Microsoft Technology Licensing, LLC

Gary Shon Katzenberger

Anders Hejlsberg

Matthew J Warren

Peter A Hallam

Luca Bolognese

Henricus Johannes Maria Meijer

Vijay Mital

Dinesh Chandrakant Kulkarni

Darryl Ellis Rubin

Olivier Colle

David George Green

Suraj T Poozhiyil

Ivan Matantsev

Costin I Eseanu

Eric Anthony Erhardt

Brian C Beckman

Samuel James McKelvie

Robert Victor Welland

Donald F Box

Andrew Douglas Reddish

Thomas William Finley

Mohammad Zeeshan Siddiqui

Petro Luferenko

Yael Dekel

Phillip Lee Bogle

John Joseph Duffy

Christopher Wellington Brumme

Vance Palmer Morrison

Erik Meijer

David G Green

R Donald Thompson, Iii

Eric Gunnerson

Robert Povey

Dinesh Chandrakant Kulkami

Pete Luferenko

Donald Glen MacKinnon

Krasimir A Aleksandrov

Henriçus Johannes Maria Meijer

Sparse data handling and/or buffer sharing are implemented. Data may be buffered in reusable buffer arrays. Data may comprise fixed or variable length vectors, which may be represented as sparse or dense vectors in a values array and indices array. Data may be materialized from a dataview comprising a non-materialized view of data in a machine-learning pipeline by cursoring over rows of the dataview and calling delegate functions to compute data for rows in an active column. A buffer and/or its set of arrays storing a first vector may be reused for a second and additional vectors, for example, when the length of buffer arrays is equal to or greater than the length of the second and additional vectors, which may be selectively stored as sparse or dense vectors to fit the array set. Shared buffers may be passed as references between delegate functions for reuse.

Sparse data handling and buffer sharing to reduce memory allocation and reclamation

Methods, systems, and computer products are herein provided for lazy evaluation of input data by a machine learning (ML) framework. An ML pipeline receives input data and compiles a chain of operators into a chain of dataviews configured for lazy evaluation of the input data. Each dataview in the chain represents a computation over data as a non-materialized view of the data. The ML pipeline receives a request for column data and selects a chain of delegates comprising one or more delegates for one or more dataviews in the chain to fulfill the request. The ML pipeline processes the input data with the selected chain of delegates. The ML pipeline performs delegate chaining on a dataview. A feature value for a feature column of the dataview is determined based on the delegate chaining and provided to an ML algorithm to predict column data.

Efficient streaming based lazily-evaluated machine learning framework

Embodiments provide a machine learning framework that enables developers to author and deploy machine learning pipelines into their applications regardless of the programming language in which the applications are structured. The framework may provide a programming language-specific API that enables the application to call a plurality of operators provided by the framework. The framework provides any number of APIs, each for a different programming language. The pipeline generated via the application is represented as an execution graph comprising node(s), where each node represents a particular operator. When a pipeline is submitted for execution, calls to the operators are detected, and nodes corresponding to the operators are generated for the execution graph. Once the execution graph is complete, the execution graph is provided to an execution engine, which reconstructs a corresponding execution graph that is structured in accordance with the operators' programming language and executes the reconstructed execution graph.

Optimized machine learning pipeline execution based on execution graphs

A business intelligence (BI) document preserves references to identities and formats of remote data sources and allows a local computing device to offload analytical operations to remote data sources. The BI document specifies a graph of entities connected by directed edges from the output of one entity to an input of another entity. An entity, for example, can represent without limitation a data structure, an external data source, a control element, an external event source, a visualization, or an update service. The entities of a BI document at a local computing device can reference data at an original data source&#x2014;rather than extracting data from the original data source to a preferred local datastore. An entity of the BI document can direct a remote data source to execute transformations on the remote data before returning a solution to the local computing device.

Business intelligence document

A business intelligence document provides functionality for testing a hypothesis on aggregated data in a business intelligence document (e.g., a spreadsheet-like document), wherein one or more of the input data values and transformation properties are designated as constrained (e.g., invariant or constrained within a range, set, enumeration, or domain). The hypothesis, which is articulated as a data mining assertion, is input through the user interface of the business intelligence document (e.g., via an expression interface or properties of a row, column, or cell) and solved over the aggregated data. The solution is then presented through the user interface of the spreadsheet-like document, such as in a table, graph, histogram, etc.

Data mining in a business intelligence document

Techniques for representing and publishing an interactive document useful for analyzing data. The document may be represented as a directed acyclic graph of entities interconnected by edges. The entities may be of multiple types. Yet, a broad range of interactive documents may be represented by a limited number of types of entities and the capabilities to interconnect entities of different types and to share a data schema across entities of different types. A tool may enable a user to author such documents. The tool may also facilitate publishing of the document. For publishing, the document may be converted to an executable form. Prior to such a conversion, the graph may be modified for more efficient processing. The graph may also be partitioned such that portions of the graph, when distributed across tiers of a computing system, such as a cloud-based platform, execute on computing devices that provide efficient operation.

Representation of an interactive document as a graph of entities

An interactive recommendation system generates one or more recommendations (e.g., recommended products, travel destinations, etc.) for a user based on a recommendation model. The recommendation model includes one or more criteria that are used to analyze a datastore of user characteristics (e.g., a user's age, location, past online behavior, etc.) and generate one or more recommendations based thereon. The interactive recommendation system further presents a user interface that allows the user to interactively modify the criteria of the recommendation model and to apply the modified recommendation model to the datastore in order to generate one or more modified recommendations. In this manner, for example, the user can customize the recommendations he or she receives by interacting with the recommendation system to modify the recommendation model used to generate such recommendations.

Interactive recommendations

A compiler supporting a language in which selected semantic objects are represented as data objects. The data objects may be used in multiple ways to expand the capabilities of the programming language. Data objects may be passed to applications and used to create executable instructions for that application. In this way, instructions written in the native language of the compiler may be used to control applications that accept programs in a language inconsistent with the native language of the compiler. The syntax checking and variable binding capabilities of the compiler may be used for those instructions that will be executed by an application separate from the object code generated by the compiler. The semantic objects represented as data objects may be selected based on express operations included in the source code or may be based on implicit type conversion.

Compiler supporting programs as data objects

Architecture that extends existing types including constructed and sealed types using extension methods. Extension methods are static methods that can be invoked using instance method syntax. Extension methods are declared by specifying the keyword 'this' as a modifier on the first parameter of the methods. Extension methods have all the capabilities of regular static methods. In addition, once imported, extension methods can be invoked using instance method syntax. Also disclosed are rules for importing extension methods, and extension method invocations. Extension methods are imported through using-namespace-directives. In addition to importing the types contained in a namespace, a using-namespace-directive imports all extension methods in all static classes in the namespace. In effect, imported extension methods appear as additional methods on the types that are given by their first parameter and have lower precedence than regular instance methods.

Architecture that extends types using extension methods

Anonymous types for a programming language. Non-denotable anonymous types are types that a compiler generates on behalf of a programmer and can then be used as nominal types. Insofar as anonymous types do not have a name, they can be used only inside a method in which they are created. Additionally, an anonymous type cannot be exposed in any way outside of the method. Syntax to create an anonymous type is useful when employed with a Select operator to generate a result with a particular shape, without the need of having a regular type for it. Anonymous types are expressible such that an expression of that type can be written. Translation of an anonymous type by a compiler generates a nominal class that implements Equals and GetHashCode methods. There is equivalence of anonymous types within the same method, and conversion of an unrealized structural type into structurally compatible nominal type.

Anonymous types for statically typed queries

Syntax for creating object instances utilizing expressions instead of statements. A syntax component facilitates extension of conventional object-creation-expression syntax. An initializer component provides for initialization of a newly-created entity via object and collection initializers. The initializer component provides initialization of the newly-created entity utilizing an object initializer where the object initializer specifies values for one or more fields or properties of the newly-created entity which is an object, and consists of a sequence of member initializers enclosed by tokens and separated by commas. The initializer component can also provides initialization of the newly-created entity utilizing a collection initializer when the underlying object implements a certain interface or implements a certain pattern. An overloading constructor can be called as part of the initialization process. In another aspect, code inferencing is disclosed whereby given a type, the corresponding code can be inferred by the compiler and inserted for compiling by a compiler.

Extending expression-based syntax for creating object instances

Growing community of inventors

Woodinville, WA, United States of America

Gary Shon Katzenberger