YData SDK for improved data quality everywhere!
ydata-sdk v0.1.0 is here! Create a YData account so you can start using today!
The YData SDK is an ecosystem of methods that allows users to, through a python interface, adopt a Data-Centric approach towards the AI development. The solution includes a set of integrated components for data ingestion, standardized data quality evaluation and data improvement, such as synthetic data generation, allowing an iterative improvement of the datasets used in high-impact business applications.
Synthetic data can be used as Machine Learning performance enhancer, to augment or mitigate the presence of bias in real data. Furthermore, it can be used as a Privacy Enhancing Technology, to enable data-sharing initiatives or even to fuel testing environments.
Under the YData-SDK hood, you can find a set of algorithms and metrics based on statistics and deep learning based techniques, that will help you to accelerate your data preparation.
YData SDK is currently composed by the following main modules:
- YData’s SDK includes several connectors for easy integration with existing data sources. It supports several storage types, like filesystems and RDBMS. Check the list of connectors.
- SDK’s Datasources run on top of Dask, which allows it to deal with not only small workloads but also larger volumes of data.
- Simplified interface to train a generative model and learn in a data-driven manner the behavior, the patterns and original data distribution. Optimize your model for privacy or utility use-cases.
- From a trained synthesizer, you can generate synthetic samples as needed and parametrise the number of records needed.
Synthetic data quality report Coming soon
- An extensive synthetic data quality report that measures 3 dimensions: privacy, utility and fidelity of the generated data. The report can be downloaded in PDF format for ease of sharing and compliance purposes or as a JSON to enable the integration in data flows.
Profiling Coming soon
- A set of metrics and algorithms summarizes datasets quality in three main dimensions: warnings, univariate analysis and a multivariate perspective.
Supported data formats
The RegularSynthesizer is perfect to synthesize high-dimensional data, that is time-indepentent with high quality results.
The TimeSeriesSynthesizer is perfect to synthesize both regularly and not evenly spaced time-series, from smart-sensors to stock.
The TimeSeriesSynthesizer supports transactional data, known to have highly irregular time intervals between records and directional relations between entities.
The MultiTableSynthesizer is perfect to learn how to replicate the data within a relational database schema.