Introduction
At Brillersys, we’ve deployed several innovative applications on the Snowflake Marketplace. You can explore our full range of offerings here. In this post, we’ll focus on our Time Series Forecaster App, available in two versions:
- Free Trial: Featuring a robust yet basic forecasting model.
- Full Version: Equipped with advanced models, extensive customizations, and a wealth of rich features.
Even with its basic design, the trial version has demonstrated impressive performance, outperforming the Snowflake Cortex forecasting model. We’ll dive into details of our Time Series Forecaster App, how it compares to Snowflake Cortex, and highlight the advantages of the full version.
Time Series Forecaster: Trial Version Features
Our Trial version of the Time Series Forecaster App offers a suite of powerful features to get you started:
- Time Series Analysis: Analyse time series data to uncover patterns, trends, and seasonality. Visualize each time series and address skewness with built-in correction options.
- Anomaly Detection and Correction: Identify anomalies in your data and choose to correct them if needed, ensuring cleaner and more reliable forecasting.
- Time Series Forecasting: Perform forecasting using a fundamental model based on XGBoost and LightGBM.
Comparison with Snowflake Cortex
Snowflake Cortex is a comprehensive suite of AI and ML features, with Time Series Forecasting being its pioneering category. To evaluate our app, we compared it against Snowflake Cortex using a range of publicly available datasets, including the M4 dataset at the item level, weather data, the M5 dataset at the item level, and air passenger data with varying granularity.
Our trial version of the Time Series Forecaster App consistently outperformed Snowflake Cortex in nearly all experiments. Additionally, our app demonstrated significantly shorter training times compared to Snowflake Cortex. We used Mean Absolute Percentage Error (MAPE) as our performance metric, and the table below illustrates the comparison of MAPE values between our forecaster app and Snowflake Cortex:
Features in the Full Version
The full version of our Time Series Forecaster App includes a range of advanced features designed to provide deeper insights and greater flexibility:
- Granularity Flexibility: Input data at any granularity, including the most detailed levels, and obtain forecasts at various granularity levels—hourly, daily, or monthly—based on user specifications.
- Custom Date Ranges: Select different date ranges for training and experimentation from the provided data. Unlike the trial version, which uses the entire date range available, the full version allows for more targeted analysis.
- Incorporation of Exogenous Variables: Enhance forecasting accuracy by adding external variables that might impact your time series data.
- Time Series Segmentation: Segment or cluster multiple time series based on their characteristics, enabling separate modelling for each segment to improve forecasting precision.
- Diverse Model Selection: Utilize a broad spectrum of models ranging from statistical methods to machine learning and deep learning. The app features smart automatic model selection tailored to your data.
- Custom User-Specific Enhancements: Access additional customizations tailored to meet specific user needs and preferences.
Conclusion
In summary, our Time Series Forecaster App offers a powerful suite of features designed to meet diverse forecasting needs. The trial version provides a solid foundation with robust analysis, anomaly detection, and basic forecasting capabilities, consistently outperforming Snowflake Cortex in our evaluations.
Explore our app on the Snowflake Marketplace here. At Brillersys, our experienced data science team is ready to support your needs with our comprehensive data science services. For inquiries or further assistance, please reach out to us at info@brillersys.com.
Leave a Reply