Sage Studio
A Model Research and Development Tool

Product Value

Low Threshold

Visualized modelling method; simply drag and drop to perform modeling.

Multi-dimensions

Compatible with multiple machine learning computing frameworks, and provides different model methods for different professional scenarios.

High Performance

Supported by various types of self-developed high-performance algorithms and techniques, with the computational efficiency hundreds or even thousands of times higher than Spark.

End-to-end

It covers the entire process from data to model, application, and to the deployment and launching of the application.

Expandable

An integrated Notebook service that supports interactive model research and algorithm package expansion.

Product Functions

  • Model Research IDE
    Diversified Development Methods
    Model Development Based on Visualized Computational Graphs (DAG)

    Modeling through programming languages such as Python is difficult and unintuitive, and therefore can be difficult for business personnel to perform the job. The visual computing diagram (DAG) displays the entire process of model training in a visual and easy-to-understand way. Without the cumbersome code debugging, users can perform modeling with simple drag-and-drop, and focus more on their businesses.

    Interactive Model Development Based on Notebook

    Based on Notebook, users can complete a one-stop and interactive process of coding, running, data visualization and result feedback. It includes data cleansing, statistical analysis, data visualization, machine learning model building and other development processes.

    Coverage of Rich and Diversified Algorithms
    Self-developed Ultra-high-dimensional AI Algorithm

    The modeling IDE integrates multiple self-developed ultra-high-dimensional AI algorithms. Reaching up to trillions of dimensions, the model achieves a breakthrough as most existing machine learning algorithms in the market perform just moderately when processing large data volumes. This has fully maximized the potential value of big data and made use of the advantage of the scale of the data. The algorithms include logic regression that is 100 times more efficient, GBDT with strong interpretability, self-developed large-scale discrete sparse neural network DSN, He-Treenet which has assisted a top bank to receive a scientific and technological innovation award, and linear classifier.

    Deep Learning Capacity

    Self-developed high-dimensional sparse deep neural network DSN; integration of Tensorflow, Pytorch and other deep learning frameworks, with distributed high-performance computing capabilities.

    Providing Value throughout the Entire Process
    Low Threshold, High Performance, Unified Online and Offline Feature Engineering

    Users are required to have rich modeling experience, strong programming capabilities and a deep understanding of the business to perform feature engineering through open source tools. Studio's feature engineering engine is equipped with many years of experience of modeling experts, and has extracted the abstract and powerful feature engineering method. Users can call upon a method with simple and low-threshold function expressions. It is easy to learn, and has timing feature processing capabilities and strong computing power, as well as the consistent and simple running state deployment online and offline. The running code for online deployment is easy to develop, which greatly shortens the development cycle and improves development efficiency.

    Multi-dimensional and Comprehensive Model Evaluation Indicators

    Model effect evaluation is inevitable in the process of model development. The application development IDE provides large variety of components, and displays a comprehensive set of of evaluation indicators such as AUC/ROC, accuracy, and recall rate, with clear and visual graphs, greatly enhancing the efficiency of model effect evaluation.

  • Application Development IDE
    Self-learning Applications

    The self-learning process of the model is one of the difficult parts in AI applications. 4Paradigm’s ‘Machine Learning Cycle’ theory has opened up the closed-loop system of AI applications with the coverage of the entire process of AI application construction. Users can quickly build self-learning applications through simple graph and form configuration, greatly reducing the amount of codes in the development process, significantly shortening the development cycle of self-learning applications, synchronizing model self-iteration with changes in business, and lowering the risk of model effect attenuation.

    Batch Forecasting Applications

    With the support of self-learning applications, Studio provides a convenient development method for batch forecasting applications for scenarios with low real-time requirements. Users can quickly build batch forecasting applications through simple graph and form configuration.

    Real-time Forecasting Applications

    With 4Paradigm’s self-developed high-performance in-memory time series database RTIDB, Studio is also suitable for scenarios with high real-time requirement such as real-time high-dimensional machine learning and real-time deep learning. Users can quickly develop real-time forecasting applications through simple form configuration, so as to significantly reduce the chance of redoing development work during the model’s “research to production” transformation process.