Scarecrow is LDC's proprietary technology for building intelligent time series models

The Lowin Data Company was founded to explore the application of novel statistical methods to time series. As a highly academic venture requiring massive experimentation, we quickly found that the typical data exploration pipeline was simply too slow and inflexible; not only were many algorithms unavailable or locked in incompatible libraries, but working with time series added complexity that some existing software packages simply couldn't handle.

We needed software that allowed rapid iteration of ideas and techniques. We wanted something that would let us focus on building algorithms, not the nuances of getting each one to run. Fortunately, we're data scientists -- so we built it ourselves.


If it only had a brain...

Our software, Scarecrow, is the result of nearly two years of development. It knows all about algorithms, data, and training methods, so our analysts simply specify the algorithms they want to work with and let Scarecrow build the appropriate data infrastructure. Moreover, Scarecrow has a vast and growing library of pre-built data processing techniques and training algorithms.

Today, Scarecrow powers all of LDC's research and models.


Scarecrow solves common data pipeline bottlenecks

  • Transparency: every algorithm can be decomposed into basic components.
  • Modularity: algorithms and their components or parameters can be swapped at any time.
  • Prototyping: algorithms go from whiteboard sketches to production models as fast as possible.
  • Universality: any algorithm can be modeled, from traditional machine learning to computational neuroscience.
  • Compatibility: all Scarecrow models are compatible with each other and can be combined into more complex architectures.