Abstract: Systems and methods disclose herein procedures for accelerated tree learning. In one class, the acceleration is based on self-adapting learning rates, while in another class, the acceleration is based on a plurality of learning rates, wherein each learning rate varies over the training; each learning rate increases linearly as a respective pseudo residual maintains a direction across sequential training iterations; and each learning rate decreases exponentially as the respective pseudo residual changes direction across sequential training iterations. The latter can be incorporated with other methodologies, such as momentum-augmented gradient boosting and Nesterov Accelerated Gradient Boosting. These systems and methods for accelerated tree learning exhibit a marked reduction in training time and resources required for gradient boosted trees.

Abstract: Systems and methods for computer memory management by a memory coordinator and a plurality of memory consumers. An urgency and memory quota of each memory consumer is initialized by the memory coordinator, which then adjusts the memory quota of each memory consumer such that the sum of the memory quota of each memory consumer does not exceed a finite amount of computer memory. Each memory consumer adjusts its memory usage in response to the quota input and urgency input from the memory coordinator.

Abstract: Systems and methods for deleting data in a versioned database, comprising: generating a version visibility data structure (VVDS) from a version graph and scenario structure; determining each combination of feasible scenarios that when deleted, delete memory; evaluating an amount of memory reclaimed for each combination of feasible scenarios; and deleting one or more combination of scenarios to free up a specific amount of memory.

Abstract: Systems and methods for efficient consolidation of record blocks in a data base. The system comprises: 1) a deletion record set; an in-memory database representation comprising: tables and records; one or more exclusive locks for the records; and a record block index; 2) a persistent database representation comprising: record blocks; and a transaction log. The method comprises: receiving, by a processor, a deletion record set; acquiring, by the processor, an exclusive lock for one or more records in the deletion record set; consolidating, by the processor, one or more record blocks; updating, by the processor, an in-memory record block index; and adding, by the processor, a transaction log entry for the updated record block index update.

Abstract: Systems and methods for replicating data in a versioned database, comprising: receiving a maximum replication size; selecting, a lead scenario for placement in a replication set; the lead scenario having a size less than the maximum replication size; marking all scenarios in the versioned database as unprocessed and the lead scenario as processed; initializing the replication set; adding the lead scenario into the replication set; obtaining a list of candidate scenarios to place in the replication set; determining a best candidate scenario from the list of candidate scenarios; adding the best candidate scenario to the replication set the best candidate scenario as processed; and iterating a new list of candidate scenarios to place in the replication set until there are no more scenario candidates to place in the replication set.

Abstract: Systems and methods for replicating data in a versioned database, comprising: receiving a maximum replication size; selecting, a lead scenario for placement in a replication set; the lead scenario having a size less than the maximum replication size; marking all scenarios in the versioned database as unprocessed and the lead scenario as processed; initializing the replication set; adding the lead scenario into the replication set; obtaining a list of candidate scenarios to place in the replication set; determining a best candidate scenario from the list of candidate scenarios; adding the best candidate scenario to the replication set the best candidate scenario as processed; and iterating a new list of candidate scenarios to place in the replication set until there are no more scenario candidates to place in the replication set.

Abstract: Systems and methods that provide a mechanism to transition static schema to dynamic schema while maintaining backwards capability. Simple removal of static schema elements, followed by replacement with dynamic schema elements, make a third-party code incompatible since the third-party code references schema entities that no longer exist. Provided is a mechanism to decrease the memory use of non-material static schema entities. Transitioning static schema to dynamic schema allows the database to avoid loading non-material schema entities, thereby decreasing overall memory usage.

Abstract: Systems and methods for partitioning data in a versioned database, comprising: initializing, a list of empty partitions; initializing a new empty partition as a current partition; selecting a lead scenario for placement in the current partition, the lead scenario having a size less than a maximum partition size; placing the lead scenario into the current partition; marking the lead scenario as processed; obtaining a list of candidate scenarios to place in the current partition; determining a best candidate scenario from the list of candidate scenarios; adding the best candidate scenario to the current partition; marking the best candidate scenario as processed; iterating a new list of candidate scenarios to place in the current partition until there are no more scenario candidates to place in the current partition; adding the current partition to a partition list; and iterating through all the scenarios in the versioned database.

Abstract: Systems and methods for efficient consolidation of record blocks in a data base. The system comprises: 1) a deletion record set; an in-memory database representation comprising: tables and records; one or more exclusive locks for the records; and a record block index; 2) a persistent database representation comprising: record blocks; and a transaction log. The method comprises: receiving, by a processor, a deletion record set; acquiring, by the processor, an exclusive lock for one or more records in the deletion record set; consolidating, by the processor, one or more record blocks; updating, by the processor, an in-memory record block index; and adding, by the processor, a transaction log entry for the updated record block index update.

Abstract: Systems and methods that provide a mechanism to transition static schema to dynamic schema while maintaining backwards capability. Simple removal of static schema elements, followed by replacement with dynamic schema elements, make a third-party code incompatible since the third-party code references schema entities that no longer exist. Provided is a mechanism to decrease the memory use of non-material static schema entities. Transitioning static schema to dynamic schema allows the database to avoid loading non-material schema entities, thereby decreasing overall memory usage.

Abstract: Systems, methods, and computer program products for deleting data in a versioned database, generating, by a processor, a version visibility data structure (VVDS) for the versioned database that maps a version to a respective Version Scenario List (VSL) from a version graph and scenario structure of the versioned database. Generating the VVDS comprises initializing the VVDS by setting a unique version as a key to an entry of the VVDS, and a respective empty VSL in the entry; accessing the respective Scenario Version List (SVL); and iterating, each entry in the VVDS from the list of all scenarios in the database and the respective SVL until there are no more unprocessed scenarios.

Abstract: Systems and methods for efficient bulk data deletion. The system comprises: 1) a deletion record set; an in-memory database representation comprising: tables and records; one or more exclusive locks for the records; and a record block index; 2) a persistent database representation comprising: record blocks; and a transaction log. The method comprises: receiving, by a processor, a deletion record set; acquiring, by the processor, an exclusive lock for one or more records in the deletion record set; deleting, by the processor, the deletion record set from an in-memory representation of the database; generating, by the processor, one or more post-delete record block sets; updating, by the processor, an in-memory record block index; writing, by the processor, the one or more post-delete record block sets to a persistent storage representation of the database; and, adding, by the processor, a transaction log entry for the record block index update.

Abstract: Systems and methods disclose herein procedures for accelerated tree learning. In one class, the acceleration is based on self-adapting learning rates, while in another class, the acceleration is based on a plurality of learning rates, wherein each learning rate varies over the training; each learning rate increases linearly as a respective pseudo residual maintains a direction across sequential training iterations; and each learning rate decreases exponentially as the respective pseudo residual changes direction across sequential training iterations. The latter can be incorporated with other methodologies, such as momentum-augmented gradient boosting and Nesterov Accelerated Gradient Boosting. These systems and methods for accelerated tree learning exhibit a marked reduction in training time and resources required for gradient boosted trees.

Abstract: Systems and methods for computer memory management by a memory coordinator and a plurality of memory consumers. An urgency and memory quota of each memory consumer is initialized by the memory coordinator, which then adjusts the memory quota of each memory consumer such that the sum of the memory quota of each memory consumer does not exceed a finite amount of computer memory. Each memory consumer adjusts its memory usage in response to the quota input and urgency input from the memory coordinator.

Abstract: Systems and methods for replicating data in a versioned database, comprising: receiving a maximum replication size; selecting, a lead scenario for placement in a replication set; the lead scenario having a size less than the maximum replication size; marking all scenarios in the versioned database as unprocessed and the lead scenario as processed; initializing the replication set; adding the lead scenario into the replication set; obtaining a list of candidate scenarios to place in the replication set; determining a best candidate scenario from the list of candidate scenarios; adding the best candidate scenario to the replication set the best candidate scenario as processed; and iterating a new list of candidate scenarios to place in the replication set until there are no more scenario candidates to place in the replication set.

Abstract: Systems and methods for computer memory management by a memory coordinator and a plurality of memory consumers. An urgency and memory quota of each memory consumer is initialized by the memory coordinator, which then adjusts the memory quota of each memory consumer such that the sum of the memory quota of each memory consumer does not exceed a finite amount of computer memory. Each memory consumer adjusts its memory usage in response to the quota input and urgency input from the memory coordinator.

Abstract: Systems and methods that provide a mechanism to transition static schema to dynamic schema while maintaining backwards capability. Simple removal of static schema elements, followed by replacement with dynamic schema elements, make a third-party code incompatible since the third-party code references schema entities that no longer exist. Provided is a mechanism to decrease the memory use of non-material static schema entities. Transitioning static schema to dynamic schema allows the database to avoid loading non-material schema entities, thereby decreasing overall memory usage.

Abstract: The systems and methods described here can reduce the storage space required (memory and/or disk) to store certain types of data, provide efficient (fast) creation, modification and retrieval of such data, and support such data within the framework of a multi-version database. In some embodiments, the systems and methods can store each field of a set of records as a vector of values, e.g., a data vector. A set of records can be represented using a vector hash vector, or “vhash” vector, wherein each element of the vhash vector contains a unique identifier of a data vector, based on a cryptographic hash of the data vector. A header table can store associations between labels and “vhash” vectors that pertain to those labels. Identical data vectors can be re-used between different record sets or vhash vectors needing that vector, thus saving space.

Abstract: Systems and methods for partitioning data in a versioned database, comprising: initializing, a list of empty partitions; initializing a new empty partition as a current partition; selecting a lead scenario for placement in the current partition, the lead scenario having a size less than a maximum partition size; placing the lead scenario into the current partition; marking the lead scenario as processed; obtaining a list of candidate scenarios to place in the current partition; determining a best candidate scenario from the list of candidate scenarios; adding the best candidate scenario to the current partition; marking the best candidate scenario as processed; iterating a new list of candidate scenarios to place in the current partition until there are no more scenario candidates to place in the current partition; adding the current partition to a partition list; and iterating through all the scenarios in the versioned database.

Abstract: Systems and methods that provide a mechanism to transition static schema to dynamic schema while maintaining backwards capability. Simple removal of static schema elements, followed by replacement with dynamic schema elements, make a third-party code incompatible since the third-party code references schema entities that no longer exist. Provided is a mechanism to decrease the memory use of non-material static schema entities. Transitioning static schema to dynamic schema allows the database to avoid loading non-material schema entities, thereby decreasing overall memory usage.