CN106779088A

CN106779088A - Perform the method and system of machine learning flow

Info

Publication number: CN106779088A
Application number: CN201611113470.4A
Authority: CN
Inventors: 栾淑君; 杨慧斌; 孙迪
Original assignee: Beijing Wusi Imagination Technology Co Ltd
Current assignee: Beijing Wusi Imagination Technology Co Ltd
Priority date: 2016-12-06
Filing date: 2016-12-06
Publication date: 2017-05-31
Anticipated expiration: 2036-12-06
Also published as: CN109993316B; CN106779088B; CN109993316A

Abstract

There is provided a kind of method and system for performing machine learning flow, methods described includes：(A) show the graphical interfaces for configuring machine learning tasks to user and detect the input operation that user is performed by graphical interfaces, wherein, machine learning task is used to perform the data processing included by machine learning flow；(B) the machine learning task is configured by the input operation that the graphical interfaces is performed according to the user for detecting；And (C) is in the case of the machine learning task for not performing configuration, the data attribute information relevant with the machine learning task is inferred, wherein, data attribute information includes the title and/or data type of data community.Correspondingly, fewer resource and time can be spent and the data attribute information in each stage in machine learning flow is effectively obtained, so as to improves the operability of Machine learning tools.

Description

Perform the method and system of machine learning flow

Technical field

All things considered of the present invention is related to artificial intelligence field, more specifically to a kind of execution machine learning flow Method and system.

Background technology

With the appearance of mass data, artificial intelligence technology is developed rapidly, wherein, the universal quilt of machine learning techniques For excavating beneficial value from the data record of magnanimity (for example, finance data, internet data etc.),

However, a great problem of artificial intelligence technology application is to lack effectively easy-to-use Machine learning tools, Hen Duoxian Some machine learning platforms are all only oriented to be proficient in the user of machine learning techniques, also, because machine learning is often targeted Be magnanimity complex data and complex data computing, therefore, the user for being even proficient in machine learning techniques is also difficult to effectively The current Machine learning tools of ground operation.In fact, wanting to turn out the technical specialist for being proficient in machine learning, it is necessary to spend a large amount of Time and efforts, this cause artificial intelligence technology application exist talent's threshold higher.On the other hand, machine learning model Prediction effect and the extraction of selection, available data and feature etc. of model have relation, for example, it is determined that feature extraction side During formula, the technological know-how of machine learning is often not only needed to be grasped, in addition it is also necessary to have deep reason to actual prediction problem Solution, and forecasting problem often combines the different practical experiences of different industries, causes to be difficult to only rely on technical specialist to reach completely The machine learning effect of meaning.As can be seen that Machine learning tools need to be improved to some extent in operability, to help different user more Machine learning flow is performed well.

As an example, Azure Machine Learning (referred to as " AML ") are that Microsoft releases on its public cloud Azure The machine learning service used based on Web, the target of the product is simplified using the process of machine learning, in order to open Hair personnel, business diagnosis teacher and data scientist are carried out extensively, easily apply.

In AML, user can complete each machine learning task in machine learning flow by DAG (directed acyclic graph) The configuration of (for example, data importing, Data Format Transform, data conversion, feature extraction, model training etc.), wherein, in user's choosing Select after the certain vertex in operation DAG, the machine learning task representated by the summit will be performed, also, conduct performs knot The field name of data is displayed on screen obtained from fruit.

Particularly, reference picture 1A, the machine learning flow set up in AML may include such as adult's investigation account of receipt The row (Select Columns in Dataset) imported in (Adult Census Income Binary), selection data set The machine learning tasks such as data (Clean Missing Data) are lost with cleaning, wherein, completing to " in selection data set Row " configuration after, downstream machine learning tasks " cleaning lose data " can be initially configured.

However, as shown in fig. 1b, due to " row in selection data set " before, this task is not carried out, so nothing Method configures " data are lost in cleaning " according to field name, correspondingly, shows that reminder message " will perform experiment on screen Column selection (the Name-based column selection will be enabled after based on title are enabled afterwards running the experiment)”。

In fig. 1 c, it can be seen that " the choosing of machine learning task is actually performed by " RUN " that clicks on below screen Select the row in data set ".Correspondingly, in Fig. 1 D, downstream machine learning tasks " data are lost in cleaning " are become able to based on name Claim to be configured accordingly.

As can be seen that in AML, during machine learning process is configured, user cannot in advance recognize any pass In the information by data field resulting after the treatment of each machine learning task, only in corresponding machine learning tasks by reality After border performs, the field name of result data can be just known.However, because machine-learning process often refers to mass data, Perform machine learning task and will take for substantial amounts of time and computing resource, this causes timely and effectively obtain or utilize each rank The data attribute information of section.

The content of the invention

Exemplary embodiment of the invention is intended to overcome cannot be had in time when machine learning flow is performed in the prior art Effect ground obtains the defect of data attribute information.

A kind of exemplary embodiment of the invention, there is provided method of execution machine learning flow, including：(A) to Family shows the graphical interfaces for configuring machine learning tasks and detects the input operation that user is performed by graphical interfaces, its In, machine learning task is used to perform the data processing included by machine learning flow；(B) institute is passed through according to the user for detecting The input operation of graphical interfaces execution is stated to configure the machine learning task；And (C) is not performing the machine of configuration In the case of learning tasks, the data attribute information relevant with the machine learning task is inferred, wherein, data attribute information bag Include the title and/or data type of data community.

Alternatively, methods described also includes：(D) it is illustrated in the data attribute information that step (C) is inferred to user.

Alternatively, in the process, in step (C), the data attribute information being inferred to is that the machine learning is appointed The data attribute information of the input data, output data and/or intermediate processing data of business.

Alternatively, methods described also includes：(E) show that for configuring with the machine learning task be upstream machine to user The graphical interfaces of the downstream machine learning tasks of device learning tasks simultaneously detects that user is grasped by the input that the graphical interfaces is performed Make；(F) the downstream machine study times is configured by the input operation that the graphical interfaces is performed according to the user for detecting Business；(G) configuration of the downstream machine learning tasks is checked based on the data attribute information being inferred in step (C).

Alternatively, in the process, in step (E), in graphical interfaces being illustrated in step (C) to user is inferred to Data attribute information so that user configures the downstream machine learning tasks based on the data attribute information of displaying.

Alternatively, in the process, the configuration in response to the machine learning task terminates to perform step automatically (C), or, in response to being opened with the configuration of the machine learning task as the downstream machine learning tasks of upstream machines learning tasks Begin to perform step (C) automatically, or, the deduction in response to user indicates to perform step (C).

Alternatively, in the process, machine learning task is implemented as the configurable summit in directed acyclic graph, its In, the configuration in response to the machine learning task terminates to perform step (C) automatically, also, represents configuration in user's connection The machine learning task configurable summit with represent with the machine learning task as upstream machines learning tasks under Step (D) is performed during the configurable summit for swimming machine learning task automatically.

Alternatively, methods described also includes：(H) execution according to user indicates to perform the machine of one or more configurations Learning tasks.

Alternatively, in the process, in step (C), by explain the machine learning task execute instruction and/ Or pushed away by performing the execute instruction for the data from the sample survey extracted among the input data of the machine learning task The intermediate processing data of the machine learning task of breaking and/or the data attribute information of output data.

In accordance with an alternative illustrative embodiment of the present invention, there is provided it is a kind of perform machine learning flow system, including：Display Device, for showing the graphical interfaces for configuring machine learning tasks to user, wherein, machine learning task is used to perform machine Data processing included by device learning process；Detection means, for detecting the input operation that user is performed by graphical interfaces；Match somebody with somebody Device is put, for configuring the machine learning times by the input operation that the graphical interfaces is performed according to the user for detecting Business；And apparatus for predicting, in the case of the machine learning task for not performing configuration, inferring and the machine learning The relevant data attribute information of task, wherein, data attribute information includes the title and/or data type of data community.

Alternatively, in the system, the data attribute that display device is also inferred to user's displaying by apparatus for predicting Information.

Alternatively, in the system, the data attribute information that apparatus for predicting is inferred to is the machine learning task The data attribute information of input data, output data and/or intermediate processing data.

Alternatively, in the system, display device also shows for configuring to user and is with the machine learning task The graphical interfaces of the downstream machine learning tasks of upstream machines learning tasks；Detection means also detects that user passes through figure circle The input operation that face performs；Configuration device is configured according to the user for detecting by the input operation that the graphical interfaces is performed The downstream machine learning tasks；Also, configuration device is based on being inferred to by apparatus for predicting with the machine learning task Relevant data attribute information checks the configuration of the downstream machine learning tasks.

Alternatively, in the system, display device is inferred to user's displaying in graphical interfaces by apparatus for predicting The data attribute information relevant with the machine learning task so that user based on displaying data attribute information to configure State downstream machine learning tasks.

Alternatively, in the system, apparatus for predicting terminates to push away automatically in response to the configuration of the machine learning task The disconnected data attribute information relevant with the machine learning task, or, apparatus for predicting is in response to the machine learning task For the configuration of the downstream machine learning tasks of upstream machines learning tasks starts to infer have with the machine learning task automatically The data attribute information of pass, or, apparatus for predicting indicates to infer have with the machine learning task in response to the deduction of user The data attribute information of pass.

Alternatively, in the system, machine learning task is implemented as the configurable summit in directed acyclic graph, its In, apparatus for predicting is inferred relevant with the machine learning task automatically in response to the configuration end of the machine learning task Data attribute information, also, the configurable summit of the machine learning task of configuration is represented in user's connection and is represented with institute When stating the configurable summit of the downstream machine learning tasks that machine learning task is upstream machines learning tasks, display device is automatic To the data attribute information relevant with the machine learning task that user's displaying is inferred to by apparatus for predicting.

Alternatively, the system also includes：Performs device, one or more are performed for being indicated according to the execution of user The machine learning task of configuration.

Alternatively, in the system, configuration device is by explaining the execute instruction of the machine learning task and/or leading to Cross and perform the execute instruction to infer for the data from the sample survey extracted among the input data of the machine learning task State the intermediate processing data of machine learning task and/or the data attribute information of output data.

In accordance with an alternative illustrative embodiment of the present invention, there is provided a kind of computer-readable for performing machine learning flow Medium, wherein, record has the computer program for performing following steps on the computer-readable medium：(A) to user Show the graphical interfaces for configuring machine learning tasks and detect the input operation that user is performed by graphical interfaces, wherein, Machine learning task is used to perform the data processing included by machine learning flow；(B) according to the user for detecting by described The input operation that graphical interfaces is performed configures the machine learning task；And (C) is not performing the engineering of configuration In the case of habit task, the data attribute information relevant with the machine learning task is inferred, wherein, data attribute information includes The title and/or data type of data community.

Alternatively, in the computer-readable medium, the computer program is additionally operable to perform step (D)：To user It is illustrated in the data attribute information that step (C) is inferred to.

Alternatively, in the computer-readable medium, in step (C), the data attribute information being inferred to is described The data attribute information of the input data, output data and/or intermediate processing data of machine learning task.

Alternatively, in the computer-readable medium, the computer program is additionally operable to perform following steps：(E) to User shows for configuring the figure with the machine learning task as the downstream machine learning tasks of upstream machines learning tasks Simultaneously detect the input operation that user is performed by the graphical interfaces in interface；(F) figure is passed through according to the user for detecting The input operation that interface performs configures the downstream machine learning tasks；(G) based on the data attribute being inferred in step (C) Information checks the configuration of the downstream machine learning tasks.

Alternatively, in the computer-readable medium, in step (E), step is illustrated in user in graphical interfaces Suddenly the data attribute information that (C) is inferred to so that user configures the downstream machine based on the data attribute information of displaying Habit task.

Alternatively, in the computer-readable medium, the configuration in response to the machine learning task terminates to come automatic Step (C) is performed, or, in response to the downstream machine learning tasks with the machine learning task as upstream machines learning tasks Configuration start automatically to perform step (C), or, the deduction instruction in response to user performs step (C).

Alternatively, in the computer-readable medium, machine learning task is implemented as matching somebody with somebody in directed acyclic graph Top set point, wherein, the configuration in response to the machine learning task terminates to perform step (C) automatically, also, in user's connection Represent configuration the machine learning task configurable summit with represents with the machine learning task be upstream machines study Step (D) is performed during the configurable summit of the downstream machine learning tasks of task automatically.

Alternatively, in the computer-readable medium, the computer program is additionally operable to perform step (H)：According to The execution at family indicates to perform the machine learning task of one or more configurations.

Alternatively, in the computer-readable medium, in step (C), by explaining the machine learning task Execute instruction and/or by among the input data of the machine learning task extract data from the sample survey perform described in hold Row instructs to infer the intermediate processing data of the machine learning task and/or the data attribute information of output data.

In accordance with an alternative illustrative embodiment of the present invention, there is provided it is a kind of perform machine learning flow computing device, including Memory unit and processor, the set of computer-executable instructions that is stored with memory unit are closed, when the computer executable instructions When set is by the computing device, following step is performed：(A) figure circle for configuring machine learning tasks is shown to user The input operation that user is performed by graphical interfaces is simultaneously detected in face, wherein, machine learning task is used to perform machine learning flow Included data processing；(B) according to the user for detecting is configured by input operation that the graphical interfaces is performed Machine learning task；And (C) is in the case of the machine learning task for not performing configuration, infer and the machine learning The relevant data attribute information of task, wherein, data attribute information includes the title and/or data type of data community.

Alternatively, in the computing device, when the set of computer-executable instructions is closed by the computing device, Also perform step (D)：The data attribute information that step (C) is inferred to is illustrated in user.

Alternatively, in the computing device, in step (C), the data attribute information being inferred to is the engineering The data attribute information of the input data, output data and/or intermediate processing data of habit task.

Alternatively, in the computing device, when the set of computer-executable instructions is closed by the computing device, Also perform following steps：(E) under showing for configuring with the machine learning task as upstream machines learning tasks to user Swim the graphical interfaces of machine learning task and detect the input operation that user is performed by the graphical interfaces；(F) according to detection To user the downstream machine learning tasks are configured by input operation that the graphical interfaces is performed；(G) it is based in step Suddenly the data attribute information that (C) is inferred to checks the configuration of the downstream machine learning tasks.

Alternatively, in the computing device, in step (E), in graphical interfaces being illustrated in step (C) to user pushes away Break the data attribute information for so that user configures the downstream machine learning tasks based on the data attribute information of displaying.

Alternatively, in the computing device, the configuration in response to the machine learning task terminates to perform step automatically Suddenly (C), or, in response to the configuration with the machine learning task as the downstream machine learning tasks of upstream machines learning tasks Start to perform step (C) automatically, or, the deduction in response to user indicates to perform step (C).

Alternatively, in the computing device, machine learning task is implemented as the configurable summit in directed acyclic graph, Wherein, the configuration in response to the machine learning task terminates to perform step (C) automatically, also, matches somebody with somebody in user's connection representative The configurable summit of the machine learning task put is with to represent with the machine learning task be upstream machines learning tasks Step (D) is performed during the configurable summit of downstream machine learning tasks automatically.

Alternatively, in the computing device, when the set of computer-executable instructions is closed by the computing device, Also perform step (H)：Execution according to user indicates to perform the machine learning task of one or more configurations.

Alternatively, in the computing device, in step (C), referred to by the execution for explaining the machine learning task Make and/or perform the execute instruction by for the data from the sample survey extracted among the input data of the machine learning task To infer the intermediate processing data of the machine learning task and/or the data attribute information of output data.

In the method and system of execution machine learning flow according to an exemplary embodiment of the present invention, can not perform In the case of the machine learning task being configured, the data attribute information relevant with the machine learning task is inferred to so that energy Enough spend fewer resource and time and effectively obtain the data attribute information in each stage in machine learning flow, so as to improve The operability of Machine learning tools.

Brief description of the drawings

From detailed description below in conjunction with the accompanying drawings to the embodiment of the present invention, it is of the invention these and/or other aspect and Advantage will become clearer and be easier to understand, wherein：

Figure 1A to Fig. 1 D show the example at the interface that machine learning flow is performed in the machine learning platform of prior art；

Fig. 2 shows the block diagram of the system of execution machine learning flow according to an exemplary embodiment of the present invention；

Fig. 3 shows the block diagram of the system of the execution machine learning flow according to another exemplary embodiment of the present invention；

Fig. 4 shows the flow chart of the method for execution machine learning flow according to an exemplary embodiment of the present invention；

Fig. 5 shows the flow chart of the method for the execution machine learning flow according to another exemplary embodiment of the present invention；

Fig. 6 shows the flow chart of the method for the execution machine learning flow according to another exemplary embodiment of the present invention；With And

Fig. 7 A to Fig. 7 F show to perform machine learning flow in machine learning platform according to an exemplary embodiment of the present invention Interface example.

Specific embodiment

In order that those skilled in the art more fully understand the present invention, with reference to the accompanying drawings and detailed description to this hair Bright exemplary embodiment is described in further detail.

In an exemplary embodiment of the present invention, machine learning flow is performed in the following manner：Configuring machine After the particular machine learning tasks being related in learning process, the data attribute letter relevant with the particular machine learning tasks is inferred Breath so that can spend less computing resource and time and be previously obtained in the case where the particular machine learning tasks are not performed By the data attribute information updated after each processing links in machine learning flow, so as in these data attributes of later use Information, so that the ease for operation of reinforcement machine learning.

Here, machine learning is the inevitable outcome that artificial intelligence study develops into certain phase, and it is devoted to by calculating Means, the performance of system itself is improved using experience.In computer systems, " experience " is generally deposited in " data " form , here, per data record can be seen as the description on an event or object, corresponding to an example or sample. In data record, including reflection performance or property in terms of certain of event or object each item, these items can be described as " category Property ".By machine learning algorithm, " model " can be produced from data, that is to say, that be supplied to machine learning to calculate empirical data Method, just can produce model based on these empirical datas, and when in face of news, model can provide corresponding judgement, i.e. prediction As a result.Machine learning can be implemented as the form of " supervised learning ", " unsupervised learning " or " semi-supervised learning ", it should be noted that The present invention does not carry out specific limitation to specific machine learning algorithm.Further, it should also be noted that in training and the mistake of application model Cheng Zhong, may also be combined with other means such as statistic algorithm.

Fig. 2 shows the block diagram of the system of execution machine learning flow according to an exemplary embodiment of the present invention.Particularly, The system can be the machine based on C/S (client/server) framework, B/S (browser/server) frameworks or unit operation Device learning platform, the machine learning platform may include the various machine learning such as training, test and/or the application of machine learning model Flow.Above-mentioned flow data record to be dealt with can be the online data for producing, previously generate and store data, also may be used Being from the data of external reception by input unit or transmission medium.These data can relate to individual, enterprise or tissue Information, for example, identity, educational background, occupation, assets, contact method, debt, income, the information such as get a profit, pay taxes.Or, these numbers According to the information that can also refer to business relevant item, for example, turnover, both parties, subject matter on deal contract, transaction ground The information such as point.It should be noted that the data attribute information mentioned in exemplary embodiment of the invention can relate to any object or affairs Performance or property in terms of certain, and be not limited to be defined individual, object, tissue, unit, mechanism, project, event etc. or Description.

These data can be derived from inside the entity for expecting to perform machine learning flow, for example, from expectation acquisition machine Bank, enterprise, school of device learning outcome etc.；These data also can be beyond above-mentioned entity, for example, being carried from data For business, internet (for example, social network sites), mobile operator, APP operator, express company, credit institution etc..Alternatively, on State internal data and external data can be combined and use, the machine learning sample of more information is carried to be formed.

In system according to an exemplary embodiment of the present invention, machine learning process can be configured by patterned mode (the machine learning flow is made up of one or more machine learning tasks), also, can not actually perform the machine that has configured In the case of device learning tasks, the data attribute information of correlation is effectively obtained.System shown in Fig. 2 can all by computer Program realized with software mode, can also be realized by special hardware unit, can also be by way of software and hardware combining come real It is existing.Correspondingly, each device of the system shown in composition Fig. 2 can only rely on computer program to realize the void of corresponding function Intend module, or the universal or special device of the function is realized by hardware configuration, can also be that operation has accordingly Hardware unit of computer program etc..

As shown in Fig. 2 display device 100 is for showing the graphical interfaces for configuring machine learning tasks to user, its In, machine learning task is used to perform the data processing included by machine learning flow.Particularly, machine learning flow can be by One or more executable machine learning task compositions, the executable such as data of these machine learning tasks are split, feature is carried Take, feature importance analysis, model training, model prediction, the data processing such as model evaluation.From execution sequence and/or data flow From the point of view of on, the relativeness between each machine learning task can be expressed as upstream machines learning tasks and downstream machine study Task, typically, downstream machine learning tasks are located at after upstream machines learning tasks, also, upstream machines learning tasks At least part of output data can be used as at least part of input data of downstream machine learning tasks.Additionally, from implementation See, these machine learning tasks can be user can be completed by selection operation configuration executable task, or What user can be write using SQL (SQL) or PySpark (Spark Python API) executable appoints Business.

In the graphical interfaces of the display of display device 100, it may include the various elements for configuring machine learning tasks, this Sample, by detecting the input operation that user performs on the graphical interfaces, you can correspondingly configure relevant machine learning and appoint Business.

Detection means 200 is used to detect the input operation that user is performed by graphical interfaces.Here, user can be by such as The various modes such as mouse, keyboard, gesture, touch, voice, action perform input operation, and correspondingly, detection means 200 can lead to Corresponding sensing means are crossed to detect various input operations of the user performed by graphical interfaces.

As an example, display device 100 can be the display screen with touch function, and in this case, detection means 200 can be integrated in display device 100, and correspondingly, user can perform touch operation come complete by the graphical interfaces of display Into the configuration of machine learning task.

Configuration device 300 is used for the input operation performed by the graphical interfaces according to the user for detecting to configure State machine learning task.Particularly, the user input operation that detection means 200 will can be detected sends configuration device 300 to, Correspondingly, configuration device 300 can determine that the implication of these input operations, and complete machine learning task according to determining result Configuration, for example, the input data of machine learning task, execution parameter, output result show.

Apparatus for predicting 400 is used in the case of the machine learning task for not performing configuration, infer and the machine The relevant data attribute information of learning tasks, wherein, data attribute information includes the title and/or data class of data community Type.

Particularly, apparatus for predicting 400 need not actually perform the machine learning task for having configured, i.e. need not be according to The execution parameter of configuration, the input data configured come actual treatment by corresponding execute instruction, but be only inferred in advance The data attribute information relevant with the machine learning task for having configured.Here, data attribute information can be on related data Any information of attribute, for example, it may be the title of attribute field, or, can be the data type of attribute field, or, Both field name and data type can also simultaneously be included.

Here, apparatus for predicting 400 can proceed by inference operations under any appropriate opportunity or triggering.As an example, In order to improve the convenience of operation, apparatus for predicting 400 can be in due course and automatically carry out inference operations, for example, inferring dress 400 configurations that may be in response to the machine learning task are put to terminate to infer the data relevant with the machine learning task automatically Attribute information, or, apparatus for predicting 400 may be in response to the downstream machine with the machine learning task as upstream machines learning tasks The configuration of device learning tasks starts to infer the data attribute information relevant with the machine learning task automatically.However, should manage Solution, the present invention is not limited to this, alternately, apparatus for predicting 400 may be in response to user deduction indicate infer with The relevant data attribute information of the machine learning task.

For example, the configuration in current machine learning tasks terminates (for example, user clicks confirmation current machine learning tasks The button that completes of configuration and correspondingly complete the actual disposition of machine learning task) when, apparatus for predicting 400 can be performed automatically Inference operations；Or, start when the configuration of the next machine learning task after current machine learning tasks (for example, The newly-built downstream machine learning tasks of user) when, apparatus for predicting 400 can perform for the current machine learning tasks automatically Inference operations.Or, when user draws an inference instruction (for example, user clicks the startup inference operations of special setting manually Button) when, apparatus for predicting 400 can correspondingly perform inference operations.

Additionally, exemplary embodiment of the invention, apparatus for predicting 400 are it can be inferred that relevant with machine learning task Various data attribute informations, as an example, these data attribute informations can relate to machine learning task processing data (for example, Input data, output data or intermediate processing data), that is to say, that apparatus for predicting 400 can be by the defeated of the machine learning task Enter data, output data and/or intermediate processing data data attribute information be inferred as it is relevant with the machine learning task Data attribute information.Additionally, data attribute information relates to other any data relevant with machine learning task, also It is to say, in the case where machine learning task is configured, configuration that can be by apparatus for predicting 40 according to machine learning task or sound Any associated data attributes information that should be configured and be inferred in machine learning task can be applied to of the invention exemplary Embodiment.

Exemplary embodiment of the invention, except directly inferring corresponding data attribute letter according to advance setting Outside breath, can also be according to the various respective characteristics of machine learning task or the position in whole machine learning flow, neatly Different deduction mechanism are set.As an example, apparatus for predicting 40 can correspondingly be inferred to according to the type of machine learning task Relevant data attribute information.That is, from the point of view of whole machine learning flow, appointing for different types of machine learning Business, can infer corresponding data attribute information according to different mechanism, easy to operate so as to strengthening system from different angles Property.

For example, apparatus for predicting 400 optionally can be appointed the machine learning according to the type of the machine learning task The data attribute information of the input data, output data and/or intermediate processing data of business is inferred as and the machine learning task Relevant data attribute information.

As an example, for some machine learning tasks, each attribute field of every data record of its output It is middle only some follow-up machine learning treatment will to be participated according to original form, therefore, downstream machine study is appointed Business may require that all words from the data record exported as some machine learning tasks described in upstream machines learning tasks A part is chosen in section.In this case, can be inferred to its according to the configuration of current machine learning tasks defeated for apparatus for predicting 400 Go out the total data attribute information (for example, each field name and/or data type) of data, to show user in advance, Allow users to effectively therefrom select a part of data community to participate in downstream machine learning tasks；Or, Yong Huke A part of data community is selected by writing the modes such as code in the case of without reference to the data field of any displaying, In this case, the attribute information being inferred to can also be used to check the configuration of downstream machine learning tasks (that is, to select data category The code of property field) whether meet specification.

It should be understood, however, that：Above-mentioned example is not intended to limit the scope of illustrative examples of the present invention, those skilled in the art The data attribute information of which data can be inferred to flexibly set according to the scene of application, if for example, current machine learns Treatment of the task to input data is directed not only to the screening of field, further relates to the conversion of form (for example, being converted to key-value (key-value pair) form), in this case, because key-value forms are readable poor, and user is in configuration downstream machine study May desire to observe the unprocessed form of screening field during task, therefore, apparatus for predicting 400 can appoint according to current machine study The configuration of business is inferred to the data attribute information of its intermediate processing data, i.e. the title of the screening field under unprocessed form and/or Data type.

Similarly, as needed, apparatus for predicting 400 can also go out its input according to the type inference of current machine learning tasks The data attribute information of data, for follow-up machine learning task.

Additionally, when inference operations are performed, apparatus for predicting 400 can be according to the configuration of machine learning task itself, with described Machine learning task is the configuration of the downstream machine learning tasks of upstream machines learning tasks, and/or upstream machines study Relevance between task and downstream machine learning tasks etc. is inferred to the attribute information of corresponding data.

As an example, apparatus for predicting 400 can using intact transparent transmission mode by with previous machine learning task (example Such as, upper machine learning task) corresponding data attribute information directly as current machine learning tasks data attribute information.

Additionally, apparatus for predicting 400 can be by explaining the execute instruction of machine learning tasks and/or by for from the machine The data from the sample survey extracted among the input data of device learning tasks performs the execute instruction to infer the machine learning task Intermediate processing data and/or output data data attribute information.Here, apparatus for predicting 400 can use single instruction solution Release or the mode of data from the sample survey operation is inferred to data attribute information, be dynamically selected also dependent on the complexity of instruction Suitable mode among above two deduction mode.

Exemplary embodiment of the invention, the data attribute information being inferred to by apparatus for predicting 400 is intended to improve The operability of machine-learning process.

For example, the data attribute information that display device 100 can be also inferred to user's displaying by apparatus for predicting 400, phase Ying Di, user can understand each stage running situation of machine learning flow or from displaying by the data attribute information for showing Data attribute information in select downstream machine learning tasks input data.

As described above, those skilled in the art can be set according to appropriate mode content produced by inference operations and The opportunity of inference operations is performed, and the content that will conclude that is shown.To configure machine learning stream according to directed acyclic graph The situation of journey as an example, wherein, machine learning task can be implemented as the configurable summit in directed acyclic graph, correspondingly, Apparatus for predicting 400 may be in response to the machine learning task configuration terminate to come it is automatic infer it is relevant with the machine learning task Data attribute information, also, user connection represent configuration the machine learning task configurable summit with represent with When the machine learning task is the configurable summit of the downstream machine learning tasks of upstream machines learning tasks, display device 100 data attribute informations relevant with the machine learning task that can be inferred to by apparatus for predicting 400 from trend user displaying.

Additionally, used as another example, the data attribute information being inferred to can also be used to check downstream machine learning tasks Configuration.Particularly, display device 100 can also show for configuring with the machine learning task as upstream machines to user The graphical interfaces of the downstream machine learning tasks of learning tasks, correspondingly, detection means 200 also detects that user passes through above-mentioned figure The input operation that interface performs, also, the input that configuration device 300 is performed according to the user for detecting by the graphical interfaces Operate to configure the downstream machine learning tasks；In this case, configuration device 300 can be based on by apparatus for predicting 400 it Preceding the be inferred to data attribute information relevant with upstream machines learning tasks checks the downstream machine learning tasks Configuration.It is seen in this example that because machine learning flow is that multiple machine learning tasks interconnections are formed, because This, it is necessary to configure each machine learning task successively, here, configure each machine learning task graphical interfaces can with identical or Difference, typically, the element included by graphical interfaces can be roughly the same, also, Partial Elements can appoint according to different machines study The characteristics of being engaged in respective and be adjusted.In this case, the data attribute letter of the upstream machines learning tasks being inferred to before Breath can be used to check the configuration of downstream machine learning tasks, for example, user can be checked to be compiled when downstream machine learning tasks are configured Whether correct Data field names are have input in the code write, whether correct data type etc. has been used.In above-mentioned example In, it is preferred that, display device 100 is inferred before can also showing by apparatus for predicting 400 to user in graphical interfaces The data attribute information relevant with upstream machines learning tasks that goes out so that the data attribute information that user refers to displaying is compiled Write the configuration item of trip machine learning task.That is, the data attribute information of the upstream machines learning tasks being inferred to was both Can be used as reference content during user configuring downstream machine learning tasks, it is also possible to make the configuration of inspection downstream machine learning tasks Basis.

Fig. 3 shows the block diagram of the system of the execution machine learning flow according to another exemplary embodiment of the present invention.In Fig. 3 In shown system, in addition to above-mentioned display device 100, detection means 200, configuration device 300 and apparatus for predicting 400, also Including performs device 500.

Particularly, in the system as shown in fig. 3, display device 100, detection means 200, configuration device 300 and deduction Device 400 can be operated according to the mode in the system shown in Fig. 2.Additionally, performs device 500 can be according to the execution of user Indicate to perform the machine learning task of one or more configurations.

Here, performs device 500 can be used to perform whole machine learning flow, or one or more of machine learning Task.Particularly, when user makes and perform instruction (example for specific one or multiple machine learning tasks for having configured Such as, executive button is pressed) when, performs device 500 can perform one or more of machine learning tasks for having configured；And work as and use Family is made when performing instruction (for example, pressing executive button) for the whole machine learning flow for having configured, and performs device 500 can Perform whole machine learning flow.

It should be understood that said apparatus can be individually configured to perform appointing for the software of specific function, hardware, firmware or above-mentioned item Meaning combination.For example, these devices may correspond to special integrated circuit, pure software code is can also correspond to, can also corresponded to The unit or module being combined with hardware in software.Additionally, the one or more functions that these devices are realized also can be by physics Component in entity device (for example, processor, client or server etc.) is sought unity of action.

The flow of the method for execution machine learning flow according to an exemplary embodiment of the present invention is described referring to Fig. 4 Figure.Here, as an example, the method shown in Fig. 4 can be as shown in Figure 2 system perform, also can completely pass through computer program Realized with software mode, the method shown in Fig. 4 can be also performed by the computing device of particular configuration.For convenience, it is false If method shown in Fig. 4 system as shown in Figure 2 is performed.

Here, machine-learning process can be performed based on the data of collection, wherein, the operation of data acquisition (or importing) can First carried out in advance outside machine learning flow, can also be held as first in machine learning flow machine learning task OK.

Here, as an example, can manually, semi- or fully automated mode carry out gathered data, or the original to gathering Beginning data are processed so that the data record after treatment has appropriate form or form.As an example, can gather in bulk Data.Here, the data record that user is manually entered can be received by input unit (for example, work station).Additionally, can be by complete Automatic mode takes out data record from data source systems, for example, by with software, firmware, hardware or its combination realization Timer mechanism obtains asked data come systematically request data source and from response.The data source may include one or Multiple databases or other servers.The mode of data can be realized automatically obtaining via internal network and/or external network, Wherein may include to be transmitted by internet the data of encryption.It is configured as what is communicated with one another in server, database, network etc. In the case of, data acquisition can be carried out automatically in the case of no manual intervention, it should be noted that can still deposit in this manner Operated in certain user input.Semiautomatic fashion is between manual mode and full-automatic mode.Semiautomatic fashion with it is complete from The difference of flowing mode is that instead of such as timer mechanism by user activated trigger mechanism.In this case, receiving In the case of specific user input, the request for extracting data is just produced.When obtaining data every time, it is preferable that will can capture Data storage in the nonvolatile memory.As an example, availability data warehouse come store during obtaining gather original Data after beginning data and treatment.

Gathered data can be carried out from identical or different data source, for example, except collection client opens credit to bank's application Outside the information data recording (its attribute information fields such as including income, educational background, post, Assets) filled in during card, as Example, can also gather other data records of the client in the bank, for example, loan documentation, current transaction data etc., these are adopted The data record of collection can subsequently be spliced into complete data record.Additionally, can also gather from other privately owned sources or public affairs The data of common source, for example, the data from metadata provider, the data from internet (for example, social network sites), source Data in mobile operator, the data from APP operators, the data from express company, from credit institution Data etc..

Alternatively, can be deposited by the data of hardware cluster (Hadoop clusters, Spark clusters etc.) to collecting Storage and/or treatment, for example, storage, classification and other off-line operations.Additionally, can also be carried out at online stream to the data for gathering Reason.

As an example, the unstructured datas such as text can be converted to the structural data for being easier to use subsequently to enter The further treatment of row is quoted.Text based data may include Email, document, webpage, figure, spreadsheet, Call center's daily record, transaction reporting etc..

So, the various data of collection are optionally configured as the input data of machine learning task.

Reference picture 4, in the step s 100, the figure for configuring machine learning tasks is shown to user from display device 100 Shape interface, and the input operation that user is performed by graphical interfaces is detected by detection means 200, wherein, machine learning task is used In the data processing included by execution machine learning flow.As an example, machine learning task here may include data split, The data processings such as feature extraction, feature importance analysis, model training, model prediction, model evaluation.

Exemplary embodiment of the invention, display device 100 may be in response to the instruction of user to show for preparing The graphical interfaces of machine learning task.As an example, display device 100 may be in response to user to expect to set up machine learning flow Indicate and show the unified graphical interfaces for configuring each machine learning task, in the interface, can show for configuring The relevant range of machine learning task, for example, for enumerating the region of all configurable machine learning tasks, for showing currently The region of the machine learning flow that configuration is finished, the region for configuring current machine learning tasks etc., here, when user's selection During configuration particular machine learning tasks, some elements on interface can correspondingly change, for example, being related to the particular machine to learn to appoint The coherent element of the concrete configuration of business can be varied from interior perhaps display format.Here, it should be appreciated that display device 100 The mode at present graphical interface is not limited to above-mentioned example, for example, display device 100 may be in response to user expects configuration machine The instruction of learning tasks and show graphical interfaces corresponding with each machine learning task respectively.

As an example, display device 100 to the graphical interfaces that user shows can be for mainly by selection operation come The input selection type interface of machine learning task configuration is completed, or, described image interface can also can directly input generation The text editing interface of code or script.Above two graphical interfaces can switch mutually.

In the step s 100, selection operation that user performs on graphical interfaces, really can be also detected by detection means 200 Recognize the various input operations such as operation and text entry operation (for example, written in code operation).Here, detection means 200 can be combined Corresponding sensor device detects the various forms of operations that user is input into for graphical interfaces, for example, voice, posture, Action, touch, key entry etc..

These input operations are intended to the intention according to user to configure the corresponding machine study times among machine learning process Business, for example, the input data of configuration machine learning tasks, parameter, data processed result to input data execution data processing Output form etc..

Next, in step S300, being performed by the graphical interfaces according to the user for detecting by configuration device 300 Input operation configure the machine learning task.

Particularly, the user input as detected by detection means 200 is operated can be converted to accordingly by configuration device 300 Configuration-direct and/or configuration parameter, also, configuration device 300 can be according to these configuration-directs and/or configuration parameter come actual Configuration machine learning tasks.For example, in the example for configuring whole machine learning flow by directed acyclic graph (DAG), such as Fruit detection means 200 is detected user and new machine learning task is connected to the engineering for configuring finish before using connecting line After habit task (wherein, connecting line from it is described configure the machine learning task that finishes before and point to the new machine learning appoint Business), then configuration device 300 can appoint the new machine learning according to the connecting object of the connecting line for detecting and the direction of arrow It is engaged in being configured as the downstream machine learning tasks for configuring the machine learning task for finishing before.What is configured is specific interior Holding (for example, perform parameter etc.) can be as configuration device 300 according to the further operation of the user as detected by detection means 200 To perform.

Similarly, it is defeated for the particular user as detected by detection means 200 under the various interaction mechanisms for designing Enter operation, configuration device 300 can complete the configuration of machine learning task accordingly based upon the input operation for detecting.For example, If detected by detection means 200 is that user is directed to the data table name of each machine learning task input (for example, user It is input into the input data table of each machine learning task and the title of output data table), then configuration device 300 can be accordingly based upon Input data table name and output data table name are configured between upstream machines learning tasks and downstream machine learning tasks Annexation, that is to say, that in the input of output data table name and another machine learning task of certain machine learning task In the case that data table name is consistent, the former is configured as the upstream machines learning tasks of the latter, and correspondingly, the latter is configured as The former downstream machine learning tasks.

Exemplary embodiment of the invention, after the configuration for completing machine learning task, not such as prior art The machine learning task that middle general execution is configured, but it is unactual perform machine learning task in the case of, anticipation go out with The relevant data attribute information of the machine learning task, also, alternately, can be further along machine learning stream The data flow of journey and transmit the data attribute information that institute's anticipation goes out.

Particularly, in step S400, by apparatus for predicting 400 the machine learning task for not performing configuration feelings Under condition, the data attribute information relevant with the machine learning task is inferred, wherein, data attribute information includes data attribute word The title and/or data type of section.

As an example, in the machine learning flow of DAG forms, when configuration is over current machine learning tasks, inferring Device 400 can automatically be inferred to the data attribute information relevant with the current machine learning tasks for having configured, for example, apparatus for predicting 400 can obtain the configuration according to current machine learning tasks, it is contemplated that implementing result data (that is, current machine learning tasks Estimated output data) each attribute field title and/or data type.Hereafter, when user is initially configured current machine The downstream machine learning tasks of learning tasks are (for example, new machine learning task is connected to current machine learning tasks by user When afterwards), the data attribute information that apparatus for predicting 400 can will conclude that is delivered to downstream machine learning tasks, the transmission of this part The data attribute information for coming can with downstream machine learning tasks will the input data of actual treatment be integrated, it is also possible to by two Person separates.

As another example, in the machine learning flow of DAG forms, when configuration is over current machine learning tasks, Apparatus for predicting 400 can not infer data attribute information, but when user is initially configured the downstream machine of current machine learning tasks During device learning tasks (for example, when be connected to new machine learning task after current machine learning tasks by user, or, When user have chosen the new machine learning task as downstream machine learning tasks and be initially configured the new machine learning During task), apparatus for predicting 400 can automatically obtain the configuration according to the current machine learning tasks, it is contemplated that implementing result The title and/or data type of each attribute field of data (that is, the estimated output datas of current machine learning tasks), and The data attribute information transmission that current machine learning tasks will conclude that in the case of being connected with each other with downstream machine learning tasks To downstream machine learning tasks.

In addition to inferring data attribute information automatically according to the configuration of upstream and downstream machine learning task needs, in this hair In bright exemplary embodiment, can also be indicated to perform inference operations according to the deduction of user, i.e. can additionally be provided for starting The input medium of inference operations so that apparatus for predicting 400 indicates to start inference process according to the deduction of user input.

Apparatus for predicting 400 can come to infer the data attribute relevant with machine learning task in advance in any suitable manner Information so that the data attribute information being inferred to can help to user and carry out follow-up operation for machine learning flow.

As an example, apparatus for predicting 400 can in itself obtain the data of correlation based on the input data of machine learning task Attribute information.For example, apparatus for predicting 400 can directly using the deduction data attribute information of upstream machines learning tasks as downstream machine The deduction data attribute information of device learning tasks and pass to downstream machine learning tasks, without consider upstream machines learning tasks Actual process, i.e. the transparent transmission of data attribute information is carried out between upstream and downstream machine learning task.

Used as another example, apparatus for predicting 400 can be inferred to and machine with reference to the treatment of the real data of machine learning tasks The relevant data attribute information of the intermediate processing data and/or output data of device learning tasks.For example, apparatus for predicting 400 can lead to Cross the execute instruction of explaining the machine learning task and/or by among the input data of the machine learning task The data from the sample survey of extraction performs the execute instruction to infer the intermediate processing data and/or output number of the machine learning task According to data attribute information.

Particularly, used as executable entity, its configuration information is used to draw the machine learning task machine learning task Execute instruction, wherein, the execute instruction has been explicitly indicated machine learning task upon execution will be for which kind of input data Perform which kind of data processing and export which kind of output data etc..Correspondingly, apparatus for predicting 400 carries out semanteme by execute instruction Explain then it can be inferred that the data attribute information of the intermediate processing data of machine learning task and/or output data, for example, output The field name and/or data of the intermediate processing data before the field name and/or data type of data, and/or format conversion Type.

In addition to the mode of interpretative order, apparatus for predicting 400 can also perform configuration by a small amount of data from the sample survey is actual Good execute instruction is inferred to the intermediate processing data of machine learning task and/or the data attribute information of output data.This In, apparatus for predicting 400 can not perform the treatment that any instruction is explained, but be extracted from pending input data a small amount of Data, and the data execution actual treatment to extracting.Corresponding result can be used to reflect the middle of machine learning task Reason data and/or output data, the data attribute information of these data can be used as the data attribute information being inferred to.

It should be noted that the various modes for inferring data attribute information may also be combined with and use, i.e. apparatus for predicting 400 can be according to pre- First setting or the type of machine learning task optionally take different deduction modes.

As described above, exemplary embodiment of the invention, can be in the unactual machine learning task for performing configuration In the case of, the data attribute information related to the machine learning is previously obtained, these data attribute informations may include data word The title and/or data type of section, so as to can be applied to follow-up machine learning task, for example, can be used to help user configuring The input data of follow-up machine learning task, or, can be used for the configuration for helping verify follow-up machine learning task, or, can It is used to help the displaying of the output data of follow-up machine learning task.It should be understood that above-mentioned application scenarios are only as an example, any energy It is enough effectively to may be applicable to exemplary embodiment of the invention using the mode of the data attribute information being inferred to.

Fig. 5 shows the flow chart of the method for the execution machine learning flow according to another exemplary embodiment of the present invention. In method shown in Fig. 5, in addition to above-mentioned steps S100, step S300 and step S400, also including step S450.Wherein, Step S100, step S300 and step S400 can be operated according to the mode in the method shown in Fig. 4, and in step S450 In, the data attribute information that can be inferred to by apparatus for predicting 400 to user's displaying from display device 100.

As an example it is supposed that being the estimated of machine learning task by the data attribute information that apparatus for predicting 400 is inferred to Output result data attribute information, and the output result can as the input data of downstream machine learning tasks come Source.In this case, display device 100 can include the data attribute information being waited in the input of downstream machine learning tasks Favored area, for example, the data can be shown in the combobox of the input data field for configuring downstream machine learning tasks Field name in attribute information, so, user can configure downstream machine by selecting corresponding field from combobox The input data of habit task.

As another example, it is assumed that the data attribute information being inferred to can help to the data of downstream machine learning tasks The bandwagon effect of result.Particularly, in the machine learning task such as feature extraction, data processing can be related to Hash (hash) result of the treatment such as conversion and data processing can have the form of the readable difference such as key-value.In this feelings Under condition, can be using the relevant primitive attribute field name of data record and/or data type as the data attribute information being inferred to Downstream machine learning tasks are transparent to from upstream machines learning tasks, also, these data attribute informations are also exposed to user. Correspondingly, when downstream machine learning task is related to the output of data processed result (for example, work as to need to show model pre-estimating result During to user), user can choose the original word together exported with model pre-estimating result from the data attribute information for showing Section, to improve the readability of model pre-estimating result.

It should be noted that these are only that data attribute information of the explanation to being inferred to is shown to be applied to follow-up engineering The example of habit task, and exemplary embodiment of the invention is not limited to this.The data attribute information being inferred to is except being opened up Outside showing, it may also be used for the configuration to downstream machine learning tasks is verified.

Fig. 6 shows the flow chart of the method for the execution machine learning flow according to another exemplary embodiment of the present invention. In method shown in Fig. 6, in addition to above-mentioned steps S100, step S300 and step S400, also including step S100 ', step S300 ' and step S600.Wherein, step S100, step S300 and step S400 can be according to the modes in the method shown in Fig. 4 Operated.

Configuration and its data of machine learning task are completed by performing step S100, step S300 and step S400 After the deduction of attribute information, user can continue downstream machine of the configuration with the machine learning task as upstream machines learning tasks Learning tasks.Here, methods described can complete the configuration of downstream machine learning tasks by step S100 ' and step S300 ', Particularly, step S100 ' and step S300 ' is similar with step S100 and step S300, and simply targeted machine learning is appointed Business difference (that is, step S100 ' and step S300 ' targeted be downstream machine learning tasks).Correspondingly, in step In S100 ', display device 100 shows the graphical interfaces for configuring downstream machine learning tasks to user, and detection means 200 The input operation that detection user is performed by the graphical interfaces；Next, in step S300 ', configuration device 300 is according to inspection The user for measuring configures the downstream machine learning tasks by the input operation that the graphical interfaces is performed.Here, in step In rapid S100 ' user interface of display can be generally identical with the user interface for showing in the step s 100 and simply be related to match somebody with somebody Put otherwise varied in the details of item, or, in the step S100 ' user interface of display also can with show in the step s 100 User interface is entirely different.Additionally, alternately, in step S300 ', display device 100 can also be in graphical interfaces The data attribute information of the upstream machines learning tasks that step S300 is inferred to is illustrated in user so that user can be based on displaying Data attribute information configure the downstream machine learning tasks.

After the configuration for completing downstream machine learning tasks, exemplary embodiment of the invention, in step In S600, the data relevant with upstream machines learning tasks for being based on being inferred to by apparatus for predicting 400 by configuration device 300 belong to Property information checks the configuration of the downstream machine learning tasks.Particularly, because data attribute information can relate to upstream machine The associated data attributes field name and/or corresponding data type of device learning tasks, therefore, configuration device 300 can be verified Whether the relevant configuration (for example, input data field, operational parameter, arithmetic type etc.) of downstream machine learning tasks meets by upper The data source that trip machine learning task is provided.

Alternatively, downstream machine is determined when configuration device 300 is based on the data attribute information of upstream machines learning tasks When the configuration of learning tasks does not meet the data source of upstream machines learning tasks offer, display device 100 will can be alerted accordingly Message is displayed on screen, and the configuration for reminding user's downstream machine learning tasks has problem.As an example, alert message In can indicate error configuration item and/or error details.

It should be noted that exemplary embodiment of the invention, for one or more machine learning tasks that configuration is finished, The machine learning task can be performed according to the instruction of user.That is, in the method for above-mentioned execution machine learning flow In, also including step：Execution according to user indicates to perform the machine learning task of one or more configurations.Here, user One or more the machine learning tasks for having configured can be started by default button or other means, these machine learning are appointed Business may make up whole machine learning flow or a part therein.

Hereinafter, reference picture 7A to Fig. 7 F are described to be performed in machine learning platform according to an exemplary embodiment of the present invention The example at the interface of machine learning flow.In the example, machine learning process is configured according to the form of DAG, however, should Understand：It is of the invention exemplary intuitively to explain that ins and outs with reference to described by Fig. 7 A to Fig. 7 F are merely possible to example Embodiment, not for the scope for limiting illustrative examples of the present invention.

Reference picture 7A, it illustrates the graphical interfaces for configuring machine learning tasks, the zone line of the graphical interfaces It is the DAG region of machine learning flow, left side lists optional machine learning task, and right side is for configuring particular machine The region of habit task.In the graphical interfaces, user can to configure, " data be torn open by the operation such as clicking on, pulling, key in Point ", for example, configuration fractionation mode and primary contract etc..As shown in Figure 7 A, " import bank data source (" bank ") " this After machine learning task, user can carry out the configuration of " data fractionation " this machine learning task by right side area, Based on the user for detecting after the configuration that the input operation of right side area completes " data fractionation ", can be based on " data are torn open Point " configuration be inferred to corresponding data attribute information.Next, user selects to continue by " SQL " on the right side of click Configure next machine learning task " SQL ".

Reference picture 7B, user can specifically be configured by clicking on " configuration " icon in right side area to " SQL ".Phase Ying Di, after user clicks on above-mentioned icon, can show interface as seen in figure 7 c.Here, it should be noted that in this example, only There is the downstream machine that the configurable summit for representing the upstream machines learning tasks that configuration is finished in user's connection will configure with representative During the configurable summit of device learning tasks, the data attribute information of the upstream machines learning tasks that will can be just inferred to is passed to Existing machine learning task, and line is not yet carried out and " SQL " the two summits between in " data fractionation " due to user, therefore User's " temporarily without input source schema, input source please be connect " is reminded in the interface of Fig. 7 C, here, schema is used as data attribute The specific example of information.

Therefore, as illustrated in fig. 7d, user can using " SQL " as downstream machine learning tasks be connected to upstream " data are torn open Point " so that the data attribute information relevant with " data fractionation " being inferred to before can be delivered to " SQL ".Correspondingly, when with When " configuration " icon is clicked at family, interface as seen in figure 7e can be shown, wherein, list the output as " data fractionation " task Whole field names of data so that user can refer to these data attribute informations to complete writing for script.

Exemplary embodiment of the invention, can also be based on the associated data attributes information of " data fractionation " (for example, word Name section) check whether configuration of the user to SQL meets specification.Reference picture 7F, " counts when being occurred in that in the script that user writes According to splitting " field name (for example, age1) for not providing as data source when, can be to user's displaying alert message " field Age 1is not found, field does not exist, please change ".

The method that execution machine learning flow according to an exemplary embodiment of the present invention is described above by reference to Fig. 2 to Fig. 7 F With system and corresponding machine learning platform application example.It should be understood that the method for above-mentioned execution machine learning flow can pass through The program in computer-readable media is recorded to realize, correspondingly, exemplary embodiment of the invention, it is possible to provide Yi Zhongyong In the computer-readable medium for performing machine learning flow, being recorded on the computer-readable medium has for performing with lower section The computer program of method step：(A) show the graphical interfaces for configuring machine learning tasks to user and detect that user passes through The input operation that graphical interfaces is performed, wherein, machine learning task is used to perform the data processing included by machine learning flow； (B) the machine learning task is configured by the input operation that the graphical interfaces is performed according to the user for detecting；And (C) in the case of the machine learning task for not performing configuration, the data category relevant with the machine learning task is inferred Property information, wherein, the title and/or data type of data attribute information including data community.

Computer program in above computer computer-readable recording medium can be in client, main frame, agent apparatus, server etc. In computer equipment dispose environment in run, it should be noted that the computer program can be additionally used in perform except above-mentioned steps with Outer additional step or performed when above-mentioned steps are performed more specifically is processed, these additional steps and further treatment Reference picture 2 is described content to Fig. 7 F, will no longer be repeated herein for repetition is avoided.

Correspondingly, the system of above-mentioned execution machine learning flow can also be completely dependent on the operation of computer program to realize phase The function of answering, i.e. each device is corresponding to each step in the function structure of computer program so that whole system is by special Software kit (for example, lib storehouses) and be called, to realize corresponding function.

On the other hand, each device and unshowned relevant apparatus shown in Fig. 2 and Fig. 3 can also be by hardware, soft Part, firmware, middleware, microcode or its any combination are realized.When being realized with software, firmware, middleware or microcode, use Can be stored in the computer-readable medium of such as storage medium in the program code or code segment for performing corresponding operating, made Processor by reading and can run corresponding program code or code segment and perform corresponding operation.

Here, exemplary embodiment of the invention is also implemented as computing device, and the computing device includes memory unit And processor, the set of computer-executable instructions that is stored with memory unit is closed, when the set of computer-executable instructions is closed by institute When stating computing device, the method for performing above-mentioned execution machine learning flow.

Particularly, the computing device can be deployed in server or client, it is also possible to be deployed in distributed network On node apparatus in network environment.Additionally, the computing device can be PC computers, board device, personal digital assistant, intelligence Can mobile phone, web applications or other be able to carry out the device of above-mentioned instruction set.

Here, the computing device is not necessarily single computing device, can also be it is any can be alone or in combination Perform the device of above-mentioned instruction (or instruction set) or the aggregate of circuit.Computing device can also be integrated control system or system A part for manager, or can be configured as with Local or Remote (for example, via be wirelessly transferred) with the portable of interface inter-link Formula electronic installation.

In the computing device, processor may include central processing unit (CPU), graphic process unit (GPU), may be programmed and patrol Collect device, dedicated processor systems, microcontroller or microprocessor.Unrestricted as an example, processor may also include simulation Processor, digital processing unit, microprocessor, polycaryon processor, processor array, network processing unit etc..

It is above-mentioned on perform machine learning flow method described in some operations can be realized by software mode, Some operations can be realized by hardware mode, additionally, can also realize that these are operated by way of software and hardware combining.

Processor can run instruction of the storage in one of memory unit or code, wherein, the memory unit can be with Data storage.Instruction and data can also pass through network via Network Interface Unit and be sent and received, wherein, the network connects Mouth device can use any of host-host protocol.

Memory unit can be integral to the processor and be integrated, for example, RAM or flash memory are arranged in into integrated circuit microprocessor etc. Within.Additionally, memory unit may include independent device, such as, outside dish driving, storage array or any Database Systems can Other storage devices for using.Memory unit and processor can be coupled operationally, or can for example pass through I/O ports, Network connection etc. is communicated so that processor can read file of the storage in memory unit.

Additionally, the computing device may also include video display (such as, liquid crystal display) and user mutual interface is (all Such as, keyboard, mouse, touch input device etc.).The all component of computing device can be connected to each other via bus and/or network.

It is above-mentioned on perform machine learning flow method involved by operation can be described as it is various interconnection or couple Functional block or function diagram.However, these functional blocks or function diagram can be equably integrated into single logic device or by Operated according to non-definite border.

Particularly, as described above, the computing device for performing machine learning flow according to an exemplary embodiment of the present invention Memory unit and processor are may include, the set of computer-executable instructions that is stored with memory unit is closed, when the computer can be held When row instruction set is by the computing device, following step is performed：(A) show for configuring machine learning tasks to user Graphical interfaces simultaneously detects the input operation that user is performed by graphical interfaces, wherein, machine learning task is used to perform engineering Practise the data processing included by flow；(B) matched somebody with somebody by the input operation that the graphical interfaces is performed according to the user for detecting Put the machine learning task；And (C) is in the case of the machine learning task for not performing configuration, infer and the machine The relevant data attribute information of device learning tasks, wherein, data attribute information includes the title and/or data of data community Type.

It should be noted that combined Fig. 2 to Fig. 7 F describe execution engineering according to an exemplary embodiment of the present invention above Each treatment details of the method for flow is practised, treatment details during computing device each step of execution is will not be described in great detail here.

Be described above each exemplary embodiment of the invention, it should be appreciated that foregoing description be only it is exemplary, not Exhaustive, and present invention is also not necessarily limited to disclosed each exemplary embodiment.Without departing from scope and spirit of the present invention In the case of, many modifications and changes will be apparent from for those skilled in the art.Therefore, originally The protection domain of invention should be defined by the scope of claim.

Claims

1. it is a kind of perform machine learning flow method, including：

(A) show graphical interfaces for configuring machine learning tasks to user and detect user by graphical interfaces perform it is defeated Enter operation, wherein, machine learning task is used to perform the data processing included by machine learning flow；

(B) the machine learning task is configured by the input operation that the graphical interfaces is performed according to the user for detecting； And

(C) in the case of the machine learning task for not performing configuration, the number relevant with the machine learning task is inferred According to attribute information, wherein, data attribute information includes the title and/or data type of data community.

2. the method for claim 1, also includes：(D) it is illustrated in the data attribute letter that step (C) is inferred to user Breath.

3. the method for claim 1, wherein in step (C), the data attribute information being inferred to is the engineering The data attribute information of the input data, output data and/or intermediate processing data of habit task.

4. the method as described in claim 1 or 3, also includes：

(E) show to user and appoint for the downstream machine study configured with the machine learning task as upstream machines learning tasks The graphical interfaces of business simultaneously detects the input operation that user is performed by the graphical interfaces；

(F) the downstream machine study times is configured by the input operation that the graphical interfaces is performed according to the user for detecting Business；

(G) configuration of the downstream machine learning tasks is checked based on the data attribute information being inferred in step (C).

5. the configuration the method for claim 1, wherein in response to the machine learning task terminates to perform step automatically Suddenly (C), or, in response to the configuration with the machine learning task as the downstream machine learning tasks of upstream machines learning tasks Start to perform step (C) automatically, or, the deduction in response to user indicates to perform step (C).

6. method as claimed in claim 2, wherein, machine learning task is implemented as the configurable top in directed acyclic graph Point, wherein, the configuration in response to the machine learning task terminates to perform step (C) automatically, also, connects representative in user The configurable summit of the machine learning task of configuration is with to represent with the machine learning task be upstream machines learning tasks Downstream machine learning tasks configurable summit when perform step (D) automatically.

7. method as claimed in claim 3, wherein, in step (C), referred to by the execution for explaining the machine learning task Make and/or perform the execute instruction by for the data from the sample survey extracted among the input data of the machine learning task To infer the intermediate processing data of the machine learning task and/or the data attribute information of output data.

8. it is a kind of perform machine learning flow system, including：

Display device, for showing the graphical interfaces for configuring machine learning tasks to user, wherein, machine learning task is used In the data processing included by execution machine learning flow；

Detection means, for detecting the input operation that user is performed by graphical interfaces；

Configuration device, for configuring the machine by the input operation that the graphical interfaces is performed according to the user for detecting Learning tasks；And

Apparatus for predicting, in the case of the machine learning task for not performing configuration, inferring and appointing with the machine learning The relevant data attribute information of business, wherein, data attribute information includes the title and/or data type of data community.

9. a kind of computer-readable medium for performing machine learning flow, wherein, remember on the computer-readable medium Record has the computer program for performing following steps：

10. a kind of computing device for performing machine learning flow, including memory unit and processor, be stored with meter in memory unit Calculation machine executable instruction set, when the set of computer-executable instructions is closed by the computing device, performs following step：