October 2020 Summaries

Gretel Blueprints are now available, simplifying the process of anonymizing and balancing datasets with just a few clicks. These collections of sample code and datasets utilize Gretel's SDKs and can be easily adapted to solve customer-specific use cases. Developers can access Gretel Blueprints via Gretel Cloud or GitHub, where they will also find new features and updates based on customer feedback. The company plans to release tutorials, walk-throughs, and customer testimonials for many of the Blueprints in the coming days.

Gretel's Premium SDK now includes detailed reporting that shows the accuracy of synthetic data's statistical distributions and correlations. The performance report provides interactive Plotly graphs and stylish HTML formatting, allowing users to assess how well their training data's distributions were maintained in the new synthetic data. Key metrics include duplicated lines between training and synthetic data, Mean Squared Error (MAE) for field correlations, and Jensen-Shannon Distance for within-field distribution maintenance. The report also includes a breakdown of JS Distance scores for each individual field, as well as a heatmap showing the differences in correlation values between the original and synthetic datasets. Users can improve their model's performance by retraining it multiple times, increasing the number of training examples or epochs, adjusting rnn_units parameter, or experimenting with dropout_rate.