Model Performance Templates
Model performance templates help you track the health of models with actuals data.Regression Model
Concisely overview aspects that impact regression model performance Key
Widgets: Count statistics, error measures (RMSE, MAE, MSE, MAPE), average
predictions, daily trends, and feature distributions, prediction vs. actual
comparison charts
Classification Model
Concisely overview aspects that impact classification model performance
Key Widgets: Statistics (prediction/actual counts, accuracy, score
averages), time series charts (class comparisons, daily values, score
distributions), and specialized charts for accuracy, F1 scores by class, and
feature distribution
Ranking Model
Concisely overview aspects that impact ranking model performance Key
Widgets: Rank performance metrics (NDCG, MRR, AUC, MAP) for overall model
performance and individual features, model performance based on rank groups,
and time-series fluctuations
Compare A/B
Compare two models in production using performance metrics Key Widgets:
Compare two models based on performance and data quality metrics over time,
implement canary vs live model comparison, and compare & track significant
production data fluctuations.
Production vs Training
Compare and visualize the difference between production performance to
training performance Key Widgets: Track slices and compare between
production and training environments. This template creates an aggregate
count tracker for key data and an overlaid view of metric history over time
Create A Performance Dashboard
Step 1: Select Performance Template
Step 2: Specify Fields
Step 3: Use dashboard, modify widgets, and share with your team!
Questions? Email us at support@arize.com or Slack us in the #arize-support channel