CN110472742A - A kind of model variable determines method, device and equipment - Google Patents
A kind of model variable determines method, device and equipment Download PDFInfo
- Publication number
- CN110472742A CN110472742A CN201910626175.6A CN201910626175A CN110472742A CN 110472742 A CN110472742 A CN 110472742A CN 201910626175 A CN201910626175 A CN 201910626175A CN 110472742 A CN110472742 A CN 110472742A
- Authority
- CN
- China
- Prior art keywords
- sample
- characteristic value
- variable
- object module
- keyword
- 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
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/30—Information retrieval; Database structures therefor; File system structures therefor of unstructured textual data
- G06F16/35—Clustering; Classification
-
- 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
- G06Q30/00—Commerce
- G06Q30/02—Marketing; Price estimation or determination; Fundraising
- G06Q30/0282—Rating or review of business operators or products
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Business, Economics & Management (AREA)
- General Physics & Mathematics (AREA)
- Physics & Mathematics (AREA)
- Software Systems (AREA)
- Accounting & Taxation (AREA)
- Strategic Management (AREA)
- Finance (AREA)
- General Engineering & Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Development Economics (AREA)
- Mathematical Physics (AREA)
- Game Theory and Decision Science (AREA)
- Artificial Intelligence (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Medical Informatics (AREA)
- Entrepreneurship & Innovation (AREA)
- Computing Systems (AREA)
- Databases & Information Systems (AREA)
- Evolutionary Computation (AREA)
- Economics (AREA)
- Marketing (AREA)
- General Business, Economics & Management (AREA)
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
This application discloses a kind of model variables to determine method, device and equipment.Method includes: then to analyze the characteristic value of sample characteristics of the sample data in initializaing variable dimension for the initializaing variable collection of model configuration in advance firstly, determining, obtain characteristic value sequence；Then, it is based on characteristic value sequence, decomposition fusion is carried out to initializaing variable, obtains influencing maximum optimal variables set to objective function.
Description
Technical field
This application involves field of computer technology more particularly to a kind of model variable to determine method, device and equipment.
Background technique
Feature Engineering refers to using specialty background knowledge and skill processing data, enables feature in machine learning model
Play the process preferably acted on.Characteristic variable used in training machine learning model is usually related technical personnel's base at present
It is artificially demarcated in factors such as experience, business demands.
Therefore, it is necessary to relatively reliable model variables to determine scheme.
Summary of the invention
This specification embodiment provides a kind of model variable and determines method, for realizing the automatic processing of model variable.
This specification embodiment also provides a kind of model variable and determines method, comprising:
Determine the initializaing variable collection of object module；
Determine the corresponding characteristic value sequence of sample data, the characteristic value sequence is for characterizing the sample data each first
The characteristic value of the sample characteristics of beginning variable dimension；
Objective function based on the characteristic value sequence and the object module determines the optimal variable of the object module
Collection.
This specification embodiment also provides a kind of model variable determining device, comprising:
Model is determined, for determining the initializaing variable collection of object module；
First processing module, for determining the corresponding characteristic value sequence of sample data, the characteristic value sequence is for characterizing
Characteristic value of the sample data in the sample characteristics of each initializaing variable dimension；
Second processing model, for the objective function based on the characteristic value sequence and the object module, determine described in
The optimal variables set of object module.
This specification embodiment also provides a kind of electronic equipment, comprising:
Processor；And
It is arranged to the memory of storage computer executable instructions, the executable instruction makes the place when executed
Device is managed to execute such as the step of above-mentioned method.
This specification embodiment also provides a kind of computer readable storage medium, deposits on the computer readable storage medium
Computer program is contained, is realized when the computer program is executed by processor such as the step of above-mentioned method.
This specification embodiment use at least one above-mentioned technical solution can reach it is following the utility model has the advantages that
By acquiring the set of the initializaing variable for training pattern configuration in advance, and sample data is analyzed in each initial change
The characteristic value of sample characteristics in the dimension of amount obtains characteristic value sequence, then, be based on characteristic value sequence, to initializaing variable into
Row decomposes fusion, to analyze the set for influencing maximum optimal variable on model objective function.With artificially match in the prior art
The scheme for setting model variable is compared, and the determination efficiency and effect of model variable can be effectively improved, and then improves model training effect
Fruit.
Detailed description of the invention
The drawings described herein are used to provide a further understanding of the present application, constitutes part of this application, this Shen
Illustrative embodiments and their description please are not constituted an undue limitation on the present application for explaining the application.In the accompanying drawings:
Fig. 1 is the flow diagram that a kind of model variable that one embodiment of this specification provides determines method；
Fig. 2 is a kind of flow diagram of the implementation for the step 106 that one embodiment of this specification provides；
Fig. 3 is the flow diagram that a kind of model variable that another embodiment of this specification provides determines method；
Fig. 4 is a kind of structural schematic diagram for model variable determining device that one embodiment of this specification provides；
Fig. 5 is the structural schematic diagram for a kind of electronic equipment that one embodiment of this specification provides.
Specific embodiment
To keep the purposes, technical schemes and advantages of the application clearer, below in conjunction with the application specific embodiment and
Technical scheme is clearly and completely described in corresponding attached drawing.Obviously, described embodiment is only the application one
Section Example, instead of all the embodiments.Based on the embodiment in the application, those of ordinary skill in the art are not doing
Every other embodiment obtained under the premise of creative work out, shall fall in the protection scope of this application.
As background technology part is stated, current model is usually artificially selected, people's meat larger workload and choosing
The effect of variable out is also irregular.Based on this, the present invention provides a kind of model variable and determines method, passes through parsing sample number
It is maximum on the influence of the objective function of model to analyze according to the characteristic value of the sample characteristics of the dimension of the initializaing variable in model
Optimal variable.
Below in conjunction with attached drawing, the technical scheme provided by various embodiments of the present application will be described in detail.
Fig. 1 is the flow diagram that a kind of model variable for providing of one embodiment of this specification determines method, referring to Fig. 1,
This method can specifically include following steps:
Step 102, the initializaing variable collection for determining object module；
Wherein, initializaing variable collection can be artificial preconfigured one group of variable, is also possible to last Optimized model and becomes
The optimum results of amount.For the former, then a kind of implementation of step 102 can be with are as follows:
Step S1, the keyword set of the sample data is determined；Specifically:
Word segmentation processing is carried out to the sample data, obtains multiple keywords, and generate keyword set.
Step S2, it is based on the keyword set, determines the initializaing variable collection of the object module.Specifically:
Clustering processing is carried out to the keyword in the keyword set of each sample data, obtains multiple crucial parts of speech；It determines every
The corresponding characteristic variable of a key part of speech, and generate initializaing variable collection.
Wherein, the initializaing variable that the initializaing variable is concentrated includes: time series variable, content variable, evaluates in variable
At least one；Time series variable is used to characterize the time of origin of sample data, and content variable is for characterizing sample data pair
The particular content of event is answered, evaluation variable refers to user to certain object in the corresponding sample event of sample data or sample event
Evaluation information.
By taking scene of doing shopping as an example, step S1 and S2 be can be exemplified are as follows:
Firstly, carrying out word segmentation processing to the shopping event gathered in advance as sample, it is corresponding to obtain each shopping event
Keyword set, such as: certain shopping event keyword set include: ' user A ', ' at 10 points in the morning ', ' refrigerator ', ' value 5000 ',
' fine ' etc.；Then, clustering processing is carried out to the keyword in each keyword, such as: it can be by ' user A ', ' cluster such as user B '
It can be shopping-time class keywords by the clusters such as ' at 10 points in the morning ', ' at 15 points in afternoon ', with such for user identity class keywords
Push away, multiple crucial parts of speech can be obtained, include at least: ' shopping-time class ', ' user class ', ' purchase commodity ', ' commodity value ',
' commodity evaluation ' etc.；Finally, converting each crucial part of speech to the form of characteristic variable, obtain including shopping-time variable, use
The initializaing variable collection of the characteristic variables such as family variable, commodity variable.
Step 104 determines the corresponding characteristic value sequence of sample data, and the characteristic value sequence is for characterizing the sample number
According to the characteristic value of the sample characteristics in each initializaing variable dimension；
Wherein, initializaing variable dimension refers to the corresponding dimension of initializaing variable that initializaing variable is concentrated, such as: user's dimension,
Shopping-time dimension etc..
Based on this, a kind of implementation of step 104 can be with are as follows:
Step S1`, the keyword in the sample data is parsed, sample feature set is obtained；
It is understandable to be, before step S1`, further includes: the step of analyzing the keyword in sample data, the analysis
Step is specifically as follows:
Step S11`, based on preset variable classification rule, extract in the sample data with each types of variables phase
The data information matched；
Wherein, design variables classifying rules is used to be arranged the variable classification of a larger dimension, includes at least: time series class
Variable, text class variable etc., corresponding, the data information to match with each types of variables includes at least: numerical value category information and text
This category information.
Step S12`, the data information to match described in parsing with each types of variables obtains the key of the sample data
Word.
Assuming that data information includes: numerical value category information and text category information, then step S12` is specifically as follows:
The numerical value category information is parsed, time-critical word is obtained；Word segmentation processing is carried out to the text category information, is obtained interior
Hold keyword；Semantic analysis processing is carried out to the text category information, obtains evaluation keyword.
With ' user A has purchased one at 10 points of the morning and is worth the refrigerator of 5000 RMB, and gives in use for some time
Favorable comment out ' for, then above-mentioned steps S11` and step S12` specifically can be exemplified are as follows:
Firstly, extracting the numerical value category information (such as :) in the shopping event at 10 points, and extracted using regular expression
Text category information therein；Then, numerical value category information is parsed, obtains this time-critical word of -10 point of year-month-day, and as this
The time series feature of shopping-time；Word segmentation processing is carried out to text category information, obtains ' user A ', ' 5000 RMB ', ' ice
Case ', ' a period of time ', the content keywords such as ' favorable comment '；Then, semantic analysis processing is carried out to text category information, understands and ' uses
Favorable comment is provided after a period of time ' the meaning of a word, obtain evaluation keyword.
Based on this, a kind of implementation of step S1` can be with are as follows:
Based on the time-critical word, timed sample sequence feature is generated；Based on the content keyword, content sample is generated
Eigen；Based on the evaluation keyword, evaluation sample characteristics are generated；
Based on the timed sample sequence feature, content sample feature and evaluation sample characteristics, sample feature set is generated.
Step S2`, determine each of sample feature set sample characteristics in corresponding initializaing variable dimension
Characteristic value, obtain characteristic value sequence.A kind of its implementation can be with are as follows:
Step S21`, be based on each initializaing variable dimension, generate multidimensional coordinate system, the reference axis of the multidimensional coordinate system and just
Beginning variable dimension corresponds；
That is, each initializaing variable dimension has corresponding reference axis.
Step S22`, pair of each of the sample feature set sample characteristics in the multidimensional coordinate system is determined
The value in reference axis is answered, and as characteristic value.
By taking shopping event as an example, step S21` and step S22` specifically be can be exemplified are as follows:
Assuming that initializaing variable dimension includes: the dimensions such as user, shopping-time, commodity, commodity value, user's evaluation, then structure
The space coordinates of five dimensions are built, five reference axis of the space coordinates of five dimension and five initializaing variable dimensions one are a pair of
It answers, such as: reference axis x, y, z, m, n are corresponded with user, shopping-time, commodity, commodity value, user's evaluation respectively；So
Afterwards, the user characteristics for the event of doing shopping are quantified as a point (value) in x-axis, are by the shopping-time characteristic quantification for the event of doing shopping
A point (value) in y-axis, the point that each sample characteristics for the event of doing shopping can be similarly quantified as on respective coordinates axis
The shopping event can be quantified as the point of a multidimensional in the space coordinates, i.e. characteristic value sequence by (value) as a result,；With this
Analogize, each shopping event can be quantified as to the point of a multidimensional in space coordinates.
In addition, if detecting, there are ready to balance sample characteristics, the ready to balance sample characteristics in the sample feature set
The sample characteristics being not present in other sample datas to exist in target sample data, then it is normal for, if it does not exist, then sample
Eigen should be quantified as 0, and there are larger differences between the normal quantized value of the sample characteristics, therefore, in order to which balanced user comments
The influence of valence, to objectively evaluate out optimal variable, the present embodiment further include: the step of optimizing the ready to balance sample characteristics, it should
Step is specifically as follows:
It step S1``, is the eigenvalue assignment weighted value of the ready to balance sample characteristics in fisrt feature value sequence；
Step S2``, virtual feature value is configured for second feature value sequence, the virtual feature value is described for fictionalizing
Characteristic value of the ready to balance sample characteristics in second sample feature set；
Wherein, the fisrt feature value sequence is the corresponding characteristic value sequence of the target sample data, and described second is special
Value indicative sequence is the corresponding characteristic value sequence of other described sample datas.
Equally by taking shopping event as an example, it is assumed that part shopping event includes this feature of user's evaluation, and another part is purchased
Formal matter part does not then include this feature of user's evaluation, then above-mentioned steps S1`` and step S2`` specifically can be exemplified are as follows:
Firstly, parsing the corresponding characteristic value sequence of each shopping event, the characteristic value sequence of no user's evaluation characteristic value is determined
Column, are denoted as fisrt feature value sequence, other characteristic value sequences are denoted as second feature value sequence；Then, it is determined that Second Eigenvalue sequence
In column, the distribution situation of user's evaluation characteristic value；It then, is that fisrt feature value sequence increases by a virtual use based on distribution situation
Family evaluating characteristic value, and be one weight of user's evaluation eigenvalue assignment of second feature value sequence.
Wherein, virtual user's evaluation characteristic value can be averaged for user's evaluation characteristic value in Second Eigenvalue sequence
Value；Weight can be a preset fixed value, generally higher than 0, to characterize the user's evaluation feature in second feature value sequence
The different degree of value is greater than virtual user's evaluation characteristic value.
Step 106, the objective function based on the characteristic value sequence and the object module, determine the object module
Optimal variables set.Referring to fig. 2, a kind of implementation can be with are as follows:
Characteristic value in the characteristic value sequence is carried out matrixing processing, and is input to default nerve net by step 202
Network；
Step 204, using the objective function of the object module as the objective function of the neural network so that the mind
Decomposition fusion treatment is carried out to the initializaing variable that initializaing variable is concentrated through network, and obtains the optimization knot of the neural network output
Fruit；
Step 206 is based on the optimum results, determines the optimal variables set of the object module.
For step 202 to step 206, specifically can be exemplified are as follows:
Firstly, each characteristic value to be converted to the form of matrix, and the input as neural network；Then, by nerve net
Network carries out resolution process to each initializaing variable, obtains the more fine-grained variable of larger number grade；Then, to more fine-grained
Variable carries out permutation and combination, obtains more multivariable；Then, by each variable of neural network analysis to the disturbance degree of objective function,
And one group of highest variable of disturbance degree is therefrom selected based on disturbance degree, as optimal variables set.
As it can be seen that the present embodiment is in advance the set of the initializaing variable of training pattern configuration by acquisition, and analyze sample number
According to the characteristic value of the sample characteristics in the dimension of each initializaing variable, characteristic value sequence is obtained, then, is based on characteristic value sequence
Column, carry out decomposition fusion to initializaing variable, to analyze the set for influencing maximum optimal variable on model objective function.With it is existing
There is the scheme of artificial allocation models variable in technology to compare, the determination efficiency and effect of model variable can be effectively improved, in turn
Improve model training effect.
Fig. 3 is the flow diagram that a kind of model variable that another embodiment of this specification provides determines method, referring to figure
3, this method can specifically include following steps:
Step 302, the efficiency relevant information for obtaining object module；
Wherein, efficiency relevant information includes at least: model effect for describing effect of the object module under the target scene
Energy, tactful efficiency and scene form situation of change；Model efficiency is for being characterized in certain measurement index drag in different fields
The superiority and inferiority degree of scape performance results is the index of measurement model quality itself, such as: AUC, F1 score etc.；Same model can be with
There are multiple strategies, it is to measure that tactful efficiency, which is used to be characterized under certain measurement index to the superiority and inferiority degree of same model performance,
The index of tactful quality, such as: accuracy rate, coverage rate etc.；Scene form refers to the form of expression of the scene of model application.
Step 304 is based on efficiency relevant information, judges whether trigger model Optimization Steps；
If so, thening follow the steps 306；Otherwise process terminates.
It should be noted that for model efficiency and tactful efficiency, if monitor its index of correlation and drop to certain value,
Risk-warning, and triggering following model optimization step can then be generated；
For scene form situation of change, varied widely if monitoring model application scenarios, such as: by prediction refrigerator
Sales situation change is for television distribution situation, then triggering following model optimization step.
Step 306 carries out variable classification to raw sample data；
Wherein, it there are many variable classification methods, specifically can be exemplified as follows:
Example 1, time series, can judge whether there is time series feature by detection as an important indicator, when
Between sequence signature be used to construct the sequence of sample data, such as: sequence of events is constructed based on Time To Event；
Example 2, text variable judge whether to belong to text variable information by regular expression；
Example 3, index accumulation, judge whether there is apparent temporal characteristics by detection, it is other it is undetected belong to work as
The statistical nature of pen.It can specifically include:
1, a certain amount of accumulation sample sliding window: is carried out to the feature except current sample；
2, feature sliding window: the transformation for carrying out time window to feature itself is accumulated；
3, expansion adjustment, such as each point exponential damping weighted average (∑ a^x, x need line to lower ginseng): are carried out to index
Between carry out being multiplied to determine last value；
4, to addition subtraction is carried out between variable, the method being multiplied Differential Characteristics, integral feature: is used to Partial Feature.
By taking purchase data as an example, the variable of following dimension can be obtained based on above-mentioned steps:
User's dimension: the product number of purchase, the order numbers under history, order frequency and, purchasing channel, the door of purchase
Family, the frequency purchased again and user buy more or less, average numbers of users recently
Product dimension: buying the user of this product, and order numbers order frequency, and purchase rate, recency are averagely added to again
The order numbers of shopping cart
Consumer products cross feature: purchase number, purchase rate, last time buy day again, and last time buys number
User time sequence signature: user most likes one day of purchase, and user most likes the time of purchase in one day.
Step 308, based on variable classification as a result, determine sample event characteristic value sequence；
Shot and long term memory network (Long Short- is used according to the variable classification method of step 306 for sample event
Term Memory, lstm) serializing generation is carried out, obtain sequence signature.Such as: to the purchase row of user in marketing event
Serializing conversion is carried out for ' user A has purchased the parallel of a value 5000 at 10 points in this morning ', obtains characteristic value sequence.
Step 310 is based on the characteristic value sequence, carries out decomposition fusion to initializaing variable, obtains optimal variables set；
Step 312 is based on optimal variables set, optimizes the object module.
For step 310 and step 312, it is specifically as follows:
Characteristic value sequence is executed into matrixing embedding, and the value of embedding is imported into the first of neural network
Layer, to complete the building of its network, being parsed by neural network influences maximum one group of variable to objective function, as final
Variable, and its variable is fused in object module.
As it can be seen that needing to carry out each scene manually under normal conditions when model performance or scene form change
Analysis, according to the difference of scene main body, obtains different characteristic variables.The present embodiment is by carrying out across time window scene main body
The accumulation of mouth, and second generation intelligence trial system is accessed, without manual configuration variable so as to carry out automated characterization generation,
Therefore, people's meat workload can be greatly reduced.Moreover, corresponding template can be used also to do in the present embodiment under other scenes, than
Such as air control scene more convenient can quickly generate some available basic samples in this way, greatly improve Feature Engineering
Modeling efficiency reduces people's meat data cleansing work.
In addition, for simple description, therefore, it is stated as a series of action groups for above method embodiment
It closes, but those skilled in the art should understand that, embodiment of the present invention is not limited by the described action sequence, because
Embodiment according to the present invention, some steps may be performed in other sequences or simultaneously.Secondly, those skilled in the art
It should be aware of, embodiment described in this description belongs to preferred embodiment, and related movement is not necessarily originally
Necessary to invention embodiment.
Fig. 4 is a kind of structural schematic diagram for model variable determining device that one embodiment of this specification provides, referring to fig. 4,
The device can specifically include: determine model 401, first processing module 402 and second processing model 403, wherein
Model 401 is determined, for determining the initializaing variable collection of object module；
First processing module 402, for determining that the corresponding characteristic value sequence of sample data, the characteristic value sequence are used for table
The sample data is levied in the characteristic value of the sample characteristics of each initializaing variable dimension；
Second processing model 403 determines institute for the objective function based on the characteristic value sequence and the object module
State the optimal variables set of object module.
Optionally, the determining model 401, is specifically used for:
Determine the keyword set of the sample data；Based on the keyword set, the initial change of the object module is determined
Quantity set.
Optionally, the determining model 401, is specifically used for:
Clustering processing is carried out to the keyword in the keyword set of each sample data, obtains multiple crucial parts of speech；It determines every
The corresponding characteristic variable of a key part of speech, and generate initializaing variable collection.
Optionally, the initializaing variable that the initializaing variable is concentrated includes: time series variable, content variable, evaluation variable
At least one of.
Optionally, first processing module 402 are specifically used for:
The keyword in the sample data is parsed, sample feature set is obtained；It determines in the sample feature set
Characteristic value of each sample characteristics in corresponding initializaing variable dimension, obtains characteristic value sequence.
Optionally, device further include:
Keyword extracting module, for based on preset variable classification rule, extract in the sample data with each change
The data information that amount type matches；The parsing data information to match with each types of variables, obtains the sample data
Keyword.
Optionally, the data information to match with each types of variables includes at least: numerical value category information and text class letter
Breath；
Wherein, the keyword extracting module, is specifically used for:
The numerical value category information is parsed, time-critical word is obtained；Word segmentation processing is carried out to the text category information, is obtained interior
Hold keyword；Semantic analysis processing is carried out to the text category information, obtains evaluation keyword.
Optionally, the first processing module 402, is specifically used for:
Based on the time-critical word, timed sample sequence feature is generated；Based on the content keyword, content sample is generated
Eigen；Based on the evaluation keyword, evaluation sample characteristics are generated；Based on the timed sample sequence feature, content sample
Feature and evaluation sample characteristics, generate sample feature set.
Optionally, the first processing module 402, is specifically used for:
Based on each initializaing variable dimension, multidimensional coordinate system is generated, the reference axis and initializaing variable of the multidimensional coordinate system are tieed up
Degree corresponds；Determine respective coordinates of each of the sample feature set sample characteristics in the multidimensional coordinate system
Value on axis, and as characteristic value.
Optionally, if institute detects that there are ready to balance sample characteristics, the ready to balance samples in the sample feature set
Feature is the sample characteristics for existing in target sample data and being not present in other sample datas, then device further include:
Equalization control module, for the eigenvalue assignment power for the ready to balance sample characteristics in fisrt feature value sequence
Weight values；Virtual feature value is configured for second feature value sequence, the virtual feature value is special for fictionalizing the ready to balance sample
Levy the characteristic value in second sample feature set；
Wherein, the fisrt feature value sequence is the corresponding characteristic value sequence of the target sample data, and described second is special
Value indicative sequence is the corresponding characteristic value sequence of other described sample datas.
Optionally, the second processing model 403, is specifically used for:
Characteristic value in the characteristic value sequence is subjected to matrixing processing, and is input to default neural network；It will be described
Objective function of the objective function of object module as the neural network, so that the neural network concentrated initializaing variable
Initializaing variable carries out decomposition fusion treatment, and obtains the optimum results of the neural network output；Based on the optimum results, really
The optimal variables set of the fixed object module.
Optionally, the object module is to have trained the model completed, device further include:
Trigger module, if the model efficiency and/or tactful efficiency for detecting the object module are reduced to pre- set
Threshold value is limited, alternatively, detecting that the scene form of model application scenarios changes, then the determining model 401 is triggered and starts work
Make.
As it can be seen that the present embodiment is in advance the set of the initializaing variable of training pattern configuration by acquisition, and analyze sample number
According to the characteristic value of the sample characteristics in the dimension of each initializaing variable, characteristic value sequence is obtained, then, is based on characteristic value sequence
Column, carry out decomposition fusion to initializaing variable, to analyze the set for influencing maximum optimal variable on model objective function.With it is existing
There is the scheme of artificial allocation models variable in technology to compare, the determination efficiency and effect of model variable can be effectively improved, in turn
Improve model training effect.
In addition, for above-mentioned apparatus embodiment, since it is substantially similar to method implementation, so description
Fairly simple, related place illustrates referring to the part of method implementation.It should be noted that in the device of the invention
In all parts, logical partitioning is carried out to component therein according to the function that it to be realized, still, the present invention is not only restricted to
This, can according to need and repartitioned or combined to all parts.
Fig. 5 is the structural schematic diagram for a kind of electronic equipment that one embodiment of this specification provides, and referring to Fig. 5, which is set
Standby includes processor, internal bus, network interface, memory and nonvolatile memory, is also possible that other business certainly
Required hardware.Processor from read in nonvolatile memory corresponding computer program into memory then run, In
Model variable determining device is formed on logic level.Certainly, other than software realization mode, other realities are not precluded in the application
Existing mode, such as logical device or the mode of software and hardware combining etc., that is to say, that the executing subject of following process flow is simultaneously
It is not limited to each logic unit, is also possible to hardware or logical device.
Network interface, processor and memory can be connected with each other by bus system.Bus can be ISA
(Industry Standard Architecture, industry standard architecture) bus, PCI (Peripheral
Component Interconnect, Peripheral Component Interconnect standard) bus or EISA (Extended Industry Standard
Architecture, expanding the industrial standard structure) bus etc..The bus can be divided into address bus, data/address bus, control always
Line etc..Only to be indicated with a four-headed arrow in Fig. 5, it is not intended that an only bus or a type of convenient for indicating
Bus.
Memory is for storing program.Specifically, program may include program code, and said program code includes computer
Operational order.Memory may include read-only memory and random access memory, and provide instruction and data to processor.It deposits
Reservoir may include high-speed random access memory (Random-Access Memory, RAM), it is also possible to further include non-volatile
Memory (non-volatile memory), for example, at least 1 magnetic disk storage.
Processor for executing the program of the memory storage, and specifically executes:
Determine the initializaing variable collection of object module；
Determine the corresponding characteristic value sequence of sample data, the characteristic value sequence is for characterizing the sample data each first
The characteristic value of the sample characteristics of beginning variable dimension；
Objective function based on the characteristic value sequence and the object module determines the optimal variable of the object module
Collection.
Model variable determining device or manager (Master) node disclosed in the above-mentioned embodiment illustrated in fig. 4 such as the application are held
Capable method can be applied in processor, or be realized by processor.Processor may be a kind of IC chip, have
The processing capacity of signal.During realization, each step of the above method can pass through the integration logic of the hardware in processor
The instruction of circuit or software form is completed.Above-mentioned processor can be general processor, including central processing unit
(Central Processing Unit, CPU), network processing unit (Network Processor, NP) etc.；It can also be number
Signal processor (Digital Signal Processor, DSP), specific integrated circuit (Application Specific
Integrated Circuit, ASIC), field programmable gate array (Field-Programmable Gate Array,
FPGA) either other programmable logic device, discrete gate or transistor logic, discrete hardware components.May be implemented or
Person executes disclosed each method, step and logic diagram in the embodiment of the present application.General processor can be microprocessor or
Person's processor is also possible to any conventional processor etc..The step of method in conjunction with disclosed in the embodiment of the present application, can be straight
Connect and be presented as that hardware decoding processor executes completion, or in decoding processor hardware and software module combination executed
At.Software module can be located at random access memory, and flash memory, read-only memory, programmable read only memory or electrically-erasable can
In the storage medium of this fields such as programmable memory, register maturation.The storage medium is located at memory, and processor reads storage
Information in device, in conjunction with the step of its hardware completion above method.
The method that model variable determining device can also carry out Fig. 1-3, and realize the method that manager's node executes.
Based on identical innovation and creation, the embodiment of the present application also provides a kind of computer readable storage medium, the meter
Calculation machine readable storage medium storing program for executing stores one or more programs, and one or more of programs are when by the electricity including multiple application programs
When sub- equipment executes, so that the electronic equipment executes the model variable that the corresponding embodiment of Fig. 1-3 provides and determines method.
All the embodiments in this specification are described in a progressive manner, same and similar portion between each embodiment
Dividing may refer to each other, and each embodiment focuses on the differences from other embodiments.Especially for system reality
For applying example, since it is substantially similar to the method embodiment, so being described relatively simple, related place is referring to embodiment of the method
Part explanation.
It is above-mentioned that this specification specific embodiment is described.Other embodiments are in the scope of the appended claims
It is interior.In some cases, the movement recorded in detail in the claims or step can be come according to the sequence being different from embodiment
It executes and desired result still may be implemented.In addition, process depicted in the drawing not necessarily require show it is specific suitable
Sequence or consecutive order are just able to achieve desired result.In some embodiments, multitasking and parallel processing be also can
With or may be advantageous.
It should be understood by those skilled in the art that, the embodiment of the present invention can provide as method, system or computer program
Product.Therefore, complete hardware embodiment, complete software embodiment or reality combining software and hardware aspects can be used in the present invention
Apply the form of example.Moreover, it wherein includes the computer of computer usable program code that the present invention, which can be used in one or more,
The computer program implemented in usable storage medium (including but not limited to magnetic disk storage, CD-ROM, optical memory etc.) produces
The form of product.
The present invention be referring to according to the method for the embodiment of the present invention, the process of equipment (system) and computer program product
Figure and/or block diagram describe.It should be understood that every one stream in flowchart and/or the block diagram can be realized by computer program instructions
The combination of process and/or box in journey and/or box and flowchart and/or the block diagram.It can provide these computer programs
Instruct the processor of general purpose computer, special purpose computer, Embedded Processor or other programmable data processing devices to produce
A raw machine, so that being generated by the instruction that computer or the processor of other programmable data processing devices execute for real
The device for the function of being specified in present one or more flows of the flowchart and/or one or more blocks of the block diagram.
These computer program instructions, which may also be stored in, is able to guide computer or other programmable data processing devices with spy
Determine in the computer-readable memory that mode works, so that it includes referring to that instruction stored in the computer readable memory, which generates,
Enable the manufacture of device, the command device realize in one box of one or more flows of the flowchart and/or block diagram or
The function of being specified in multiple boxes.
These computer program instructions also can be loaded onto a computer or other programmable data processing device, so that counting
Series of operation steps are executed on calculation machine or other programmable devices to generate computer implemented processing, thus in computer or
The instruction executed on other programmable devices is provided for realizing in one or more flows of the flowchart and/or block diagram one
The step of function of being specified in a box or multiple boxes.
In a typical configuration, calculating equipment includes one or more processors (CPU), input/output interface, net
Network interface and memory.
Memory may include the non-volatile memory in computer-readable medium, random access memory (RAM) and/or
The forms such as Nonvolatile memory, such as read-only memory (ROM) or flash memory (flash RAM).Memory is computer-readable medium
Example.
Computer-readable medium includes permanent and non-permanent, removable and non-removable media can be by any method
Or technology come realize information store.Information can be computer readable instructions, data structure, the module of program or other data.
The example of the storage medium of computer includes, but are not limited to phase change memory (PRAM), static random access memory (SRAM), moves
State random access memory (DRAM), other kinds of random access memory (RAM), read-only memory (ROM), electric erasable
Programmable read only memory (EEPROM), flash memory or other memory techniques, read-only disc read only memory (CD-ROM) (CD-ROM),
Digital versatile disc (DVD) or other optical storage, magnetic cassettes, tape magnetic disk storage or other magnetic storage devices
Or any other non-transmission medium, can be used for storage can be accessed by a computing device information.As defined in this article, it calculates
Machine readable medium does not include temporary computer readable media (transitory media), such as the data-signal and carrier wave of modulation.
It should also be noted that, the terms "include", "comprise" or its any other variant are intended to nonexcludability
It include so that the process, method, commodity or the equipment that include a series of elements not only include those elements, but also to wrap
Include other elements that are not explicitly listed, or further include for this process, method, commodity or equipment intrinsic want
Element.In the absence of more restrictions, the element limited by sentence "including a ...", it is not excluded that including described want
There is also other identical elements in the process, method of element, commodity or equipment.
It will be understood by those skilled in the art that embodiments herein can provide as method, system or computer program product.
Therefore, complete hardware embodiment, complete software embodiment or embodiment combining software and hardware aspects can be used in the application
Form.It is deposited moreover, the application can be used to can be used in the computer that one or more wherein includes computer usable program code
The shape for the computer program product implemented on storage media (including but not limited to magnetic disk storage, CD-ROM, optical memory etc.)
Formula.
The above description is only an example of the present application, is not intended to limit this application.For those skilled in the art
For, various changes and changes are possible in this application.All any modifications made within the spirit and principles of the present application are equal
Replacement, improvement etc., should be included within the scope of the claims of this application.
Claims (15)
1. a kind of model variable determines method, comprising:
Determine the initializaing variable collection of object module；
Determine the corresponding characteristic value sequence of sample data, the characteristic value sequence is for characterizing the sample data in each initial change
Measure the characteristic value of the sample characteristics of dimension；
Objective function based on the characteristic value sequence and the object module determines the optimal variables set of the object module.
2. according to the method described in claim 1, the initializaing variable collection of the determining object module, comprising:
Determine the keyword set of the sample data；
Based on the keyword set, the initializaing variable collection of the object module is determined.
3. determining the initializaing variable of the object module according to the method described in claim 2, described be based on the keyword set
Collection, comprising:
Clustering processing is carried out to the keyword in the keyword set of each sample data, obtains multiple crucial parts of speech；
It determines the corresponding characteristic variable of each crucial part of speech, and generates initializaing variable collection.
4. according to the method described in claim 2, the initializaing variable concentrate initializaing variable include: time series variable, it is interior
Hold at least one of variable, evaluation variable.
5. according to the method described in claim 1, the corresponding characteristic value sequence of the determining sample data, comprising:
The keyword in the sample data is parsed, sample feature set is obtained；
It determines characteristic value of each of the sample feature set sample characteristics in corresponding initializaing variable dimension, obtains
Characteristic value sequence.
6. according to the method described in claim 5, further include:
Based on preset variable classification rule, the data information to match in the sample data with each types of variables is extracted；
The parsing data information to match with each types of variables, obtains the keyword of the sample data.
7. according to the method described in claim 6, the data information to match with each types of variables includes at least: numerical value class
Information and text category information；
Wherein, the data information to match described in the parsing with each types of variables obtains the keyword of the sample data, packet
It includes:
The numerical value category information is parsed, time-critical word is obtained；
Word segmentation processing is carried out to the text category information, obtains content keyword；
Semantic analysis processing is carried out to the text category information, obtains evaluation keyword.
8. according to the method described in claim 7, the keyword in the parsing sample data, obtains sample characteristics collection
It closes, comprising:
Based on the time-critical word, timed sample sequence feature is generated；
Based on the content keyword, content sample feature is generated；
Based on the evaluation keyword, evaluation sample characteristics are generated；
Based on the timed sample sequence feature, content sample feature and evaluation sample characteristics, sample feature set is generated.
9. according to the method described in claim 5, each of described sample feature set of determination sample characteristics are right
The characteristic value in initializaing variable dimension answered, comprising:
Based on each initializaing variable dimension, multidimensional coordinate system, the reference axis and initializaing variable dimension one of the multidimensional coordinate system are generated
One is corresponding；
Determine each of sample feature set sample characteristics on the respective coordinates axis in the multidimensional coordinate system
Value, and as characteristic value.
10. according to the method described in claim 9, if detecting in the sample feature set there are ready to balance sample characteristics,
The ready to balance sample characteristics are the sample characteristics for existing in target sample data and being not present in other sample datas, then method
Further include:
For the eigenvalue assignment weighted value of the ready to balance sample characteristics in fisrt feature value sequence；
Virtual feature value is configured for second feature value sequence, the virtual feature value is for fictionalizing the ready to balance sample characteristics
In the characteristic value of second sample feature set；
Wherein, the fisrt feature value sequence is the corresponding characteristic value sequence of the target sample data, the Second Eigenvalue
Sequence is the corresponding characteristic value sequence of other described sample datas.
11. according to the method described in claim 1, the target letter based on the characteristic value sequence and the object module
Number, determines the optimal variables set of the object module, comprising:
Characteristic value in the characteristic value sequence is subjected to matrixing processing, and is input to default neural network；
Using the objective function of the object module as the objective function of the neural network, so that the neural network is to initial
Initializaing variable in variables set carries out decomposition fusion treatment, and obtains the optimum results of the neural network output；
Based on the optimum results, the optimal variables set of the object module is determined.
12. according to the method described in claim 1, the object module be trained complete model, the method also includes:
If detecting, the model efficiency of the object module and/or tactful efficiency are reduced to predetermined lower threshold value, alternatively, detecting
The step of scene form of model application scenarios changes, then triggers the initializaing variable collection for executing the determining object module.
13. a kind of model variable determining device, comprising:
Model is determined, for determining the initializaing variable collection of object module；
First processing module, for determining that the corresponding characteristic value sequence of sample data, the characteristic value sequence are described for characterizing
Characteristic value of the sample data in the sample characteristics of each initializaing variable dimension；
Second processing model determines the target for the objective function based on the characteristic value sequence and the object module
The optimal variables set of model.
14. a kind of electronic equipment, comprising:
Processor；And
It is arranged to the memory of storage computer executable instructions, the executable instruction makes the processor when executed
The step of executing the method as described in any one of claims 1 to 12.
15. a kind of computer readable storage medium, computer program, the meter are stored on the computer readable storage medium
The step of method as described in any one of claims 1 to 12 is realized when calculation machine program is executed by processor.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910626175.6A CN110472742B (en) | 2019-07-11 | 2019-07-11 | Model variable determination method, device and equipment |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910626175.6A CN110472742B (en) | 2019-07-11 | 2019-07-11 | Model variable determination method, device and equipment |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN110472742A true CN110472742A (en) | 2019-11-19 |
| CN110472742B CN110472742B (en) | 2023-04-04 |
Family
ID=68508012
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910626175.6A Active CN110472742B (en) | 2019-07-11 | 2019-07-11 | Model variable determination method, device and equipment |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN110472742B (en) |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111626898A (en) * | 2020-03-20 | 2020-09-04 | 贝壳技术有限公司 | Method, device, medium and electronic equipment for realizing attribution of events |
| WO2021196843A1 (en) * | 2020-03-31 | 2021-10-07 | 支付宝(杭州)信息技术有限公司 | Derived variable selection method and apparatus for risk identification model |
| CN116402113A (en) * | 2023-06-08 | 2023-07-07 | 之江实验室 | Task execution method and device, storage medium and electronic equipment |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20180060730A1 (en) * | 2016-08-29 | 2018-03-01 | International Business Machines Corporation | Leveraging class information to initialize a neural network langauge model |
| CN108510080A (en) * | 2018-03-21 | 2018-09-07 | 华南理工大学 | A kind of multi-angle metric learning method based on DWH model many-many relationship type data |
| CN109933662A (en) * | 2019-02-15 | 2019-06-25 | 北京奇艺世纪科技有限公司 | Model training method, information generating method, device, electronic equipment and computer-readable medium |
-
2019
- 2019-07-11 CN CN201910626175.6A patent/CN110472742B/en active Active
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20180060730A1 (en) * | 2016-08-29 | 2018-03-01 | International Business Machines Corporation | Leveraging class information to initialize a neural network langauge model |
| CN108510080A (en) * | 2018-03-21 | 2018-09-07 | 华南理工大学 | A kind of multi-angle metric learning method based on DWH model many-many relationship type data |
| CN109933662A (en) * | 2019-02-15 | 2019-06-25 | 北京奇艺世纪科技有限公司 | Model training method, information generating method, device, electronic equipment and computer-readable medium |
Non-Patent Citations (1)
| Title |
|---|
| 袁宏源等: "多站径流的模糊随机生成模型", 《武汉水利电力大学学报》 * |
Cited By (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111626898A (en) * | 2020-03-20 | 2020-09-04 | 贝壳技术有限公司 | Method, device, medium and electronic equipment for realizing attribution of events |
| CN111626898B (en) * | 2020-03-20 | 2022-03-15 | 贝壳找房(北京)科技有限公司 | Method, device, medium and electronic equipment for realizing attribution of events |
| WO2021196843A1 (en) * | 2020-03-31 | 2021-10-07 | 支付宝(杭州)信息技术有限公司 | Derived variable selection method and apparatus for risk identification model |
| CN116402113A (en) * | 2023-06-08 | 2023-07-07 | 之江实验室 | Task execution method and device, storage medium and electronic equipment |
| CN116402113B (en) * | 2023-06-08 | 2023-10-03 | 之江实验室 | Task execution method and device, storage medium and electronic equipment |
Also Published As
| Publication number | Publication date |
|---|---|
| CN110472742B (en) | 2023-04-04 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN106485562B (en) | Commodity information recommendation method and system based on user historical behaviors | |
| CN107689008A (en) | A kind of user insures the method and device of behavior prediction | |
| CN111475680A (en) | Method, device, equipment and storage medium for detecting abnormal high-density subgraph | |
| CN108921569B (en) | Method and device for determining complaint type of user | |
| CN110472742A (en) | A kind of model variable determines method, device and equipment | |
| CN107230108A (en) | The processing method and processing device of business datum | |
| US10394631B2 (en) | Anomaly detection and automated analysis using weighted directed graphs | |
| CN112329816A (en) | Data classification method and device, electronic equipment and readable storage medium | |
| CN110069545B (en) | Behavior data evaluation method and device | |
| CN107316200B (en) | Method and device for analyzing user behavior period | |
| CN113656558B (en) | Method and device for evaluating association rule based on machine learning | |
| CN103365829A (en) | Information processing apparatus, information processing method, and program | |
| JP2022546571A (en) | Generate training data for machine learning models | |
| CN106327230B (en) | Abnormal user detection method and equipment | |
| CN111079937A (en) | Rapid modeling method | |
| CN111242318A (en) | Business model training method and device based on heterogeneous feature library | |
| CN110458600A (en) | Portrait model training method, device, computer equipment and storage medium | |
| CN112884569A (en) | Credit assessment model training method, device and equipment | |
| CN112070559A (en) | State acquisition method and device, electronic equipment and storage medium | |
| CN116362823A (en) | Recommendation model training method, recommendation method and recommendation device for behavior sparse scene | |
| CN112328869A (en) | User loan willingness prediction method and device and computer system | |
| CN108647714A (en) | Acquisition methods, terminal device and the medium of negative label weight | |
| CN109460474B (en) | User preference trend mining method | |
| CN108090794A (en) | A kind of merchandise items sort method and device based on dynamic sliding time window | |
| CN107291722B (en) | Descriptor classification method and 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 |