CN107273979A - The method and system of machine learning prediction are performed based on service class - Google Patents

The method and system of machine learning prediction are performed based on service class Download PDF

Info

Publication number
CN107273979A
CN107273979A CN201710427869.8A CN201710427869A CN107273979A CN 107273979 A CN107273979 A CN 107273979A CN 201710427869 A CN201710427869 A CN 201710427869A CN 107273979 A CN107273979 A CN 107273979A
Authority
CN
China
Prior art keywords
machine learning
learning model
service class
subset
training
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Granted
Application number
CN201710427869.8A
Other languages
Chinese (zh)
Other versions
CN107273979B (en
Inventor
陈雨强
戴文渊
杨强
罗远飞
涂威威
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
4Paradigm Beijing Technology Co Ltd
Original Assignee
4Paradigm Beijing Technology Co Ltd
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by 4Paradigm Beijing Technology Co Ltd filed Critical 4Paradigm Beijing Technology Co Ltd
Priority to CN201710427869.8A priority Critical patent/CN107273979B/en
Publication of CN107273979A publication Critical patent/CN107273979A/en
Application granted granted Critical
Publication of CN107273979B publication Critical patent/CN107273979B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N20/00—Machine learning
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q10/00—Administration; Management
    • G06Q10/04—Forecasting or optimisation specially adapted for administrative or management purposes, e.g. linear programming or "cutting stock problem"

Landscapes

  • Engineering & Computer Science (AREA)
  • Business, Economics & Management (AREA)
  • Theoretical Computer Science (AREA)
  • Software Systems (AREA)
  • General Physics & Mathematics (AREA)
  • Strategic Management (AREA)
  • Human Resources & Organizations (AREA)
  • Physics & Mathematics (AREA)
  • Economics (AREA)
  • Development Economics (AREA)
  • Medical Informatics (AREA)
  • Mathematical Physics (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Artificial Intelligence (AREA)
  • Evolutionary Computation (AREA)
  • Computing Systems (AREA)
  • Game Theory and Decision Science (AREA)
  • Data Mining & Analysis (AREA)
  • General Engineering & Computer Science (AREA)
  • Entrepreneurship & Innovation (AREA)
  • Marketing (AREA)
  • Operations Research (AREA)
  • Quality & Reliability (AREA)
  • Tourism & Hospitality (AREA)
  • General Business, Economics & Management (AREA)
  • Management, Administration, Business Operations System, And Electronic Commerce (AREA)

Abstract

There is provided it is a kind of based on service class come perform machine learning prediction method and system, including：(a) prediction data record is obtained；(b) forecast sample of machine learning model corresponding with service class is generated based on the attribute information of prediction data record, wherein, the forecast sample of basic machine learning model includes essential characteristic subset, or, the forecast sample of reinforcement machine learning model includes essential characteristic subset and at least one supplementary features subset；(c) forecast sample is supplied to machine learning model, to obtain predicting the outcome for forecast sample, wherein, reinforcement machine learning model includes basic machine learning model and at least one additional submodel identical with basic machine learning model type and trained according to lift frame.Machine learning is realized due to generating corresponding machine learning sample for service class, and then according to corresponding characteristic Design and model framework, therefore, machine learning service flexibly can be effectively provided.

Description

The method and system of machine learning prediction are performed based on service class
Technical field
The exemplary embodiment all things considered of the present invention is related to artificial intelligence field, is based on more specifically to one kind Service class come perform machine learning prediction method and system and it is a kind of based on service class come training machine learning model Method and system.
Background technology
With the appearance of mass data, artificial intelligence technology is developed rapidly, in order to be excavated from mass data Value, machine learning techniques are applied in the concrete scene in the various fields such as internet, finance, security protection.
In practice, when providing the related service of machine learning application, this can be weighed in terms of any one or more The quality of kind of service, such as it is, the accuracy of machine learning model prediction, stability, ageing, resource-consuming etc..With clothes The related factor of quality of being engaged in is a lot, and the relation between each factor is also more complicated, generally requires to consider each factor, example Such as, the model algorithm of machine learning model, related data scale, available computing resource etc..
, it is necessary to produce the training and/or prediction suitable for machine learning based on data record in machine learning techniques Sample.Here, per data, record can be seen as the description as described in an event or object, corresponding to an example or sample. In data record, include each item of the performance or property of reflection event or object in terms of certain, these items can be described as " attribute ".The processing such as Feature Engineering is carried out by the attribute information to data record, the machine including various features can be produced Device learning sample.
The attribute information of data record respectively has feature in terms of form or implication, correspondingly, and produced feature also can There is each species diversity in terms of form or implication.This species diversity can directly influence the service quality of machine learning, but skill Art personnel are but difficult to effectively hold or utilize this influence.
Therefore, how to provide machine learning service turns into the technical problem that this area is paid close attention to efficient, flexible.
The content of the invention
The exemplary embodiment of the present invention is intended to overcome existing machine learning model to provide engineering with being difficult to efficient, flexible The problem of acclimatization is engaged in.
According to the present invention exemplary embodiment there is provided it is a kind of based on service class come perform machine learning prediction side Method, including：(a) prediction data record is obtained；(b) attribute information recorded based on prediction data is corresponding with service class to generate Machine learning model forecast sample, wherein, basic machine learning corresponding with the basic level of service among service class The forecast sample of model includes essential characteristic subset, or, enhancing machine corresponding with the enhancing service class among service class The forecast sample of device learning model includes essential characteristic subset and at least one supplementary features subset；And (c) is by forecast sample Machine learning model corresponding with service class is supplied to, to obtain predicting the outcome for forecast sample, wherein, strengthen machine Learning model includes basic machine learning model and identical with basic machine learning model type and trained according to lift frame At least one additional submodel, wherein, basic machine learning model correspond to essential characteristic subset, additional submodel pair Should be in supplementary features subset.
Alternatively, in the process, machine learning model corresponding with service class is to be in advance based on the seeervice level The unique machine learning model not trained.
Alternatively, in the process, machine learning model corresponding with service class is from being in advance based on multiple services The machine learning model corresponding with the service class chosen among multiple machine learning models that rank is trained.
Alternatively, in the process, service class is used at least one the aspect correlation for weighing machine learning service.
Alternatively, in the process, the model algorithm of service class and machine learning model, data scale and/or meter Calculate resource related.
Alternatively, in the process, by being chosen and the service class pair by user determines the service class The machine learning model answered；Or, machine corresponding with the service class is chosen by automatically determining the service class Learning model.
Alternatively, in the process, supplementary features are based on essential characteristic and produced.
In accordance with an alternative illustrative embodiment of the present invention there is provided it is a kind of based on service class come perform machine learning prediction Medium, wherein, record has the computer program for performing any of the above-described method on the computer-readable medium.
In accordance with an alternative illustrative embodiment of the present invention there is provided it is a kind of based on service class come perform machine learning prediction Computing device, including memory unit and processor, wherein, the set of computer-executable instructions that is stored with memory unit is closed, and works as institute When stating set of computer-executable instructions conjunction by the computing device, any of the above-described method is performed.
In accordance with an alternative illustrative embodiment of the present invention there is provided it is a kind of based on service class come training machine learning model Method, including：(A) training data record is obtained；(B) generated and service class pair based on the attribute information of training data record The training sample for the machine learning model answered, wherein, basic engineering corresponding with the basic level of service among service class Practising the training sample of model includes essential characteristic subset, or, enhancing corresponding with the enhancing service class among service class The training sample of machine learning model includes essential characteristic subset and at least one supplementary features subset；And (C) utilizes generation Training sample train machine learning model corresponding with service class, wherein, reinforcement machine learning model includes basic machine Device learning model and it is identical with basic machine learning model type and according to lift frame train at least one add Submodel, wherein, basic machine learning model corresponds to essential characteristic subset, and additional submodel corresponds to supplementary features subset.
Alternatively, in the process, methods described is performed for the service class selected among multiple service class, To obtain unique machine learning model.
Alternatively, in the process, performed respectively for each service class among multiple service class described Method, to obtain multiple machine learning models.
Alternatively, in the process, in step (C), in the case where training reinforcement machine learning model, by solid The fixed basic machine learning model wherein trained and additional submodel train remaining additional submodel successively.
Alternatively, in the process, service class is used at least one aspect for weighing machine learning service.
Alternatively, in the process, the model algorithm of service class and machine learning model, data scale and/or meter Calculate resource related.
Alternatively, in the process, supplementary features are based on essential characteristic and produced.
Alternatively, in the process, basic machine learning model and each additional submodel are based respectively on identical or not Same training data record training is formed.
In accordance with an alternative illustrative embodiment of the present invention there is provided it is a kind of based on service class come training machine learning model Medium, wherein, record has the computer program for performing any of the above-described method on the computer-readable medium.
In accordance with an alternative illustrative embodiment of the present invention there is provided it is a kind of based on service class come training machine learning model Computing device, including memory unit and processor, wherein, the set of computer-executable instructions that is stored with memory unit is closed, and works as institute When stating set of computer-executable instructions conjunction by the computing device, any of the above-described method is performed.
In accordance with an alternative illustrative embodiment of the present invention there is provided it is a kind of based on service class come perform machine learning prediction System, including：Prediction data records acquisition device, for obtaining prediction data record；Forecast sample generation device, for based on The attribute information of prediction data record generates the forecast sample of machine learning model corresponding with service class, wherein, with clothes The forecast sample of the corresponding basic machine learning model of basic level of service among rank of being engaged in includes essential characteristic subset, or Person, the forecast sample of reinforcement machine learning model corresponding with the enhancing service class among service class includes essential characteristic Collection and at least one supplementary features subset；And prediction meanss, for forecast sample to be supplied into machine corresponding with service class Device learning model, to obtain predicting the outcome for forecast sample, wherein, reinforcement machine learning model includes basic machine learning Model and at least one additional submodel identical with basic machine learning model type and trained according to lift frame, Wherein, basic machine learning model corresponds to essential characteristic subset, and additional submodel corresponds to supplementary features subset.
Alternatively, in the system, machine learning model corresponding with service class is to be in advance based on the seeervice level The unique machine learning model not trained.
Alternatively, in the system, machine learning model corresponding with service class is from being in advance based on multiple services The machine learning model corresponding with the service class chosen among multiple machine learning models that rank is trained.
Alternatively, in the system, service class is used at least one the aspect correlation for weighing machine learning service.
Alternatively, in the system, the model algorithm of service class and machine learning model, data scale and/or meter Calculate resource related.
Alternatively, in the system, by being chosen and the service class pair by user determines the service class The machine learning model answered；Or, machine corresponding with the service class is chosen by automatically determining the service class Learning model.
Alternatively, in the system, supplementary features are based on essential characteristic and produced.
In accordance with an alternative illustrative embodiment of the present invention there is provided it is a kind of based on service class come training machine learning model System, including：Training data records acquisition device, for obtaining training data record；Training sample generation device, for based on The attribute information of training data record generates the training sample of machine learning model corresponding with service class, wherein, with clothes The training sample of the corresponding basic machine learning model of basic level of service among rank of being engaged in includes essential characteristic subset, or Person, the training sample of reinforcement machine learning model corresponding with the enhancing service class among service class includes essential characteristic Collection and at least one supplementary features subset；And trainer, trained for the training sample using generation and service class Corresponding machine learning model, wherein, reinforcement machine learning model include basic machine learning model and with basic engineering Habit types of models is identical and according at least one additional submodel of lift frame training, wherein, basic machine learning mould Type corresponds to essential characteristic subset, and additional submodel corresponds to supplementary features subset.
Alternatively, in the system, the system is performed for the service class selected among multiple service class Processing, to obtain unique machine learning model.
Alternatively, in the system, the system is distinguished for each service class among multiple service class Processing is performed, to obtain multiple machine learning models.
Alternatively, in the system, trainer is in the case where training reinforcement machine learning model, by fixing it In the basic machine learning model that has trained and additional submodel train remaining additional submodel successively.
Alternatively, in the system, service class is used at least one aspect for weighing machine learning service.
Alternatively, in the system, the model algorithm of service class and machine learning model, data scale and/or meter Calculate resource related.
Alternatively, in the system, supplementary features are based on essential characteristic and produced.
Alternatively, in the system, basic machine learning model and each additional submodel are based respectively on identical or not Same training data record training is formed.
Machine learning prediction and/or training machine performed based on service class according to an exemplary embodiment of the present invention In the method and system of learning model, corresponding machine learning sample is generated for service class, and then according to corresponding spy Levy design and model framework to realize machine learning, so as to flexibly effectively provide machine learning service.
Brief description of the drawings
From detailed description below in conjunction with the accompanying drawings to the embodiment of the present invention, these and/or other aspect of the invention and Advantage will become clearer and be easier to understand, wherein：
Fig. 1 shows the system according to an exemplary embodiment of the present invention for performing machine learning prediction based on service class Block diagram；
Fig. 2 shows the method according to an exemplary embodiment of the present invention for performing machine learning prediction based on service class Flow chart；
Fig. 3 show it is according to an exemplary embodiment of the present invention based on service class come the system of training machine learning model Block diagram；And
Fig. 4 show it is according to an exemplary embodiment of the present invention based on service class come the method for training machine learning model Flow chart.
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.
Machine learning is the inevitable outcome that artificial intelligence study develops into certain phase, and it is directed to the hand by calculating Section, improves the performance of system itself using experience.In computer systems, " experience " generally exists in " data " form, leads to Machine learning algorithm is crossed, " model " can be produced from data, that is to say, that empirical data is supplied to machine learning algorithm, just Model can be produced based on these empirical datas, when in face of news, model can provide corresponding judgement, i.e. predict the outcome. Machine learning can be implemented as the form of " supervised learning ", " unsupervised learning " or " semi-supervised learning ".According to the present invention's Exemplary embodiment, the process related to machine learning application scenarios is (for example, training machine learning model, offer machine learning Predict the outcome, receive machine learning and the process such as predict the outcome) on the whole can as one or more machine learning services, here, The machine learning service can be provided online, can also be provided offline.It should be noted that the exemplary embodiment of the present invention is being trained and answered During with machine learning model, also using statistic algorithm, business rule and/or expertise etc., further to improve The effect of machine learning.
Particularly, exemplary embodiment of the invention is related to trains and/or utilizes engineering in machine learning service Model is practised, wherein, based on service class come processing attribute information to generate each character subset, and then based on corresponding lift frame Carry out training machine learning model or provide service using machine learning model.Here, service class is used to weigh engineering acclimatization At least one aspect of business, for example, accuracy, stability, ageing, resource-consuming etc..As an example, service class can With related to the factor such as the model algorithm, data scale and/or computing resource of machine learning model.According to the exemplary of the present invention Embodiment, after service class is set, may correspondingly determine that the composition submodel of machine learning model and corresponding spy Levy subset.Here, the specific dividing mode of service class is not limited, it is any can to service quality carry out area it is equal otherwise It can be applied to the exemplary embodiment of the present invention.
Fig. 1 shows the system according to an exemplary embodiment of the present invention for performing machine learning prediction based on service class Block diagram.Particularly, the forecasting system can be used for being directed to forecast sample, and its pass is provided using corresponding machine learning model In predicting the outcome for specific transactions problem (that is, predicting target), wherein, the machine learning model corresponds to service class and had Standby corresponding one or more submodels, i.e. basic machine learning model or additional submodel, these submodel types it is identical and Follow lift frame (for example, gradient lift frame etc.).
Here, the basic machine learning model or additional submodel for constituting machine learning model are unrestricted on particular type System, any types of models that composite construction can be trained for according to lift frame can be as according to an exemplary embodiment of the present Submodel.For example, machine learning model and additional submodel can be linear models (for example, logarithm probability returns mould substantially Type etc.).
As described above, according to the exemplary embodiment of the present invention, machine learning model corresponds to service class in itself, specifically In fact, for special services rank, will using corresponding machine learning model come perform prediction, wherein, the machine learning mould Type has one or more submodels based on lift frame.It should be understood that machine learning model corresponding from different service class Had differences in terms of the number or the corresponding character subset of each submodel of submodel, in this way, can be effective The machine learning service of various service class is neatly provided.
System shown in Fig. 1 can be realized all by computer program with software mode, can also be filled by special hardware Put to realize, can also be realized by way of software and hardware combining.Correspondingly, each device of the system shown in composition Fig. 1 can To be to only rely on computer program to realize the virtual module of corresponding function or realize the work(by hardware configuration The universal or special device of energy, can also be that operation has processor of corresponding computer program etc..
Shown in Fig. 1, prediction data record acquisition device 100 is used to obtain prediction data record.These prediction data are recorded Can in any way it be produced by any side, for example, it may be the online data for generating or collecting, the number for previously generating or storing According to, can also be data from external reception.The attribute information of these data can relate to customer information, for example, identity, educational background, The information such as occupation, assets, contact method.Or, the attribute information of these data can also refer to the information of business relevant item, example Such as, on information such as the turnover of deal contract, both parties, subject matter, locos.It should be noted that the present invention's is exemplary The attribute for the data mentioned in embodiment can relate to the performance or property of any object or affairs in terms of certain, and be not limited to individual People, object, tissue, unit, mechanism, project, event etc. are defined or described.In fact, any can be by carrying out to it The information data of machine learning can be applied to the exemplary embodiment of the present invention.
Prediction data record acquisition device 100 can obtain separate sources (for example, data from metadata provider, coming Come from the data of internet (for example, social network sites), the data from mobile operator, the data from APP operators, Data from express company, from data of credit institution etc.) structuring or unstructured data, for example, literary Notebook data or numeric data etc..These data can be input to prediction data record acquisition device 100, Huo Zheyou by input unit Prediction data record acquisition device 100 is automatically generated according to existing data, or can record acquisition device by prediction data 100 (for example, storage medium (for example, data warehouse) on the network) acquisitions from network, in addition, the mediant of such as server Prediction data, which is can help to, according to switch records acquisition device 100 from the corresponding data of external data source acquisition.Here, obtain The data conversion module such as the text analysis model that can be predicted in data record acquisition device 100 of data be converted to and be easily processed Form.It should be noted that prediction data record acquisition device 100 can be configured as being made up of software, hardware and/or firmware each Module, some of these modules module or whole modules can be integrated into one or common cooperation to complete specific function.
The attribute information that forecast sample generation device 200 is used to record based on prediction data is corresponding with service class to generate Machine learning model forecast sample, wherein, basic machine learning corresponding with the basic level of service among service class The forecast sample of model includes essential characteristic subset, or, enhancing machine corresponding with the enhancing service class among service class The forecast sample of device learning model includes essential characteristic subset and at least one supplementary features subset.
In forecasting system, service class can be determined in any suitable manner, as an example, service class can be with It is the preset value consistent with model training, for example, it is assumed that only obtaining engineering corresponding with special services rank in the training stage Model is practised, then in perform prediction, it is also desirable to generate corresponding forecast sample using the special services rank, and then obtain The machine learning model predicts the outcome.Correspondingly, in this case, the machine corresponding with service class in forecasting system Device learning model is to be in advance based on the unique machine learning model that the service class is trained.
As another example, service class can be independently determined in forecasting system, for example, it is assumed that obtained in the training stage with The corresponding multiple machine learning models of multiple service class, then it needs to be determined that machine learning mould actual use in forecasting system Which service class is type correspond to.That is, machine learning model corresponding with service class is from pre- in forecasting system One corresponding with the service class chosen among the multiple machine learning models first trained based on multiple service class Machine learning model., also can be according to as an example, it is contemplated that the factor such as environment of online service when choosing machine learning model Service class according to user's actual subscription etc..Here, for how to choose machine learning model, any appropriate side can be used Formula, for example, by choosing machine learning model corresponding with the service class by user determines the service class, as Example, user can be by the interactive interface of the software systems such as machine learning platform come specified services rank；Or, by automatic true Determine the service class to choose machine learning model corresponding with the service class, as an example, can be by considering Influence the factor (for example, prediction data record scale, computing resource, response time etc.) of service class suitable to automatically determine Service class.
Here, it is pre- produced by forecast sample generation device 200 depending on the service class corresponding to machine learning model Test sample originally can only include essential characteristic subset (in the case where service class is basic level of service)；Or, except substantially special Levy outside subset, forecast sample can further comprise that one or more supplementary features subsets (are enhancing seeervice level in service class In the case of other), it can be seen that the supplementary features subset in forecast sample corresponds to certain specific enhancing seeervice level on the whole Not, that is to say, that the supplementary features subset of the forecast sample under difference enhancing service class has differences on the whole, for example, attached Plus the quantity of character subset is different, the feature of at least a portion supplementary features subset and differ.
As an example, forecast sample generation device 200 can by being screened to the attribute information that prediction data is recorded, point Group or further additional treatments etc. and obtain multiple features, and obtain forecast sample by being divided to the multiple feature Essential characteristic subset and/or supplementary features subset (wherein, each feature can be divided into one or more subsets), this In, forecast sample is corresponding with prediction data record, is usually implemented as directly inputting for machine learning model.According to showing for the present invention Example property embodiment, forecast sample generation device 200 can generate character subset in any suitable fashion, for example, it is contemplated that The factors such as content, implication, value continuity, span, valued space scale, the Deletional, importance of attribute information, or Person, can be with reference to submodel feature in reinforcement machine learning model etc..It is available to be based on carrying by the design of supplementary features subset The reinforcement machine learning model that trains of framework is risen to provide various non-main levels other machine learning service.
As an example, forecast sample generation device 200 can be produced based on the essential characteristic in essential characteristic subset it is additional Supplementary features in character subset, that is to say, that supplementary features are based on essential characteristic and produced.For example, forecast sample produces dress The combination of essential characteristic can be regard as supplementary features by putting 200.Here, forecast sample generation device 200 can be by essential characteristic Carry out any appropriate change and bring to obtain supplementary features.Correspondingly, as supplementary features via additional submodel are incorporated into machine In device study, the rank of machine learning prediction service can be effectively influenceed.
Prediction meanss 300 are used to forecast sample being supplied to machine learning model corresponding with service class, to obtain pin Forecast sample is predicted the outcome.Here, each character subset of forecast sample can be provided correspondingly to machine by prediction meanss 300 Each submodel of device learning model, for example, essential characteristic subset is supplied into basic machine learning model, by supplementary features Collection is supplied to corresponding additional submodel.That is, basic machine learning model corresponds to essential characteristic subset, additional submodule Type corresponds to supplementary features subset.Especially, it is assumed that the machine learning model corresponding with service class is enhancing engineering Model is practised, then reinforcement machine learning model may include basic machine learning model and identical with basic machine learning model type And at least one the additional submodel trained according to lift frame.
Particularly, prediction meanss 300 can to machine learning model each submodel (that is, basic machine learning model Or additional submodel) corresponding character subset in forecast sample is provided respectively, here, any two submodel can be provided that completely The identical or entirely different character subset in identical, part.That is, each submodel of machine learning model is carried for it The character subset of confession is estimated to perform, correspondingly, can integrate the estimation results of all submodels and to obtain machine learning model whole Body acupuncture predicts the outcome to forecast sample.Especially, the discardable some character subsets of prediction meanss 300, i.e. not by these features Subset is supplied to corresponding submodel, thus causes corresponding submodel not work or only provide default default value.
Describe according to an exemplary embodiment of the present invention to perform machine learning based on service class hereinafter with reference to Fig. 2 The flow chart of the method for prediction.Here, as an example, method shown in Fig. 2 can be as shown in Figure 1 forecasting system perform, It can be realized, can also be performed by the computing device of particular configuration shown in Fig. 2 with software mode by computer program completely Method.
For convenience, it is assumed that the forecasting system of method as shown in Figure 1 shown in Fig. 2 is performed, as illustrated, in step In rapid S100, prediction data record is obtained by prediction data record acquisition device 100.
Here, as an example, every prediction data record may correspond to an item to be predicted on particular prediction problem (for example, event or object), correspondingly, prediction data record may include the performance or property of reflection event or object in terms of certain The various attribute informations of (that is, attribute)., can be further by carrying out screening, be grouped or handling accordingly to these attribute informations Obtain the sample characteristics for carrying out machine learning.Here, prediction data record acquisition device 100 can be by manual, semi-automatic Or full automatic mode carrys out gathered data, as an example, prediction data record acquisition device 100 can gathered data in bulk.
Prediction data record acquisition device 100 can receive what user was manually entered by input unit (for example, work station) Prediction data is recorded.In addition, prediction data record acquisition device 100 can be taken out by full automatic mode from data source systems Prediction data record, for example, by with software, firmware, hardware or its combination realize timer mechanism come systematically number of request Asked data are obtained according to source and from response.The data source may include one or more databases or other servers. Can be realized via internal network and/or external network it is full-automatic obtain the mode of data, wherein may include by internet come Transmit the data of encryption.In the case where server, database, network etc. are configured as communicating with one another, it can not do manually It is automatic in the case of pre- to carry out data acquisition, it should be noted that certain user's input operation still may be present in this manner. Semiautomatic fashion is between manual mode and full-automatic mode.Semiautomatic fashion and the difference of full-automatic mode are by user The trigger mechanism of activation instead of timer mechanism.In this case, in the case where receiving specific user's input, Produce the request for extracting data.When obtaining data every time, it is preferable that can be by the data storage of capture in nonvolatile memory In.As an example, availability data warehouse come be stored in obtain during the data that gather.Alternatively, can be (all by hardware cluster Such as Hadoop clusters) data collected are stored and/or subsequent treatment, for example, storage, classification and other offline behaviour Make.In addition, also online stream process can be carried out to the data of collection.
As an example, may include the data conversion modules such as text analysis model in prediction data record acquisition device 100, use The structural data that uses is easier to be further processed or quote in the unstructured datas such as text are converted to.Base It may include Email, document, webpage, figure, spreadsheet, call center's daily record, suspicious transaction report in the data of text Accuse etc..
Next, in step s 200, by forecast sample generation device 200 based on the attribute information that prediction data is recorded come The forecast sample of generation machine learning model corresponding with service class, wherein, with the basic level of service among service class The forecast sample of corresponding basic machine learning model includes essential characteristic subset, or, taken with the enhancing among service class The forecast sample of the corresponding reinforcement machine learning model of rank of being engaged in includes essential characteristic subset and at least one supplementary features subset.
Here, be converted to by prediction data record and can directly input the pre- of machine learning model corresponding with service class During test sample sheet, the essential characteristic or additional in each character subset of forecast sample can be generated based on each attribute information Feature.According to the exemplary embodiment of the present invention, forecast sample can have multiple character subsets, and each submodel has respective Character subset.
As described above, service class can be the rank preset, or, service class can be entered from multiple candidate's ranks Row is chosen.In this case, forecast sample generation device 200 also needs to be indicated according to user or selects pre- according to application scenario Used service class during survey.In the case where service class is determined, corresponding machine learning model is able to determine or selected In.
In addition, forecast sample generation device 200 can in any suitable manner, the attribute letter recorded based on prediction data Cease to produce the individual features of forecast sample, and these features are combined as each character subset according to ad hoc fashion.It should be noted that Forecast sample generation device 200, can be according to any relevant with attribute information, submodel or data etc. when producing character subset Factor, to cause the submodel based on each character subset to corresponding affect on the quality of machine learning service under lift frame, Therefore, exemplary embodiment of the invention is not intended to limit the specific producing method of character subset.
Here, the screening or packet of attribute information during producing feature, can not only carried out based on attribute information, The attribute information that screening or packet are obtained can be also further processed, i.e. alternately, forecast sample generation device 200 can carry out Feature Engineering processing to the prediction data record of acquisition, for example, forecast sample generation device 200 can be to prediction number Discretization, field combination are carried out according to the primitive attribute information of record, part field value is extracted, the various features engineering such as rounds Processing, and the feature after processing is combined as each character subset according to ad hoc rules.
As an example, forecast sample generation device 200 can be produced during forecast sample is produced based on essential characteristic Raw supplementary features.Here, forecast sample generation device 200 can be by performing such as discretization, spy at least one essential characteristic Combination is levied, part field value is extracted, rounds etc. to produce supplementary features.For example, forecast sample generation device 200 can be by base Eigen is combined to produce supplementary features, here, while essential characteristic is combined, alternately, can also carry out Other extra processing.
In step S300, forecast sample is supplied to machine learning mould corresponding with service class by prediction meanss 300 Type, to obtain predicting the outcome for forecast sample, wherein, reinforcement machine learning model include basic machine learning model and At least one additional submodel identical with basic machine learning model type and trained according to lift frame, wherein, base Machine learning model corresponds to essential characteristic subset, and additional submodel corresponds to supplementary features subset.
Here, machine learning model can be stored among the forecasting system shown in Fig. 1, or, machine learning model can be protected Exist outside the forecasting system shown in Fig. 1；As an example, the machine learning can be read by prediction meanss 300 or other devices Model so that forecast sample can directly be supplied to submodel (that is, the base of the machine learning model read out by prediction meanss 300 Eigen model or additional submodel).
In addition, machine learning model can be also always positioned at outside the forecasting system shown in Fig. 1, and it is straight by prediction meanss 300 Connect or training sample is supplied to externally-located machine learning model via other devices.In this case, prediction meanss 300 can also predicting the outcome from external reception machine learning model.
Under lift frame, predicting the outcome for each submodel is applied, alternately, and stack result can be by pre- The conversion first defined is to obtain final predict the outcome.In this way, can via character subset design, in lift frame Under, constitute to provide the machine learning service of specific rank via different submodels.
The service class according to an exemplary embodiment of the present invention that is based on is described below in conjunction with Fig. 3 and Fig. 4 come training machine The system and its training method of learning model.
According to the exemplary embodiment of the present invention, the machine learning model may include basic machine learning model, or, It can also additionally include additional submodel with basic machine learning model same type, also, train according to lift frame work For the basic machine learning model and additional submodel of submodel.Here, submodel can be quantitatively it is one or more, no There can be the identical or entirely different character subset in identical, part with submodel.
Particularly, Fig. 3 shows that the service class according to an exemplary embodiment of the present invention that is based on learns mould come training machine The block diagram of the system of type.Training system shown in Fig. 3 can be realized all by computer program with software mode, also can be by special The hardware unit of door is realized, can also be realized by way of software and hardware combining.Correspondingly, the training system shown in composition Fig. 3 Each device of system can only rely on computer program to realize the virtual module of corresponding function or rely on hardware knot Structure realizes the universal or special device of the function, can also be that operation has processor of corresponding computer program etc..
As shown in figure 3, training data record acquisition device 1000 is used to obtain training data record.Here, training data Record acquisition device 1000 can using it is various it is appropriate by the way of come offline or acquisition training data is recorded online.According to the present invention Exemplary embodiment, training data record acquisition device 1000 can use with prediction data record acquisition device 100 it is similar Mode performs operation, and the specific data that only both obtain are different, therefore here will no longer be described in greater detail. In the case of supervised learning, the training data obtained by training data record acquisition device 1000 is recorded except including each kind Property information outside, in addition to the data record relative to forecasting problem mark (label).
Training sample generation device 2000 is used to the attribute information based on training data record generate and service class pair The training sample for the machine learning model answered, wherein, basic engineering corresponding with the basic level of service among service class Practising the training sample of model includes essential characteristic subset, or, enhancing corresponding with the enhancing service class among service class The training sample of machine learning model includes essential characteristic subset and at least one supplementary features subset.Here, training sample is produced Generating apparatus 2000 can generate character subset in any suitable fashion, for example, it is contemplated that the content of attribute information, implication, The factors such as value continuity, span, valued space scale, Deletional, importance, or, machine learning model can be combined In submodel feature etc. so that each submodel of feature based subset can effectively from some or it is some in terms of influence The rank of machine learning service.According to the present invention exemplary embodiment, training sample generation device 2000 can according to prediction The corresponding mode of sample generation device 200 generates each feature of training sample, i.e. training sample is with feature samples in feature Correspondence is respectively provided with terms of character subset.It should be understood that because prediction data record in practice can relative to training data record Can there can be the attribute information of some missings, it is therefore, relevant with missing attribute information in the generation of forecast sample generation device 200 During feature, the corresponding missing attribute information in prediction data record can be set as null value or default value.
According to the exemplary embodiment of the present invention, formed between each submodel based on lift frame training, correspondingly, respectively Individual features subset that individual submodel is corresponded respectively in training sample is trained.
As can be seen here, basic machine learning model and each additional submodel can be based respectively on identical or different training number Formed according to record training.For example, all submodels can be formed based on all training data record training, or, it can also distinguish Formed based on a part of training data record training sampled from all training datas record.As an example, can be according to default Sampling policy distribute corresponding training data record for each submodel, distributed for example, more training data can be recorded To basic machine learning model, and less training data record is distributed into additional submodel, here, different submodel distribution Training data record between can have it is a certain proportion of common factor or not occur simultaneously completely.By being determined according to sampling policy Training data used in each submodel is recorded, and can further lift the effect of whole machine learning model.
Trainer 3000 is used to train machine learning model corresponding with service class using the training sample of generation, Wherein, reinforcement machine learning model include basic machine learning model and it is identical with basic machine learning model type and according to At least one additional submodel of lift frame training, wherein, basic machine learning model corresponds to essential characteristic subset, Additional submodel corresponds to supplementary features subset.
Particularly, trainer 3000 can learn according to lift frame (for example, gradient lift frame) come training machine Each submodel (that is, basic machine learning model and additional submodel) of the type identical that model includes, wherein, per height Model is trained based on respective character subset.Here, trainer 3000 can based on loading model training configure come Train the submodel included by machine learning model by stage (stage).Particularly, in the basic machine of first stage-training During learning model, trainer 3000 can perform initialization process according to the parameter of configuration.In addition, being instructed in follow-up each stage When practicing additional submodel, it can be configured determine that the character subset of submodel that this stage trained is drawn according to the model training of loading Point.After all submodels are trained, complete machine learning model can be correspondingly made available, the machine learning model can quilt Storage in the system of figure 3 subsequently to use, or, the machine learning model trained can be supplied to external system or dress Put.
As an example, in the case where training reinforcement machine learning model, trainer 3000 can be by fixing wherein It is trained go out basic machine learning model and additional submodel train remaining additional submodel successively.That is, for The submodel come has been trained, the coefficient of these submodels can be fixed so that after training during continuous submodel, can have been saved Operand.
The service class according to an exemplary embodiment of the present invention that is based on is described hereinafter with reference to Fig. 4 come training machine to learn The flow chart of the method for model.Here, as an example, method shown in Fig. 4 can be as shown in Figure 3 training system perform, It can be realized, can also be performed by the computing device of particular configuration shown in Fig. 4 with software mode by computer program completely Method.
For convenience, it is assumed that the training system of method as shown in Figure 3 shown in Fig. 4 is performed, as illustrated, in step In rapid S1000, training data record is obtained by training data record acquisition device 1000.Here, can according to step S100 classes As mode perform step S1000, the specific data only obtained in the two steps are different, for example, there is supervision In the case of study, training data is recorded in addition to including various attribute informations, in addition to data record is relative to pre- The mark (label) of survey problem.
Next, in step S2000, the attribute information recorded by training sample generation device 2000 based on training data To generate the training sample of machine learning model corresponding with service class, wherein, with the basic service level among service class The training sample of not corresponding basic machine learning model includes essential characteristic subset, or, with the enhancing among service class The training sample of the corresponding reinforcement machine learning model of service class includes essential characteristic subset and at least one supplementary features Collection.It should be understood that step S2000 can be performed according to mode corresponding with step S200, simply except feature in training sample Outside subset, in addition it is also necessary to including corresponding mark, therefore, some duplicate contents and details are will not be described in great detail here.
In step S3000, trainer 3000 can be corresponding with service class to train using the training sample of generation Machine learning model, wherein, reinforcement machine learning model include basic machine learning model and with basic machine learning model Type is identical and according at least one additional submodel of lift frame training, wherein, basic machine learning model correspondence In essential characteristic subset, additional submodel corresponds to supplementary features subset.
Particularly, trainer 3000 can configure at least one among the following items of machine learning model：Submodule Type sum, submodule shape parameter, submodel Parameters variation mode.The model training configuration formed can be used for instructing follow-up for each Every stage-training of individual submodel.Especially, in this step, submodel parameter can be set to gradually change.By this Parameter adaptive (parameter adaptation), can allow model population parameter (such as learning rate) and submodule shape parameter is (such as Linear model iteration wheel number, regularization coefficient etc.) gradually changed.
Here, trainer 3000 can be trained first with the training sample being made up of essential characteristic subset together with mark Obtain basic machine learning model.
On this basis, the reinforcement machine learning model under lift frame is represented by basic machine learning model and at least The splicing result of one additional submodel, the result may correspond to a relatively strong model.Here, basic machine learning mould Type corresponds to basic level of service, and different reinforcement machine learning models are each on the whole due to wherein included additional submodel Differ, so corresponding to respective enhancing service class.
As an example, after training draws basic machine learning model, further training reinforcement machine learning model it In each additional submodel process can it is abstract be on the basis of the submodel trained, according to lift frame come The process of follow-up additional submodel is trained successively.Herein, the submodel trained can be basic machine learning model, It can also be the set of basic machine learning model and additional submodel.
Assuming that reinforcement machine learning model is expressed as F, here, F can be made up of (herein, by basic machine m submodel f Learning model and the unification of additional submodel are represented with symbol f), it is assumed that input data record is expressed as x, is passing through corresponding sample After the processing of generation device, the feature of the corresponding sample portion of k-th of submodel is xk.Correspondingly, can be according to following etc. Formula 1 builds reinforcement machine learning model F：
According to the exemplary embodiment of the present invention, the input of each submodel may correspond to character subset, this feature subset It can regard as by carrying out eigentransformation (for example, Φ to input data recordk()) and obtain, i.e. xk=Φk(x).That is, The reinforcement machine learning model that equation 1 is limited is represented by as shown in following equation 2：
That is, in an exemplary embodiment of the present invention, each submodel is fk(Φk(x)).Correspondingly, per single order Section can train a corresponding submodel.
Particularly, it is assumed that have been completed the training of m submodel, it can be correspondingly made available what is be made up of m submodel Machine learning model
Assuming that there is training sample set the D={ (Φ obtained based on N (N is the integer more than 1) individual training data record (xi),yi) | i=1,2 ..., N }, wherein, xiIndicate i-th of training data record, Φ (xi) it is corresponding training sample feature, yiFor xiMark, moreover, it is assumed that loss function is l, then Fm(x) total losses on training sample set D is represented by following Equation 3：
In the following description, the D in above-mentioned expression formula can be omitted, L (F are only written asm)。
In the case where currently having trained m submodel, the m+1 submodel can be obtained by minimizing function fm+1, i.e.,：
In general, the no closed solutions of above-mentioned minimum, accordingly, it would be desirable to carry out corresponding iteration for different types of f Processing.
As an example it is supposed that submodel is linear submodel (for example, logarithm probability regression model), reinforcement machine learning Model is represented by：
In above formula, fkEach linear submodel of trained completion is represented,Partly refer to and be currently needed for The linear submodel of training.Correspondingly, the coefficient of current linear submodel can be updated according to following equation：
In above formula,For xiAfter trained sample generation device, the training sample of k-th of submodel of correspondence of generation is special Levy；λ, γ are regularization coefficient (regularizer coefficient), for the complexity of control line sub-model.Here, FTRL-Proximal algorithms can be used to carry out iterative wm+1。
The exemplary training method of submodel is enumerated above, however, it should be understood that the exemplary embodiment of the present invention is not It is limited to above-mentioned example.For example, in training machine learning model, each submodel is not necessarily limited by same training data It is trained in space, that is to say, that each submodel can be based on respective training data space.So, each submodel institute Based on training data record can be identical or entirely different with identical, part.
Those skilled in the art can train what reinforcement machine learning model included successively in any suitable manner Each submodel.For some reinforcement machine learning model, one or more additional submodels that it is included are on the whole Reflect corresponding service class.The service class difference of different reinforcement machine learning models is mostly derived from respective additional son Model part is distinct.
According to the present invention exemplary embodiment, can be trained only for some default service class it is its corresponding only One machine learning model (basic machine learning model or reinforcement machine learning model), i.e. for being selected among multiple service class The service class selected performs model training method, to obtain unique machine learning model.Or, it can also be directed to multiple seeervice levels Multiple machine learning models (including basic machine learning model and/or at least one reinforcement machine learning mould are not respectively trained out Type), i.e. model training method is performed respectively for each service class among multiple service class, to obtain multiple machines Device learning model.
As described above, in the case where training reinforcement machine learning model, the fixed base wherein trained can be passed through Machine learning model and additional submodel train remaining additional submodel successively.Therefore, multiple enhancing engineerings are being trained In the case of practising model, for the training mission of submodel is fixed jointly, parallel training can be easily performed, Further to improve operation efficiency.
It should be understood that Fig. 1 and device illustrated in fig. 3 can be individually configured to perform software, hardware, the firmware of specific function Or any combination of above-mentioned item.For example, these devices may correspond to special integrated circuit, pure software generation can also correspond to Code, also corresponds to unit or module that software is combined with hardware.In addition, the one or more functions that these devices are realized Also it can be sought unity of action by the component in physical entity equipment (for example, processor, client or server etc.).
Described above by reference to Fig. 1 and Fig. 2 and according to an exemplary embodiment of the present invention machine is performed based on service class Learn the system and method for prediction.It should be understood that above-mentioned Forecasting Methodology can be by recording the program in computer-readable media come real It is existing, correspondingly, according to the exemplary embodiment of the present invention, it is possible to provide a kind of to perform machine learning prediction based on service class Medium, wherein, record has the computer program for performing following methods step on the computer-readable medium：(a) obtain Prediction data is taken to record；(b) machine learning mould corresponding with service class is generated based on the attribute information of prediction data record The forecast sample of type, wherein, the pre- test sample of basic machine learning model corresponding with the basic level of service among service class This includes essential characteristic subset, or, reinforcement machine learning model corresponding with the enhancing service class among service class Forecast sample includes essential characteristic subset and at least one supplementary features subset；And forecast sample is supplied to and serviced by (c) The corresponding machine learning model of rank, to obtain predicting the outcome for forecast sample, wherein, reinforcement machine learning model includes Basic machine learning model and it is identical with basic machine learning model type and trained according to lift frame at least one Individual additional submodel, wherein, basic machine learning model corresponds to essential characteristic subset, and additional submodel corresponds to supplementary features Subset.
The service class according to an exemplary embodiment of the present invention that is based on is described come training machine above by reference to Fig. 3 and Fig. 4 The system and method for learning model.It should be understood that above-mentioned training method can be by recording the program in computer-readable media come real It is existing, correspondingly, according to the exemplary embodiment of the present invention, it is possible to provide it is a kind of based on service class come training machine learning model Medium, wherein, record has the computer program for performing following methods step on the computer-readable medium：(A) obtain Training data is taken to record；(B) machine learning mould corresponding with service class is generated based on the attribute information of training data record The training sample of type, wherein, the training sample of basic machine learning model corresponding with the basic level of service among service class This includes essential characteristic subset, or, reinforcement machine learning model corresponding with the enhancing service class among service class Training sample includes essential characteristic subset and at least one supplementary features subset；And (C) is instructed using the training sample of generation Practice corresponding with service class machine learning model, wherein, reinforcement machine learning model including basic machine learning model and At least one additional submodel identical with basic machine learning model type and trained according to lift frame, wherein, base Machine learning model corresponds to essential characteristic subset, and additional submodel corresponds to supplementary features subset.
Computer program in above computer computer-readable recording medium can be in client, main frame, agent apparatus, server etc. In the environment disposed in computer equipment 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 performing above-mentioned steps more specifically handles, and these additional steps and further handles Content is described referring to figs. 1 to Fig. 4, here in order to avoid repetition will be repeated no longer.
It should be noted that forecasting system according to an exemplary embodiment of the present invention or training system can be completely dependent on computer program Operation realize corresponding function, i.e. each device is corresponding with each step to the function structure of computer program so that whole Individual system is called by special software kit (for example, lib storehouses), to realize corresponding forecast function.
On the other hand, each device shown in Fig. 1 or Fig. 3 can also pass through hardware, software, firmware, middleware, microcode Or it is combined to realize.When being realized with software, firmware, middleware or microcode, the program for performing corresponding operating Code or code segment can be stored in the computer-readable medium of such as storage medium so that processor can be by reading simultaneously Corresponding program code or code segment is run to 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 conjunction, when the set of computer-executable instructions is closed by institute When stating computing device, perform using machine learning model come the method for perform prediction and/or the training machine learning model Method.
Particularly, the computing device can be deployed in server or client, can also be deployed in distributed network On node apparatus in network environment.In addition, 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), graphics processor (GPU), may be programmed and patrol Collect device, dedicated processor systems, microcontroller or microprocessor.Unrestricted as example, processor may also include simulation Processor, digital processing unit, microprocessor, polycaryon processor, processor array, network processing unit etc..
Some operations described in Forecasting Methodology and training method according to an exemplary embodiment of the present invention can be by soft Part mode realizes that some operations can be realized by hardware mode, in addition, can also be realized by way of software and hardware combining These operations.
Processor can run the instruction being stored in one of memory unit or code, wherein, the memory unit can be with Data storage.Instruction and data can be also sent and received by network via Network Interface Unit, 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.In addition, memory unit may include independent device, such as, outside dish driving, storage array or any Database Systems can Other storage devices used.Memory unit and processor can be coupled operationally, or can for example by I/O ports, Network connection etc. is communicated so that processor can read the file being stored in memory unit.
In addition, 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.
Operation involved by Forecasting Methodology and/or training method according to an exemplary embodiment of the present invention can be described as respectively Plant the functional block or function diagram of interconnection or coupling.However, these functional blocks or function diagram can be equably integrated into it is single Logic device or operated according to non-definite border.
Particularly, as described above, according to an exemplary embodiment of the present invention perform machine learning based on service class The computing device of prediction may include memory unit and processor, wherein, be stored with set of computer-executable instructions in memory unit Close, when the set of computer-executable instructions is closed by the computing device, perform following step：(a) prediction data is obtained Record；(b) the pre- test sample of machine learning model corresponding with service class is generated based on the attribute information of prediction data record This, wherein, the forecast sample of basic machine learning model corresponding with the basic level of service among service class is included substantially Character subset, or, the forecast sample bag of reinforcement machine learning model corresponding with the enhancing service class among service class Include essential characteristic subset and at least one supplementary features subset；And (c) forecast sample is supplied to it is corresponding with service class Machine learning model, to obtain predicting the outcome for forecast sample, wherein, reinforcement machine learning model includes basic engineering Practise model and at least one additional submodule identical with basic machine learning model type and trained according to lift frame Type, wherein, basic machine learning model corresponds to essential characteristic subset, and additional submodel corresponds to supplementary features subset.
It should be noted that above combined Fig. 1 and Fig. 2 describe it is according to an exemplary embodiment of the present invention be based on service class To perform each processing details of machine learning prediction, processing details when computing device performs each step is will not be described in great detail here.
In addition, it is according to an exemplary embodiment of the present invention based on service class come the computing device of training machine learning model It may include the set of computer-executable instructions conjunction that is stored with memory unit and processor, memory unit, when the computer can be held When row instruction set is by the computing device, following step is performed：(A) training data record is obtained；(B) it is based on training data The attribute information of record generates the training sample of machine learning model corresponding with service class, wherein, with service class it In the corresponding basic machine learning model of basic level of service training sample include essential characteristic subset, or, with service The training sample of the corresponding reinforcement machine learning model of enhancing service class among rank is including essential characteristic subset and at least One supplementary features subset；And (C) trains machine learning model corresponding with service class using the training sample of generation, Wherein, reinforcement machine learning model include basic machine learning model and it is identical with basic machine learning model type and according to At least one additional submodel of lift frame training, wherein, basic machine learning model corresponds to essential characteristic subset, Additional submodel corresponds to supplementary features subset.
It should be noted that above combined Fig. 3 and Fig. 4 describe it is according to an exemplary embodiment of the present invention be based on service class Carry out each processing details of training machine learning model, processing details when computing device performs each step is will not be described in great detail here.
Be described above the present invention each exemplary embodiment, 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 (10)

1. a kind of method that machine learning prediction is performed based on service class, including：
(a) prediction data record is obtained；
(b) the pre- test sample of machine learning model corresponding with service class is generated based on the attribute information of prediction data record This, wherein, the forecast sample of basic machine learning model corresponding with the basic level of service among service class is included substantially Character subset, or, the forecast sample bag of reinforcement machine learning model corresponding with the enhancing service class among service class Include essential characteristic subset and at least one supplementary features subset；And
(c) forecast sample is supplied to machine learning model corresponding with service class, to obtain the prediction for forecast sample As a result,
Wherein, reinforcement machine learning model include basic machine learning model and it is identical with basic machine learning model type and At least one the additional submodel trained according to lift frame, wherein, basic machine learning model corresponds to essential characteristic Subset, additional submodel corresponds to supplementary features subset.
2. the method for claim 1, wherein machine learning model corresponding with service class is to be in advance based on the clothes The unique machine learning model that business rank is trained.
3. the method for claim 1, wherein machine learning model corresponding with service class is multiple from being in advance based on The machine learning mould corresponding with the service class chosen among multiple machine learning models that service class is trained Type.
4. the method for claim 1, wherein service class is used at least one the aspect phase for weighing machine learning service Close.
5. the method for claim 1, wherein the model algorithm of service class and machine learning model, data scale and/ Or computing resource is related.
6. method as claimed in claim 3, wherein, by being chosen and the seeervice level by user determines the service class Not corresponding machine learning model；Or, chosen by automatically determining the service class corresponding with the service class Machine learning model.
7. the method for claim 1, wherein supplementary features are based on essential characteristic and produced.
8. it is a kind of based on service class come the method for training machine learning model, including：
(A) training data record is obtained；
(B) the training sample of machine learning model corresponding with service class is generated based on the attribute information of training data record This, wherein, the training sample of basic machine learning model corresponding with the basic level of service among service class is included substantially Character subset, or, the training sample bag of reinforcement machine learning model corresponding with the enhancing service class among service class Include essential characteristic subset and at least one supplementary features subset；And
(C) machine learning model corresponding with service class is trained using the training sample of generation,
Wherein, reinforcement machine learning model include basic machine learning model and it is identical with basic machine learning model type and At least one the additional submodel trained according to lift frame, wherein, basic machine learning model corresponds to essential characteristic Subset, additional submodel corresponds to supplementary features subset.
9. a kind of system that machine learning prediction is performed based on service class, including：
Prediction data records acquisition device, for obtaining prediction data record；
Forecast sample generation device, machine corresponding with service class is generated for the attribute information recorded based on prediction data The forecast sample of learning model, wherein, basic machine learning model corresponding with the basic level of service among service class Forecast sample includes essential characteristic subset, or, reinforcement machine learning corresponding with the enhancing service class among service class The forecast sample of model includes essential characteristic subset and at least one supplementary features subset；And
Prediction meanss, for forecast sample to be supplied into machine learning model corresponding with service class, to obtain for prediction Sample predicts the outcome,
Wherein, reinforcement machine learning model include basic machine learning model and it is identical with basic machine learning model type and At least one the additional submodel trained according to lift frame, wherein, basic machine learning model corresponds to essential characteristic Subset, additional submodel corresponds to supplementary features subset.
10. it is a kind of based on service class come the system of training machine learning model, including：
Training data records acquisition device, for obtaining training data record；
Training sample generation device, machine corresponding with service class is generated for the attribute information recorded based on training data The training sample of learning model, wherein, basic machine learning model corresponding with the basic level of service among service class Training sample includes essential characteristic subset, or, reinforcement machine learning corresponding with the enhancing service class among service class The training sample of model includes essential characteristic subset and at least one supplementary features subset；And
Trainer, machine learning model corresponding with service class is trained for the training sample using generation,
Wherein, reinforcement machine learning model include basic machine learning model and it is identical with basic machine learning model type and At least one the additional submodel trained according to lift frame, wherein, basic machine learning model corresponds to essential characteristic Subset, additional submodel corresponds to supplementary features subset.
CN201710427869.8A 2017-06-08 2017-06-08 Method and system for performing machine learning prediction based on service level Active CN107273979B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201710427869.8A CN107273979B (en) 2017-06-08 2017-06-08 Method and system for performing machine learning prediction based on service level

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201710427869.8A CN107273979B (en) 2017-06-08 2017-06-08 Method and system for performing machine learning prediction based on service level

Publications (2)

Publication Number Publication Date
CN107273979A true CN107273979A (en) 2017-10-20
CN107273979B CN107273979B (en) 2020-12-01

Family

ID=60066046

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201710427869.8A Active CN107273979B (en) 2017-06-08 2017-06-08 Method and system for performing machine learning prediction based on service level

Country Status (1)

Country Link
CN (1) CN107273979B (en)

Cited By (9)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
WO2019148669A1 (en) * 2018-01-30 2019-08-08 平安科技（深圳）有限公司 Method and apparatus for generating machine learning model, computer device, and storage medium
CN110188910A (en) * 2018-07-10 2019-08-30 第四范式（北京）技术有限公司 The method and system of on-line prediction service are provided using machine learning model
CN110766164A (en) * 2018-07-10 2020-02-07 第四范式（北京）技术有限公司 Method and system for performing a machine learning process
CN111523676A (en) * 2020-04-17 2020-08-11 第四范式（北京）技术有限公司 Method and device for assisting machine learning model to be online
CN111832735A (en) * 2019-04-18 2020-10-27 第四范式（北京）技术有限公司 Method and system for performing a machine learning process based on a template
CN111985637A (en) * 2019-05-21 2020-11-24 苹果公司 Machine learning model with conditional execution of multiple processing tasks
CN112470160A (en) * 2018-07-31 2021-03-09 三星电子株式会社 Apparatus and method for personalized natural language understanding
WO2023273720A1 (en) * 2021-06-28 2023-01-05 京东科技控股股份有限公司 Method and apparatus for training model, and device, and storage medium
CN112470160B (en) * 2018-07-31 2024-06-04 三星电子株式会社 Device and method for personalized natural language understanding

Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN105760950A (en) * 2016-02-05 2016-07-13 北京物思创想科技有限公司 Method for providing or obtaining prediction result and device thereof and prediction system
CN106407999A (en) * 2016-08-25 2017-02-15 北京物思创想科技有限公司 Rule combined machine learning method and system
US20170060993A1 (en) * 2015-09-01 2017-03-02 Skytree, Inc. Creating a Training Data Set Based on Unlabeled Textual Data
CN106503787A (en) * 2016-10-26 2017-03-15 腾讯科技（深圳）有限公司 A kind of method for obtaining game data and electronic equipment
CN106575246A (en) * 2014-06-30 2017-04-19 亚马逊科技公司 Machine learning service

Patent Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN106575246A (en) * 2014-06-30 2017-04-19 亚马逊科技公司 Machine learning service
US20170060993A1 (en) * 2015-09-01 2017-03-02 Skytree, Inc. Creating a Training Data Set Based on Unlabeled Textual Data
CN105760950A (en) * 2016-02-05 2016-07-13 北京物思创想科技有限公司 Method for providing or obtaining prediction result and device thereof and prediction system
CN106407999A (en) * 2016-08-25 2017-02-15 北京物思创想科技有限公司 Rule combined machine learning method and system
CN106503787A (en) * 2016-10-26 2017-03-15 腾讯科技（深圳）有限公司 A kind of method for obtaining game data and electronic equipment

Cited By (10)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
WO2019148669A1 (en) * 2018-01-30 2019-08-08 平安科技（深圳）有限公司 Method and apparatus for generating machine learning model, computer device, and storage medium
CN110188910A (en) * 2018-07-10 2019-08-30 第四范式（北京）技术有限公司 The method and system of on-line prediction service are provided using machine learning model
CN110766164A (en) * 2018-07-10 2020-02-07 第四范式（北京）技术有限公司 Method and system for performing a machine learning process
CN112470160A (en) * 2018-07-31 2021-03-09 三星电子株式会社 Apparatus and method for personalized natural language understanding
CN112470160B (en) * 2018-07-31 2024-06-04 三星电子株式会社 Device and method for personalized natural language understanding
CN111832735A (en) * 2019-04-18 2020-10-27 第四范式（北京）技术有限公司 Method and system for performing a machine learning process based on a template
CN111985637A (en) * 2019-05-21 2020-11-24 苹果公司 Machine learning model with conditional execution of multiple processing tasks
CN111523676A (en) * 2020-04-17 2020-08-11 第四范式（北京）技术有限公司 Method and device for assisting machine learning model to be online
CN111523676B (en) * 2020-04-17 2024-04-12 第四范式（北京）技术有限公司 Method and device for assisting machine learning model to be online
WO2023273720A1 (en) * 2021-06-28 2023-01-05 京东科技控股股份有限公司 Method and apparatus for training model, and device, and storage medium

Also Published As

Publication number Publication date
CN107273979B (en) 2020-12-01

Similar Documents

Publication Publication Date Title
CN107273979A (en) The method and system of machine learning prediction are performed based on service class
CN107169573A (en) Using composite machine learning model come the method and system of perform prediction
CN103502899B (en) Dynamic prediction Modeling Platform
CN109919684A (en) For generating method, electronic equipment and the computer readable storage medium of information prediction model
CN112100295A (en) User data classification method, device, equipment and medium based on federal learning
CN108268934A (en) Recommendation method and apparatus, electronic equipment, medium, program based on deep learning
CN107392319A (en) Generate the method and system of the assemblage characteristic of machine learning sample
CN106407999A (en) Rule combined machine learning method and system
CN108021984A (en) Determine the method and system of the feature importance of machine learning sample
US20230153857A1 (en) Recommendation model training method, recommendation method, apparatus, and computer-readable medium
CN107169574A (en) Using nested machine learning model come the method and system of perform prediction
CN107704871A (en) Generate the method and system of the assemblage characteristic of machine learning sample
CN107220217A (en) Characteristic coefficient training method and device that logic-based is returned
CN108090516A (en) Automatically generate the method and system of the feature of machine learning sample
CN107729915A (en) For the method and system for the key character for determining machine learning sample
CN107316082A (en) For the method and system for the feature importance for determining machine learning sample
CN110188910A (en) The method and system of on-line prediction service are provided using machine learning model
CN108090570A (en) For selecting the method and system of the feature of machine learning sample
CN107103036A (en) Using acquisition methods, equipment and the programmable device for downloading probability
US20210350175A1 (en) Key-value memory network for predicting time-series metrics of target entities
CN107679549A (en) Generate the method and system of the assemblage characteristic of machine learning sample
CN108182472A (en) For generating the method and apparatus of information
CN111461757B (en) Information processing method and device, computer storage medium and electronic equipment
CN115034836A (en) Model training method and related device
CN116308640A (en) Recommendation method and related device

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination
GR01 Patent grant
GR01 Patent grant