The FullRank Machine Learning Technology differs from widely available open source and commercial options in the following ways.


Probabilistic Results

Most ML Frameworks give only an answer; for example, that an image is a cat or a dog. Others also provide probabilities - as in a 90% probability that an image is a cat. FullRank gives the full result: a probability distribution.




This tool allows one to interactively explore the data, different hypotheses and hypothesis ordering.

Hypo Inspector

The inspector allows one to view previously-created hypothesis ensembles, typically created by the AutoML (AGS) tool.

AUto ML Tool

The AutoML tool (AGS) automatically searches for the 'best' hypotheses, builds efficient ensemble trees, and saves them for later inspection and use.

The Full Explanation

For each result, the framework breaks down the results into constituents, each with particular reasoning, expected values, dispersion and strength. 



Most functions are available through an API. The key concepts are:

  • Client - Each API key is associated with a client.

  • Project - A Client can have multiple projects and list them. Projects contain DataSets, problems and hypotheses.

  • DataSet - A Project has typically many accessible data sources. The system currently supports spreadsheets, CSV files, RDBMS tables and Parquet files.

  • ProblemContext - These bind together data sets, loss or error metrics (RSME, for example), data set annotation, and other configurations. There can be many versions of these. ProblemContexts contain many Hypotheses.

  • Hypothesis - A hypothesis, given an input vector, X, returns an output Y. The output is in the form of a Result.

  • Result - Results are generated by a Hypothesis and store a probabilistic response and a explanation thereof.