictions with Grad-CAM
  • 1. Loading our model and data
  • 2. Explaining our model’s prediction
  • 3. Choosing the target class (target prediction)
  • 4. Choosing a hidden activation layer
  • 5. Under the hood - explain_prediction() and format_as_image()
  • 6. Extra arguments to format_as_image()
  • 7. Removing softmax
  • 8. Comparing explanations of different models