Download the Brochure Read the Handbook Get Started In MATLAB
CatFish simulates and forecasts count outcomes using a Bayesian categorical variable multiplicative Poisson model. It combines observed counts with contextual information such as expert knowledge, external forecasts, and independent covariates to infer full joint distributions with interpretable factors. Its simulations and forecasts aggregate and disaggregate consistently across levels of detail. One application is generating time and location-specific, product-level demand forecasts for supply chain optimization.
CatFish is not a black box artificial neural network (ANN). It provides identified values for potentially millions of categorical variables. Leverage spatial and temporal data on products, weather, demographics, and more to understand what drives outcomes accross locations and over time.
Get the full joint distribution, not point estimates. Distributions are key to sophisticated business decision making, including supply-chain management and risk analysis. CatFish’s mixed Poisson and negative binomial meet most use cases.
CatFish supports any number and type of dimensions: Compare results across locations, time, and other dimensions, with forecasts that aggregate and disaggregate consistently.
CatFish is a cloud-scalable fully Bayesian model: Use expert beliefs to refine predictions or conduct what-if analysis when data is limited. Use hierarchical priors to learn distributional parameters. Use demand and contextual data to build up to petabyte-scale models when information is available.
Unlike other forecasting models, CatFish is primarily cross-sectional. However, its time dimensions support random walks with drift, extrapolated trends, and other time series functionality. Trends can also be imported for any time-related variable from outside models.†
CatFish is competitive with state-of-the-art 1D and 2D statistical and ANN-based time-series models in the time dimension. And it can use their trends as a starting point.† But every meaningful dimension and factor added to a CatFish model increases its capabilities and accuracy.