Google Inc.

Jordan M Breckenridge

Gideon S Mann

Wei-Hao Lin

Robert Kaplow

Travis H K Green

David Konerding

Daniel Belov

Travis Green

Travis H Green

Methods, systems, and apparatus, including computer programs encoded on one or more computer storage devices, for training predictive models. Multiple training data records are received that each include an input data portion and an output data portion. A training data type is determined that corresponds to the training data. For example, a training data type can be determined by inputting the output data portions into one or more trained predictive classifiers. In other example, the training data type can be determined by comparison of the output data portions to data formats. Based on the determined training data type, a set of training functions are identified that are compatible with the training data of the determined training data type. The training data and the identified set of training functions are used to train multiple predictive models.

Customized predictive analytical model training

A global-level manager access a work order from a client and parameters associated with the work order. A service level agreement to meet the work order parameters is determined. The service level agreement includes a price. An indication is received from the client that the service level agreement is accepted. The one or more input files are partitioned into multiple shards, and the work order into multiple jobs. The jobs are distributed among a plurality of clusters to be processed using underutilized computing resources in the clusters. The job outputs are combined to form the work order output. The jobs are monitored to insure that the deadline for completion of the work order will be met.

Opportunistic job processing using worker processes comprising instances of executable processes created by work order binary code

Opportunistic job processing of input data divided into partitions and distributed amongst task level managers via a peer-to-peer mechanism supplied by a cluster cache

Opportunistic job processing of input data divided into partitions of different sizes

Opportunistic job Processing of input data divided into partitions of different sizes

A global-level manager accesses a work order referencing at least one executable file and one or more input files. The one or more input files are partitioned into multiple shards. The work order is partitioned into multiple jobs. The jobs are distributed among a plurality of to be processed by a task level manager at each of the plurality of clusters. The executable file is loaded into the native client environment through a validator. The validator is configured to insure that the executable file does not include one or more of a defined set of instructions, does not call instructions outside of the executable file, and does not read or write data outside of a data region associated with the executable file.

Opportunistic job processing in distributed computing resources with an instantiated native client environment with limited read/write access

Methods, systems, and apparatus, including computer programs encoded on one or more computer storage devices, for training a predictive model. In one aspect, a method includes receiving over a network predictive modeling training data from a client computing system. The training data and multiple training functions obtained from a repository of training functions are used to train multiple predictive models. A score is generated for each of the trained predictive models, where each score represents an estimation of the effectiveness of the respective trained predictive model. A first trained predictive model is selected from among the trained predictive models based on the generated scores. Access to the first trained predictive model is provided over the network.

Predictive analytic modeling platform

A global-level manager receives a work order referencing at least one executable file and one or more input files. The one or more input files include multiple input resources. A first type of input resource is identified in the one or more input files and a second type of input resource in the one or more input files. The first type of input resource is split into a plurality of first-type input shards. The second type of input resource is split into a plurality of second-type input shards. The plurality of second-type input shards are associated with each of the first-type input shards. For each of the first-type input shards, the global-level manager distributes the first-type input shards, the associated second-type input shards, and the executable file to a single.

Opportunistic job processing

A central storage configured to store one or more input files and an executable file. A work order frontend is configured to receive, from the client, a representational state transfer (RESTful) message that contains a reference to the one or more input files in the central storage. The work order frontend is further configured to transmit, to a global-level manager, a work order. The global-level manager is configured to access the work order. The global-level manager is further configured to partition the one or more input files into multiple shards, the work order into multiple jobs, each job being associated with one or more of the multiple shards and the executable file. The global-level manager is further configured to distribute the jobs among a plurality of clusters.

Opportunistic job processing in a distributed computer environment

Methods, systems, and apparatus, including computer programs encoded on one or more computer storage devices, for training a predictive model. In one aspect, a method includes receiving over a network predictive modeling training data from a client computing system. The training data and multiple training functions obtained from a repository of training functions are used to train multiple predictive models. A score is generated for each of the trained predictive models, where each score represents an estimation of the effectiveness of the respective trained predictive model. A first trained predictive model is selected from among the trained predictive models based on the generated scores. Access to the first trained predictive model is provided to the client computing system.

Growing community of inventors

Menlo Park, CA, United States of America

Jordan M Breckenridge