What is ZipCast
ZipCast puts forecasting on steroids. It enables forecasting at scale and complexity at a base unit level (e.g. store-SKU), that can be used for demand planning and inventory optimization.
Advantages of ZipCast on IBM Cloud Pak for Data
1. Automates model building and forecasting with high levels of accuracy.
2. Offers a range of prebuilt models that take into account industry best practices.
3. Gives data scientists teams a head-start and significantly reduces the build time.
4. No prior knowledge of machine learning required
Key Features of ZipCast
- Event Fuzzification
- Select External Regressors
- Auto discover significant regressors
- GA for estimating lags
- 24 techniques across Regression, Time
series and ML
- Linear regression
- Neural nets
- Multiple performance metrics available
- Most accurate model chosen using holdout period validation
ZipCast in Action
ZipCast on IBM Cloud Pak for Data
- Zipcast Demand Forecasting Module is built on top of IBM Cloud Pak for Data.
- Leveraging capabilities like Virtualization to create virtual assets within Cloud Pak for Data without moving Product and Sales Data from the source RDBMS systems.
- Data refinery was used for data wrangling and light weight transformation ability, and visual analysis to detect outlier.
- The cleansed data fed into Zipcast Module runs on the Watson Studio R- IDE instance. Easily customizable run-time to increase vCPU and RAM on demand for an increased workload.
- The output of the model can be visually analysed using built-in Cognos Dashboard or fed into third party systems for further integration and consumption.
- IBM Cloud Pak for data allows the use cases to expand beyond the current module by easily and quickly setting up more models.