Batched boundary cut generator

An apparatus for calculation of an estimation solution within a constraint polytope, the apparatus comprising:

• a database comprising data indicative of a constraint polytope;
• a communication interface configured to receive a point batch from an external computer system, the point batch including a plurality of data points representing transactions;
• and
• a polytope controller implemented by a processor coupled with the database and the communications interface and configured to analyze the plurality of data points for a projection attempt of the plurality of data points to the constraint polytope and generate a dynamic adjustment for use with at least one subsequent point batch for the constraint polytope in response to the projection attempt;
• and
• the polytope controller further configured to use the constraint polytope, sample during batch processing, use the sample to generate a linear approximation of a non-linear constraint and combine the linear approximation with already existing linear constraints;
• wherein some points in each batch are discarded when outside of a boundary and as more and more batches are processed, the processor learns about the boundary thereby decreasing the discard rate over time, without the need to recalculate the internal states of the linear constraints of the constrain polytope.

A non-transitory computer readable medium including instructions that when executed by a processor are configured to cause the processor to perform:

• initializing, at a polytope controller, a constraint polytope;
• receiving, at the polytope controller, a plurality of point batches from an external computer system, each point batch including a plurality of data points representing transactions, wherein the plurality of data points include both feasible and infeasible data points;
• for each received point batch:
• projecting, at the polytope controller, at least a subset of the plurality of data points to the constraint polytope to modify parameters of the sets of transactions of the infeasible data point so as to render them feasible;
• mapping, at the polytope controller, the plurality of data points relative to a boundary defined by the non-linear constraint and storing, in a memory coupled with the polytope controller, a subset of the plurality of data points characterized by a distance from the boundary less than a first threshold value;
• determining, at the polytope controller, whether a number of stored data points exceeds a second threshold value, wherein if the number of stored data points does not exceed the second threshold value, performing the projecting and mapping on a subsequently received point batch;
• wherein if the number of stored data points exceeds the second threshold value, performing, at the polytope controller, Binary Space Partitioning (BSP) on the stored data points to partition the stored points into two partitions, wherein if a number of data points in either of the two partitions is less than a third threshold value, the partition is discarded;
• for each remaining of the two partitions, generating, at the polytope controller using a Support Vector Machine (SVM), a linear approximation of the non-linear constraint which classifies the most feasible data points from the most non-feasible data points of the partition;
• determining, at the polytope controller, whether a number of correctly classified infeasible data points therein exceeds a fourth threshold and a number of incorrectly classified feasible data points does not exceed a fifth threshold, wherein if the fourth threshold is not exceeded or the fifth threshold is exceeded, the BSP and linear approximation using the SVM is performed again on the partition, otherwise storing the generated linear approximation in the memory;
• updating, at the polytope controller, the constraint polytope with an additional constraint comprising the linear approximation;
• determining, at the polytope controller, whether the projection satisfies the constraint polytope;
• and
• generating, at the polytope controller, a dynamic adjustment for use with a second point batch for the constraint polytope;
• and
• wherein some points in each batch are discarded when outside of a boundary and as more and more batches are processed, the processor learns about the boundary thereby decreasing the discard rate over time, without the need to recalculate the internal states of the linear constraints of the constrain polytope.