CN109784377A - Multiple recognition model building method, device, computer equipment and storage medium - Google Patents
Multiple recognition model building method, device, computer equipment and storage medium Download PDFInfo
- Publication number
- CN109784377A CN109784377A CN201811601941.5A CN201811601941A CN109784377A CN 109784377 A CN109784377 A CN 109784377A CN 201811601941 A CN201811601941 A CN 201811601941A CN 109784377 A CN109784377 A CN 109784377A
- Authority
- CN
- China
- Prior art keywords
- data
- vector
- identification
- training
- model
- 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.)
- Pending
Links
Landscapes
- Image Analysis (AREA)
Abstract
The invention discloses Multiple recognition model building method, device, computer equipment and storage mediums.This method is by obtaining source data, pretreatment and vector combination are carried out to source data, it obtains primitive character matrix and the column vector wherein chosen is subjected to statistical calculation to obtain newly-increased column vector, column vector low with the Pearson correlation coefficient of each column vector in primitive character matrix in newly-increased column vector is deleted, rear eigenmatrix is simplified;Eigenmatrix after simplification is increased by a column result vector and obtains raw data set;The row vector for extracting default line number by row is concentrated to concentrate non-extracted row vector as test set as training set, and using initial data in initial data;And it is obtained to including being trained to training pattern for multiple submodels for the anti-identification model for cheating identification by training set.The method achieve by training set to it is multiple son simultaneously models be trained include to obtain Multiple Velocity Model identification model, improve the recognition accuracy of identification model.
Description
Technical field
The present invention relates to prediction model technical field more particularly to a kind of Multiple recognition model building methods, device, calculating
Machine equipment and storage medium.
Background technique
Current anti-fraud identifying system is more based on single model, and single model has quasi- on specific field
The advantage closed and predicted, and on other fields or scene, do not have advantage;Such as linear regression model (LRM), in linear mould
In type have advantage, and neural network model on nonlinear model have advantage, due to single model on specific field
Tool advantage is not easy to promote the comprehensive performance of single model.
And when carrying out model training to single model, used data are usually initial data, to initial data
It pre-processes less, causes data dimension low, reduce accuracy of the single model in specific identification.
Summary of the invention
The embodiment of the invention provides a kind of Multiple recognition model building method, device, computer equipment and storage medium,
Aim to solve the problem that anti-fraud identifying system is more to carry out model training when institute to single model based on single model in the prior art
The data of use are usually initial data, reduce accuracy of the single model in specific identification, and when single model uses by
The problem of specific field or scene restriction.
In a first aspect, the embodiment of the invention provides a kind of Multiple recognition model building methods comprising:
Source data is obtained, source data is pre-processed, data after being handled;
Using the data at time point each in data after processing as a row vector, by data after processing in chronological sequence
Sequentially corresponding row vector sequentially combines from top to bottom, obtains primitive character matrix;
The column vector chosen in primitive character matrix is subjected to statistical calculation to obtain newly-increased column vector, deletion is newly added
It is located at preset ranking threshold with the size ranking of the Pearson correlation coefficient of each column vector in primitive character matrix in vector
Column vector after value is simplified rear eigenmatrix;
It will increase by a column result vector in eigenmatrix after simplification and obtain raw data set;
The row vector for extracting preset percentage by row is concentrated to concentrate not as training set, and by initial data in initial data
The row vector being extracted is as test set;And
By training set to including being trained to training pattern for multiple submodels, obtain for the anti-knowledge for cheating identification
Other model.
Second aspect, the embodiment of the invention provides a kind of Multiple recognition model construction devices comprising:
Pretreatment unit pre-processes source data for obtaining source data, data after being handled;
Original matrix acquiring unit, the data for each time point in data after handling, will as a row vector
In chronological sequence the corresponding row vector of sequence sequentially combines from top to bottom in data after processing, obtains primitive character matrix;
Simplification matrix acquiring unit, for the column vector chosen in primitive character matrix to be carried out statistical calculation to obtain
Newly-increased column vector deletes the size in newly-increased column vector with the Pearson correlation coefficient of each column vector in primitive character matrix
Ranking is located at the column vector after preset rank threshold, is simplified rear eigenmatrix;
Raw data set acquiring unit obtains initial data for increasing by a column result vector in eigenmatrix after simplifying
Collection;
Raw data set split cells extracts the row vector of preset percentage as instruction by row for concentrating in initial data
Practice collection, and concentrates non-extracted row vector as test set initial data;
Model training unit, for, to including being trained to training pattern for multiple submodels, being obtained by training set
For the anti-identification model for cheating identification.
The third aspect, the embodiment of the present invention provide a kind of computer equipment again comprising memory, processor and storage
On the memory and the computer program that can run on the processor, the processor execute the computer program
Multiple recognition model building method described in the above-mentioned first aspect of Shi Shixian.
Fourth aspect, the embodiment of the invention also provides a kind of computer readable storage mediums, wherein the computer can
It reads storage medium and is stored with computer program, it is above-mentioned that the computer program when being executed by a processor executes the processor
Multiple recognition model building method described in first aspect.
The embodiment of the invention provides a kind of Multiple recognition model building method, device, computer equipment and storage mediums.
This method pre-processes source data, data after being handled by obtaining source data;Each time in data after handling
The data of point are as a row vector, by the in chronological sequence corresponding row vector of sequence sequentially group from top to bottom in data after processing
It closes, obtains primitive character matrix;By the column vector chosen in primitive character matrix carry out statistical calculation with newly added to
Amount is deleted in newly-increased column vector and is located at the size ranking of the Pearson correlation coefficient of each column vector in primitive character matrix
Column vector after preset rank threshold is simplified rear eigenmatrix;It will increase by a column result in eigenmatrix after simplification
Vector obtains raw data set;It concentrates by the row vector of row extraction preset percentage in initial data as training set, and will be former
Non-extracted row vector is as test set in beginning data set;And by training set to the mould to be trained including multiple submodels
Type is trained, and is obtained for the anti-identification model for cheating identification.The method achieve pass through training set to multiple sons while moulds
Type be trained include to obtain Multiple Velocity Model identification model, improve the recognition accuracy of identification model.
Detailed description of the invention
Technical solution in order to illustrate the embodiments of the present invention more clearly, below will be to needed in embodiment description
Attached drawing is briefly described, it should be apparent that, drawings in the following description are some embodiments of the invention, general for this field
For logical technical staff, without creative efforts, it is also possible to obtain other drawings based on these drawings.
Fig. 1 is the flow diagram of Multiple recognition model building method provided in an embodiment of the present invention;
Fig. 2 is the sub-process schematic diagram of Multiple recognition model building method provided in an embodiment of the present invention;
Fig. 3 is another sub-process schematic diagram of Multiple recognition model building method provided in an embodiment of the present invention;
Fig. 4 is another sub-process schematic diagram of Multiple recognition model building method provided in an embodiment of the present invention;
Fig. 5 is another flow diagram of Multiple recognition model building method provided in an embodiment of the present invention;
Fig. 6 is another flow diagram of Multiple recognition model building method provided in an embodiment of the present invention;
Fig. 7 is another flow diagram of Multiple recognition model building method provided in an embodiment of the present invention;
Fig. 8 is the schematic block diagram of Multiple recognition model construction device provided in an embodiment of the present invention;
Fig. 9 is the subelement schematic block diagram of Multiple recognition model construction device provided in an embodiment of the present invention;
Figure 10 is another subelement schematic block diagram of Multiple recognition model construction device provided in an embodiment of the present invention;
Figure 11 is another subelement schematic block diagram of Multiple recognition model construction device provided in an embodiment of the present invention;
Figure 12 is another schematic block diagram of Multiple recognition model construction device provided in an embodiment of the present invention;
Figure 13 is another schematic block diagram of Multiple recognition model construction device provided in an embodiment of the present invention;
Figure 14 is another schematic block diagram of Multiple recognition model construction device provided in an embodiment of the present invention;
Figure 15 is the schematic block diagram of computer equipment provided in an embodiment of the present invention.
Specific embodiment
Following will be combined with the drawings in the embodiments of the present invention, and technical solution in the embodiment of the present invention carries out clear, complete
Site preparation description, it is clear that described embodiments are some of the embodiments of the present invention, instead of all the embodiments.Based on this hair
Embodiment in bright, every other implementation obtained by those of ordinary skill in the art without making creative efforts
Example, shall fall within the protection scope of the present invention.
It should be appreciated that ought use in this specification and in the appended claims, term " includes " and "comprising" instruction
Described feature, entirety, step, operation, the presence of element and/or component, but one or more of the other feature, whole is not precluded
Body, step, operation, the presence or addition of element, component and/or its set.
It is also understood that mesh of the term used in this description of the invention merely for the sake of description specific embodiment
And be not intended to limit the present invention.As description of the invention and it is used in the attached claims, unless on
Other situations are hereafter clearly indicated, otherwise " one " of singular, "one" and "the" are intended to include plural form.
It will be further appreciated that the term "and/or" used in description of the invention and the appended claims is
Refer to any combination and all possible combinations of one or more of associated item listed, and including these combinations.
Referring to Fig. 1, Fig. 1 is the flow diagram of Multiple recognition model building method provided in an embodiment of the present invention, it should
Multiple recognition model building method is applied in management server, and this method passes through the application software that is installed in management server
It is executed, management server is the enterprise terminal for carrying out Multiple recognition model construction.
As shown in Figure 1, the method comprising the steps of S110~S160.
S110, source data is obtained, source data is pre-processed, data after being handled.
In the present embodiment, acquired source data can be the data under multiple application scenarios.Such as it settles a claim in vehicle insurance
Under scene, acquired source data can be the driving recording data of automobile data recorder, such as source data collected is to come from
In vehicle insurance APP, when user drives, spontaneous starting GPS positioning module is to GPS data into acquisition, and wherein source data includes longitude and latitude
The time series datas such as degree, speed, height above sea level and direction, period are that every 1s acquires a data.
Since GPS positioning module source data collected is related with satellite-signal, it is easy in the place of high building
There is the inaccurate situation of longitude and latitude data;And since APP needs outgoing data packet, in order to guarantee that data packet has been sent into
Function can inevitably encounter data packet and send out the case where leading to Data duplication again.In order to ensure understanding the abnormal data in source data, need
Source data is pre-processed.
In one embodiment, as shown in Fig. 2, step S110 includes:
If increasing degree of the included time series data there are current data with respect to last data exceeds in S111, source data
The value of current data delete or the value of current data is set to last data * amount of increase threshold by preset amount of increase threshold value
Value, obtains data after dealing of abnormal data;
S112, data are taken out by preset time cycle constant duration in number after dealing of abnormal data, is referred to
Mark data;
S113, achievement data is normalized, data after being handled.
In the present embodiment, in order to eliminate the abnormal data in source data, can judge according to the sequencing at time point
Included time series data exceeds preset amount of increase with respect to the increasing degree of last data with the presence or absence of current data in source data
Threshold value (such as amount of increase threshold value is set as 150%).If increasing degree of the time series data there are current data with respect to last data exceeds
Preset amount of increase threshold value then indicates that there are abnormal datas, and two ways can be used at this time and handled, and one is delete the timing
Data, another kind are that the value of the time series data is set to last data * amount of increase threshold value, obtain data after dealing of abnormal data.
Then, data are taken out to number medium time interval after dealing of abnormal data, obtains achievement data.Constant duration
Taking-up data, be the time-serial position in order to construct relative smooth.
Finally achievement data is normalized, is to be standardized data in order to subsequent mould
Type training.
Wherein, when logarithm type data are normalized, numerical value normalizing is carried out using with minimax laws for criterion
Change, the formula of minimax laws for criterion are as follows:Wherein XnomIndicate normalization after numerical value, X indicate to
The numerical value of normalized, XminIndicate the minimum value of the parameter, XmaxIndicate the maximum value of the parameter.To classification type data, adopt
With one-hot coding.
S120, using the data at time point each in data after processing as a row vector, by data after processing on time
Between the corresponding row vector of sequencing sequentially combine from top to bottom, obtain primitive character matrix.
In the present embodiment, the corresponding timing of time point t is 100,100,60,300,230 (preceding two such as in source data
A numerical value respectively represents longitude and dimension, third numerical value representation speed, and the 4th numerical value represents height above sea level, the 5th numerical value
Represent direction), corresponding row vector is [0.2 0.3 0.5 0.1 0.3];The corresponding row vector of time point 2t is [0.2
0.1 0.1 0.3 0.2];..., the corresponding row vector of time point Nt is [0.1 0.1 0.1 0.1 0.1].By this N number of row to
Amount sequentially combines from top to bottom according to chronological order, obtains the primitive character matrix of a N row 5 column.By by source data
It is converted to primitive character matrix according to timing, initial data can be normalized standard processing using the standard as training pattern
Input data.
S130, the column vector chosen in primitive character matrix is subjected to statistical calculation to obtain newly-increased column vector, deleted
With the size ranking of the Pearson correlation coefficient of each column vector in primitive character matrix positioned at preset in newly-increased column vector
Column vector after rank threshold is simplified rear eigenmatrix.
In this embodiment, the dimension of primitive character matrix is less, can will be in primitive character matrix in order to promote its dimension
The column vector chosen carries out statistical calculation to obtain newly-increased column vector realization and extend to the dimension of primitive character matrix.For example,
The primitive character matrix that obtained N row 5 arranges, after being extended using statistical calculation, after obtaining the increase feature arranged for N row M
Matrix.In order to retain increase feature after increase the higher column vector of data value in column vector in matrix newly, can delete and newly add
It is located at preset ranking threshold with the size ranking of the Pearson correlation coefficient of each column vector in primitive character matrix in vector
Column vector after value is simplified rear eigenmatrix.
Wherein, the Pearson correlation coefficient between any two column vector is calculated, can be calculate by the following formula:
Wherein, E indicates mathematic expectaion;
ρX,YValue range be (0,1), work as ρX,YIt indicates that the similarity degree of two column vectors is higher closer to 1, works as ρX,YMore
Indicate that the similarity degree of two column vectors is lower close to 0.
In one embodiment, as shown in figure 3, step S130 includes:
S131, preset statistical calculation species number is obtained;
Two column vectors in S132, random selection primitive character matrix, obtain identical with the statistical calculation species number
Column vector combination;
S133, combine each column vector and calculate by one type operation in statistical calculation, obtain number be greater than or
Equal to the newly-increased column vector of the statistical calculation species number.
Wherein, the statistical calculation includes summation operation, asks difference operation, multiplying and division arithmetic.For example, when default
Statistical calculation species number be 4 kinds, be summation operation respectively, ask difference operation, multiplying and division arithmetic.At this time in original spy
2 column vectors of selection random in matrix are levied, a and b is denoted as respectively, then sums respectively to each numerical value in the two column vectors
Operation a+b, difference operation a-b, multiplying a*b and division arithmetic a/b are asked.
For example, primitive character matrix is as follows:
For the data for arranging primitive character matrix-expand for 2 rows 9, the first column data can be added work with the second column data
For the 6th column data, the second column data and third column data are subtracted each other as the 7th column data, the first column data and third columns
It is used as the 8th column data according to being multiplied, third column data and the 5th column data are divided by as the 9th column data, and obtaining following matrix (can
To be interpreted as increasing matrix after feature):
The 4 column column vectors newly increased in matrix after increasing feature seek Pearson came phase with the column vector of front 5 respectively respectively
Relationship number, wherein the Pearson correlation coefficient between the column vector that the column vector of the 6th column and the column vector of first row are arranged to the 5th
RespectivelyChoose maximum correlation coefficient thereinColumn vector as the 6th column
With the related coefficient in the column vector of first five column.
Similarly, the related coefficient ... ... for calculating the column vector of the 7th column and the column vector of first five column, when obtaining each column
With the related coefficient of the column vector of first five column, can and will increase after feature in matrix in each column vector and original feature vector
Each column vector calculation Pearson correlation coefficient, if there are in column vector and primitive character matrix in vector after increasing feature
The size ranking of the Pearson correlation coefficient of each column vector is located at after preset rank threshold, and corresponding column vector is carried out
It deletes, is simplified rear eigenmatrix, such as preset rank threshold is 2, if then assuming the column vector and the 8th column of the 6th column
Column vector be not deleted column vector, then eigenmatrix is as follows after simplifying:
The dimension of primitive character matrix can be quicklyd increase through the above way, and can be more by primitive character matrix-expand
Eigenmatrix after the simplification of embodiment source data feature comprehensively.
S140, it will increase by a column result vector in eigenmatrix after simplification and obtain raw data set.
In the present embodiment, it is trained, each row vector of eigenmatrix after simplification need to be made in order to treat training pattern
For the input to training pattern, while needing to increase in eigenmatrix after simplification a column result vector also with each row vector of correspondence
And using the corresponding end value of each row vector as the output to training pattern, the training for treating training pattern can be completed.
For example, eigenmatrix is as follows after simplifying:
Raw data set is obtained after increasing by a column result vector are as follows:
Initial data concentrates each row vector to represent a training data, wherein preceding 7 data are made in each row vector
For the input to training pattern, the last one data in each row vector are as the output to training pattern.
S150, it is concentrated in initial data and extracts the row vector of preset percentage as training set by row, and by initial data
Concentrate non-extracted row vector as test set.
In the present embodiment, preset percentage is taken to indicate that the line number of extracted row vector accounts for the total of entire raw data set
The percentage of line number, such as preset percentage is taken to may be configured as 70% indicates to choose initial data and concentrates 70% (i.e. preset the
One ratio is row vector 70%) as training set, will not be chosen as the row vector of the residue 30% of training set as test set.
By the way that raw data set is split as training set and test set, it can realize the training for treating training pattern, it can also be to instruction
The identification model got is tested.
S160, by training set to including that multiple submodels being trained to training pattern, obtain knowing for instead cheating
Other identification model.
In the present embodiment, selected submodel have support vector machines (i.e. SupportVector Machine, SVM,
It is supervised learning model related to relevant learning algorithm, can analyze data, recognition mode, for classifying and returning point
Analysis), XGBoost model (eXtreme Gradient Boosting, it can be understood as extreme gradient is promoted, XGBoost model
Be exactly a kind of monitor model), Random Forest model etc..
By training set to including being trained to training pattern for multiple submodels, it includes polynary for can be obtained one
The Multiple recognition model of submodel can accurately judge data to be identified.And since identification model is multiple
Model can efficiently and dynamically adjust entire multi-model cooperation, have stronger fitness to different data set.
In one embodiment, as shown in figure 4, step S160 includes:
The total packet number of S161, acquisition to submodel in training pattern;
S162, training set is subjected to equal part according to total packet number, obtains sub- training set identical with the several numbers of total packet;
S163, carry out model training for each sub- training set as the training data of corresponding grouping submodel, obtain with
Submodel identifies submodel correspondingly, to form multiple identification model.
In the present embodiment, first obtain the total packet number to submodel in training pattern, be in order to according to total packet number into
The equal part of the corresponding number of row.For example, the total packet number to submodel in training pattern is 3, then training set is divided into 3 equal parts,
Each sub- training set carries out model training respectively as the training data of corresponding grouping submodel, obtains corresponding with submodel
Identification submodel, to form multiple identification model.In addition to using the above-mentioned total packet number according to submodel to training set etc.
Point mode treat each submodel in training pattern and be trained, training set can also not divided directly but directly with
Training set is respectively trained each submodel.Using including multiple submodels to training pattern, it can increase or reduce son
Model has stronger scalability.
In one embodiment, as shown in figure 5, after step S160 further include:
S171, test set is inputted into the identification model progress model verifying, if input vector included in test set
The difference for importing the end value in the result and test set that identification model is exported is less than default error threshold, and identification model is by testing
It demonstrate,proves and saves the identification model.
In the present embodiment, in order to ensure the recognition accuracy of identification model, identification mould is being obtained by training set training
After type, test set need to be inputted to the identification model and carry out model verifying.When the input vector in test set is input to identification
The difference of model, obtained output result and the corresponding end value in test set is (included i.e. in test set within the allowable range
The difference of end value that imports in the identification model result and test set that are exported of input vector be less than default error threshold), then
Indicate the identification model can by verifying and it is specifically used.Once the identification model is not verified by model, then it represents that selected
The submodel selected is unreasonable, needs to input more training datas again to be trained to the submodel.It is verified by increasing
Link can effectively ensure the availability and recognition accuracy of identification model.
In one embodiment, as shown in fig. 6, after step S160 further include:
S1721, data to be identified are input to identification model, are corresponded to obtain sub- identification knot according to each submodel
Fruit;
S1722, each sub- recognition result multiplied by the default weighted value of corresponding submodel and is summed, obtains recognition result.
In the present embodiment, as the first specific embodiment for obtaining recognition result, in order to make multi-model assist to be known
It not as a result, can be for the corresponding default weighted value of each submodel, by each sub- recognition result multiplied by the default weight of corresponding submodel
It is worth and sums, obtain recognition result, can efficiently and dynamically adjusts entire multi-model cooperation.
In one embodiment, as shown in fig. 7, after step S160 further include:
S1731, data to be identified are input to identification model, are corresponded to obtain sub- identification knot according to each submodel
Fruit;
S1732, the sub- recognition result that numerical value output frequency in sub- recognition result is maximum value is obtained, and as identification model
Recognition result.
It in the present embodiment, can also be defeated by counting a subpattern as the second specific embodiment for obtaining recognition result
The highest sub- recognition result of the frequency of occurrences also achieves one kind in this way and fills as final recognition result in sub- recognition result out
Divide the identification process for considering each model assist process.
The method achieve by training set to it is multiple son simultaneously models be trained include to obtain Multiple Velocity Model knowledge
Other model, improves the recognition accuracy of identification model.
The embodiment of the present invention also provides a kind of Multiple recognition model construction device, which is used for
Execute any embodiment of aforementioned Multiple recognition model building method.Specifically, referring to Fig. 8, Fig. 8 is that the embodiment of the present invention mentions
The schematic block diagram of the Multiple recognition model construction device of confession.The Multiple recognition model construction device 100 can be configured at management
In server.
As shown in figure 8, Multiple recognition model construction device 100 includes pretreatment unit 110, original matrix acquiring unit
120, simplification matrix acquiring unit 130, raw data set acquiring unit 140, raw data set split cells 150 and model training
Unit 160.
Pretreatment unit 110 pre-processes source data for obtaining source data, data after being handled.
In the present embodiment, acquired source data can be the data under multiple application scenarios.Such as it settles a claim in vehicle insurance
Under scene, acquired source data can be the driving recording data of automobile data recorder, such as source data collected is to come from
In vehicle insurance APP, when user drives, spontaneous starting GPS positioning module is to GPS data into acquisition, and wherein source data includes longitude and latitude
The time series datas such as degree, speed, height above sea level and direction, period are that every 1s acquires a data.
Since GPS positioning module source data collected is related with satellite-signal, it is easy in the place of high building
There is the inaccurate situation of longitude and latitude data;And since APP needs outgoing data packet, in order to guarantee that data packet has been sent into
Function can inevitably encounter data packet and send out the case where leading to Data duplication again.In order to ensure understanding the abnormal data in source data, need
Source data is pre-processed.
In one embodiment, as shown in figure 9, pretreatment unit 110 includes:
Dealing of abnormal data unit 111, if for time series data included in source data there are current data relatively before
The increasing degree of one data exceeds preset amount of increase threshold value, and the value of current data delete or by the value of current data
It is set to last data * amount of increase threshold value, obtains data after dealing of abnormal data;
Extraction unit 112 at equal intervals, for pressing preset time cycle constant duration in number after dealing of abnormal data
Ground takes out data, obtains achievement data;
Normalization unit 113, for achievement data to be normalized, data after being handled.
In the present embodiment, in order to eliminate the abnormal data in source data, can judge according to the sequencing at time point
Included time series data exceeds preset amount of increase with respect to the increasing degree of last data with the presence or absence of current data in source data
Threshold value (such as amount of increase threshold value is set as 150%).If increasing degree of the time series data there are current data with respect to last data exceeds
Preset amount of increase threshold value then indicates that there are abnormal datas, and two ways can be used at this time and handled, and one is delete the timing
Data, another kind are that the value of the time series data is set to last data * amount of increase threshold value, obtain data after dealing of abnormal data.
Then, data are taken out to number medium time interval after dealing of abnormal data, obtains achievement data.Constant duration
Taking-up data, be the time-serial position in order to construct relative smooth.
Finally achievement data is normalized, is to be standardized data in order to subsequent mould
Type training.
Wherein, when logarithm type data are normalized, numerical value normalizing is carried out using with minimax laws for criterion
Change, the formula of minimax laws for criterion are as follows:Wherein XnomIndicate normalization after numerical value, X indicate to
The numerical value of normalized, XminIndicate the minimum value of the parameter, XmaxIndicate the maximum value of the parameter.To classification type data, adopt
With one-hot coding.
Original matrix acquiring unit 120, the data for each time point in data after handling as a row vector,
By in chronological sequence the corresponding row vector of sequence sequentially combines from top to bottom in data after processing, primitive character matrix is obtained.
In the present embodiment, the corresponding timing of time point t is 100,100,60,300,230 (preceding two such as in source data
A numerical value respectively represents longitude and dimension, third numerical value representation speed, and the 4th numerical value represents height above sea level, the 5th numerical value
Represent direction), corresponding row vector is [0.2 0.3 0.5 0.1 0.3];The corresponding row vector of time point 2t is [0.2
0.1 0.1 0.3 0.2];..., the corresponding row vector of time point Nt is [0.1 0.1 0.1 0.1 0.1].By this N number of row to
Amount sequentially combines from top to bottom according to chronological order, obtains the primitive character matrix of a N row 5 column.By by source data
It is converted to primitive character matrix according to timing, initial data can be normalized standard processing using the standard as training pattern
Input data.
Simplification matrix acquiring unit 130, column vector for will be chosen in primitive character matrix carry out statistical calculation with
Obtain newly-increased column vector, delete in newly-increased column vector with the Pearson correlation coefficient of each column vector in primitive character matrix
Size ranking is located at the column vector after preset rank threshold, is simplified rear eigenmatrix.
In this embodiment, the dimension of primitive character matrix is less, can will be in primitive character matrix in order to promote its dimension
The column vector chosen carries out statistical calculation to obtain newly-increased column vector realization and extend to the dimension of primitive character matrix.For example,
The primitive character matrix that obtained N row 5 arranges, after being extended using statistical calculation, after obtaining the increase feature arranged for N row M
Matrix.In order to retain increase feature after increase the higher column vector of data value in column vector in matrix newly, can delete and newly add
It is located at preset ranking threshold with the size ranking of the Pearson correlation coefficient of each column vector in primitive character matrix in vector
Column vector after value is simplified rear eigenmatrix.
Wherein, the Pearson correlation coefficient between any two column vector is calculated, can be calculate by the following formula:
Wherein, E indicates mathematic expectaion;
ρX,YValue range be (0,1), work as ρX,YIt indicates that the similarity degree of two column vectors is higher closer to 1, works as ρX,YMore
Indicate that the similarity degree of two column vectors is lower close to 0.
In one embodiment, as shown in Figure 10, simplification matrix acquiring unit 130 includes:
Operation species number acquiring unit 131, for obtaining preset statistical calculation species number;
Column vector selection unit 132 obtains and the statistics for randomly choosing two column vectors in primitive character matrix
The identical column vector combination of operation species number;
Newly-increased column vector acquiring unit 133, for by the combination of each column vector by one type operation in statistical calculation into
Row calculates, and obtains the newly-increased column vector that number is greater than or equal to the statistical calculation species number.
Wherein, the statistical calculation includes summation operation, asks difference operation, multiplying and division arithmetic.For example, when default
Statistical calculation species number be 4 kinds, be summation operation respectively, ask difference operation, multiplying and division arithmetic.At this time in original spy
2 column vectors of selection random in matrix are levied, a and b is denoted as respectively, then sums respectively to each numerical value in the two column vectors
Operation a+b, difference operation a-b, multiplying a*b and division arithmetic a/b are asked.
Raw data set acquiring unit 140, for increase in eigenmatrix after simplifying a column result vector obtain it is original
Data set.
In the present embodiment, it is trained, each row vector of eigenmatrix after simplification need to be made in order to treat training pattern
For the input to training pattern, while needing to increase in eigenmatrix after simplification a column result vector also with each row vector of correspondence
And using the corresponding end value of each row vector as the output to training pattern, the training for treating training pattern can be completed.
Raw data set split cells 150, for concentrating the row vector for extracting preset percentage by row to make in initial data
For training set, and concentrate non-extracted row vector as test set initial data.
In the present embodiment, preset percentage is taken to indicate that the line number of extracted row vector accounts for the total of entire raw data set
The percentage of line number, such as preset percentage is taken to may be configured as 70% indicates to choose initial data and concentrates 70% (i.e. preset the
One ratio is row vector 70%) as training set, will not be chosen as the row vector of the residue 30% of training set as test set.
By the way that raw data set is split as training set and test set, it can realize the training for treating training pattern, it can also be to instruction
The identification model got is tested.
Model training unit 160, for, to including being trained to training pattern for multiple submodels, being obtained by training set
To for the anti-identification model for cheating identification.
In the present embodiment, selected submodel have support vector machines (i.e. Support Vector Machine, SVM,
It is supervised learning model related to relevant learning algorithm, can analyze data, recognition mode, for classifying and returning point
Analysis), XGBoost model (eXtreme Gradient Boosting, it can be understood as extreme gradient is promoted, XGBoost model
Be exactly a kind of monitor model), Random Forest model etc..
By training set to including being trained to training pattern for multiple submodels, it includes polynary for can be obtained one
The Multiple recognition model of submodel can accurately judge data to be identified.And since identification model is multiple
Model can efficiently and dynamically adjust entire multi-model cooperation, have stronger fitness to different data set.
In one embodiment, as shown in figure 11, model training unit 160 includes:
Submodel packet count acquiring unit 161, for obtaining the total packet number to submodel in training pattern;
Training set division unit 162 obtains and the several numbers of total packet for training set to be carried out equal part according to total packet number
Identical sub- training set;
Identification model acquiring unit 163, for being grouped the training data of submodel using each sub- training set as correspondence
Model training is carried out, obtains identifying submodel correspondingly with submodel, to form multiple identification model.
In the present embodiment, first obtain the total packet number to submodel in training pattern, be in order to according to total packet number into
The equal part of the corresponding number of row.For example, the total packet number to submodel in training pattern is 3, then training set is divided into 3 equal parts,
Each sub- training set carries out model training respectively as the training data of corresponding grouping submodel, obtains corresponding with submodel
Identification submodel, to form multiple identification model.In addition to using the above-mentioned total packet number according to submodel to training set etc.
Point mode treat each submodel in training pattern and be trained, training set can also not divided directly but directly with
Training set is respectively trained each submodel.Using including multiple submodels to training pattern, it can increase or reduce son
Model has stronger scalability.
In one embodiment, as shown in figure 12, Multiple recognition model construction device 100 further include:
Model measurement authentication unit 171 carries out model verifying for test set to be inputted the identification model, if test set
In the difference of end value that imports in the result and test set that identification model is exported of included input vector be less than default error
Threshold value, identification model is by verifying and saving the identification model.
In the present embodiment, in order to ensure the recognition accuracy of identification model, identification mould is being obtained by training set training
After type, test set need to be inputted to the identification model and carry out model verifying.When the input vector in test set is input to identification
The difference of model, obtained output result and the corresponding end value in test set is (included i.e. in test set within the allowable range
The difference of end value that imports in the identification model result and test set that are exported of input vector be less than default error threshold), then
Indicate the identification model can by verifying and it is specifically used.Once the identification model is not verified by model, then it represents that selected
The submodel selected is unreasonable, needs to input more training datas again to be trained to the submodel.It is verified by increasing
Link can effectively ensure the availability and recognition accuracy of identification model.
In one embodiment, as shown in figure 13, Multiple recognition model construction device 100 further include:
First input unit 1721, it is a pair of according to each submodel one for data to be identified to be input to identification model
It should obtain sub- recognition result;
Weight calculation unit 1722, for each sub- recognition result multiplied by the default weighted value of corresponding submodel and to be summed,
Obtain recognition result.
In the present embodiment, as the first specific embodiment for obtaining recognition result, in order to make multi-model assist to be known
It not as a result, can be for the corresponding default weighted value of each submodel, by each sub- recognition result multiplied by the default weight of corresponding submodel
It is worth and sums, obtain recognition result, can efficiently and dynamically adjusts entire multi-model cooperation.
In one embodiment, as shown in figure 14, Multiple recognition model construction device 100 further include:
Second input unit 1731, it is a pair of according to each submodel one for data to be identified to be input to identification model
It should obtain sub- recognition result;
Frequency statistics unit 1732, for obtaining the son identification knot that numerical value output frequency in sub- recognition result is maximum value
Fruit, and the recognition result as identification model.
It in the present embodiment, can also be defeated by counting a subpattern as the second specific embodiment for obtaining recognition result
The highest sub- recognition result of the frequency of occurrences also achieves one kind in this way and fills as final recognition result in sub- recognition result out
Divide the identification process for considering each model assist process.
The arrangement achieves by training set to it is multiple son simultaneously models be trained include to obtain Multiple Velocity Model knowledge
Other model, improves the recognition accuracy of identification model.
Above-mentioned Multiple recognition model construction device can be implemented as the form of computer program, which can be
It is run in computer equipment as shown in figure 15.
Figure 15 is please referred to, Figure 15 is the schematic block diagram of computer equipment provided in an embodiment of the present invention.
Refering to fig. 15, which includes processor 502, memory and the net connected by system bus 501
Network interface 505, wherein memory may include non-volatile memory medium 503 and built-in storage 504.
The non-volatile memory medium 503 can storage program area 5031 and computer program 5032.The computer program
5032 are performed, and processor 502 may make to execute Multiple recognition model building method.
The processor 502 supports the operation of entire computer equipment 500 for providing calculating and control ability.
The built-in storage 504 provides environment for the operation of the computer program 5032 in non-volatile memory medium 503, should
When computer program 5032 is executed by processor 502, processor 502 may make to execute Multiple recognition model building method.
The network interface 505 is for carrying out network communication, such as the transmission of offer data information.Those skilled in the art can
To understand, structure shown in Figure 15, only the block diagram of part-structure relevant to the present invention program, is not constituted to this hair
The restriction for the computer equipment 500 that bright scheme is applied thereon, specific computer equipment 500 may include than as shown in the figure
More or fewer components perhaps combine certain components or with different component layouts.
Wherein, the processor 502 is for running computer program 5032 stored in memory, to realize following function
Can: source data is obtained, source data is pre-processed, data after being handled;By the number at time point each in data after processing
According to as a row vector, by chronological sequence the corresponding row vector of sequence sequentially combines from top to bottom in data after processing, obtain
To primitive character matrix;The column vector chosen in primitive character matrix is subjected to statistical calculation to obtain newly-increased column vector, is deleted
It is default except being located in newly-increased column vector with the size ranking of the Pearson correlation coefficient of each column vector in primitive character matrix
Rank threshold after column vector, be simplified rear eigenmatrix;It will increase by a column result vector in eigenmatrix after simplification
Obtain raw data set;It is concentrated in initial data and extracts the row vector of preset percentage as training set by row, and by original number
According to the non-extracted row vector of concentration as test set;And by training set to include multiple submodels to training pattern into
Row training, obtains for the anti-identification model for cheating identification.
In one embodiment, processor 502 execute it is described source data is pre-processed, the step of data after being handled
It when rapid, performs the following operations: if the growth width of time series data included in source data there are current data with respect to last data
Degree exceeds preset amount of increase threshold value, and the value of current data delete or the value of current data is set to last data *
Amount of increase threshold value obtains data after dealing of abnormal data;By between the times such as preset time cycle in number after dealing of abnormal data
Data are taken out every ground, obtain achievement data;Achievement data is normalized, data after being handled.
In one embodiment, processor 502 described unites the column vector chosen in primitive character matrix executing
When meter operation is to obtain the step of newly-increased column vector, performs the following operations: obtaining preset statistical calculation species number;Random selection
Two column vectors in primitive character matrix obtain column vector identical with the statistical calculation species number and combine;By it is each arrange to
Amount combination is calculated by one type operation in statistical calculation, obtains number more than or equal to the statistical calculation species number
Newly-increased column vector;Wherein, the statistical calculation includes summation operation, asks difference operation, multiplying and division arithmetic.
In one embodiment, processor 502 is executing the training set that passes through to the mould to be trained including multiple submodels
Type is trained, and when obtaining the step for the anti-identification model for cheating identification, is performed the following operations: being obtained in training pattern
The total packet number of submodel;Training set is subjected to equal part according to total packet number, obtains son training identical with the several numbers of total packet
Collection;Model training is carried out using each sub- training set as the training data of corresponding grouping submodel, is obtained with submodel one by one
Corresponding identification submodel, to form multiple identification model.
In one embodiment, processor 502 is executing the training set that passes through to the mould to be trained including multiple submodels
After the step of type is trained, and is obtained for the anti-identification model for cheating identification, also performs the following operations: test set is inputted
The identification model carries out model verifying, if input vector included in test set import result that identification model is exported with
The difference of end value in test set is less than default error threshold, and identification model is by verifying and saving the identification model.
In one embodiment, processor 502 is executing the training set that passes through to the mould to be trained including multiple submodels
It after the step of type is trained, and is obtained for the anti-identification model for cheating identification, also performs the following operations: by data to be identified
It is input to identification model, is corresponded to obtain sub- recognition result according to each submodel;By each sub- recognition result multiplied by corresponding son
The default weighted value of model is simultaneously summed, and recognition result is obtained.
In one embodiment, processor 502 is executing the training set that passes through to the mould to be trained including multiple submodels
It after the step of type is trained, and is obtained for the anti-identification model for cheating identification, also performs the following operations: by data to be identified
It is input to identification model, is corresponded to obtain sub- recognition result according to each submodel;Numerical value in sub- recognition result is obtained to export
Frequency is the sub- recognition result of maximum value, and the recognition result as identification model.
It will be understood by those skilled in the art that the embodiment of computer equipment shown in Figure 15 is not constituted to computer
The restriction of equipment specific composition, in other embodiments, computer equipment may include components more more or fewer than diagram, or
Person combines certain components or different component layouts.For example, in some embodiments, computer equipment can only include depositing
Reservoir and processor, in such embodiments, the structure and function of memory and processor are consistent with embodiment illustrated in fig. 15,
Details are not described herein.
It should be appreciated that in embodiments of the present invention, processor 502 can be central processing unit (Central
Processing Unit, CPU), which can also be other general processors, digital signal processor (Digital
Signal Processor, DSP), specific integrated circuit (Application Specific Integrated Circuit,
ASIC), ready-made programmable gate array (Field-Programmable GateArray, FPGA) or other programmable logic devices
Part, discrete gate or transistor logic, discrete hardware components etc..Wherein, general processor can be microprocessor or
The processor is also possible to any conventional processor etc..
A kind of computer readable storage medium is provided in another embodiment of the invention.The computer readable storage medium
It can be non-volatile computer readable storage medium.The computer-readable recording medium storage has computer program, wherein
Acquisition source data is performed the steps of when computer program is executed by processor, source data is pre-processed, after obtaining processing
Data;It, will be in chronological sequence suitable in data after processing using the data at time point each in data after processing as a row vector
The corresponding row vector of sequence sequentially combines from top to bottom, obtains primitive character matrix;By the column chosen in primitive character matrix to
Amount carries out statistical calculation to obtain newly-increased column vector, delete in newly-increased column vector with each column vector in primitive character matrix
The size ranking of Pearson correlation coefficient is located at the column vector after preset rank threshold, is simplified rear eigenmatrix;It will
Increase by a column result vector after simplification in eigenmatrix and obtains raw data set;It is concentrated in initial data and extracts default percentage by row
The row vector of ratio concentrates non-extracted row vector as test set as training set, and using initial data;And pass through training
Collect to including being trained to training pattern for multiple submodels, obtains for the anti-identification model for cheating identification.
In one embodiment, described that source data is pre-processed, data after being handled, comprising: if institute in source data
Including time series data there are current data with respect to last data increasing degree exceed preset amount of increase threshold value, by current data
Value carry out delete or the value of current data is set to last data * amount of increase threshold value, obtain data after dealing of abnormal data;
Data are taken out by preset time cycle constant duration in number after dealing of abnormal data, obtain achievement data;By index
Data are normalized, data after being handled.
In one embodiment, the statistical calculation includes summation operation, asks difference operation, multiplying and division arithmetic;Institute
It states and the column vector chosen in primitive character matrix is subjected to statistical calculation to obtain newly-increased column vector, comprising: obtain preset
Statistical calculation species number;Two column vectors in primitive character matrix are randomly choosed, are obtained identical as the statistical calculation species number
Column vector combination;Combine each column vector and calculate by one type operation in statistical calculation, obtain number be greater than or
Equal to the newly-increased column vector of the statistical calculation species number.
In one embodiment, it is described by training set to including that multiple submodels being trained to training pattern, obtain
For the anti-identification model for cheating identification, comprising: obtain the total packet number to submodel in training pattern;By training set according to total
Packet count carries out equal part, obtains sub- training set identical with the several numbers of total packet;It is grouped using each sub- training set as corresponding
The training data of submodel carries out model training, obtains identifying submodel correspondingly with submodel, to form multiple knowledge
Other model.
In one embodiment, it is described by training set to including that multiple submodels being trained to training pattern, obtain
After the anti-identification model for cheating identification, further includes: test set is inputted the identification model and carries out model verifying, if surveying
The difference for the end value that examination concentrates included input vector to import in the result and test set that identification model is exported is less than default
Error threshold, identification model is by verifying and saving the identification model.
In one embodiment, it is described by training set to including that multiple submodels being trained to training pattern, obtain
After the anti-identification model for cheating identification, further includes: data to be identified are input to identification model, according to each submodel
One-to-one correspondence obtains sub- recognition result;Each sub- recognition result multiplied by the default weighted value of corresponding submodel and is summed, is known
Other result.
In one embodiment, it is described by training set to including that multiple submodels being trained to training pattern, obtain
After the anti-identification model for cheating identification, further includes: data to be identified are input to identification model, according to each submodel
One-to-one correspondence obtains sub- recognition result;It obtains numerical value output frequency in sub- recognition result and is the sub- recognition result of maximum value, and make
For the recognition result of identification model.
It is apparent to those skilled in the art that for convenience of description and succinctly, foregoing description is set
The specific work process of standby, device and unit, can refer to corresponding processes in the foregoing method embodiment, and details are not described herein.
Those of ordinary skill in the art may be aware that unit described in conjunction with the examples disclosed in the embodiments of the present disclosure and algorithm
Step can be realized with electronic hardware, computer software, or a combination of the two, in order to clearly demonstrate hardware and software
Interchangeability generally describes each exemplary composition and step according to function in the above description.These functions are studied carefully
Unexpectedly the specific application and design constraint depending on technical solution are implemented in hardware or software.Professional technician
Each specific application can be used different methods to achieve the described function, but this realization is it is not considered that exceed
The scope of the present invention.
In several embodiments provided by the present invention, it should be understood that disclosed unit and method, it can be with
It realizes by another way.For example, the apparatus embodiments described above are merely exemplary, for example, the unit
It divides, only logical function partition, there may be another division manner in actual implementation, can also will be with the same function
Unit set is at a unit, such as multiple units or components can be combined or can be integrated into another system or some
Feature can be ignored, or not execute.In addition, shown or discussed mutual coupling, direct-coupling or communication connection can
Be through some interfaces, the indirect coupling or communication connection of device or unit, be also possible to electricity, mechanical or other shapes
Formula connection.
The unit as illustrated by the separation member may or may not be physically separated, aobvious as unit
The component shown may or may not be physical unit, it can and it is in one place, or may be distributed over multiple
In network unit.Some or all of unit therein can be selected to realize the embodiment of the present invention according to the actual needs
Purpose.
It, can also be in addition, the functional units in various embodiments of the present invention may be integrated into one processing unit
It is that each unit physically exists alone, is also possible to two or more units and is integrated in one unit.It is above-mentioned integrated
Unit both can take the form of hardware realization, can also realize in the form of software functional units.
If the integrated unit is realized in the form of SFU software functional unit and sells or use as independent product
When, it can store in one storage medium.Based on this understanding, technical solution of the present invention is substantially in other words to existing
The all or part of part or the technical solution that technology contributes can be embodied in the form of software products, should
Computer software product is stored in a storage medium, including some instructions are used so that a computer equipment (can be
Personal computer, server or network equipment etc.) execute all or part of step of each embodiment the method for the present invention
Suddenly.And storage medium above-mentioned include: USB flash disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), magnetic disk or
The various media that can store program code such as person's CD.
The above description is merely a specific embodiment, but scope of protection of the present invention is not limited thereto, any
Those familiar with the art in the technical scope disclosed by the present invention, can readily occur in various equivalent modifications or replace
It changes, these modifications or substitutions should be covered by the protection scope of the present invention.Therefore, protection scope of the present invention should be with right
It is required that protection scope subject to.
Claims (10)
1. a kind of Multiple recognition model building method characterized by comprising
Source data is obtained, source data is pre-processed, data after being handled;
Using the data at time point each in data after processing as a row vector, by chronological sequence sequence in data after processing
Corresponding row vector sequentially combines from top to bottom, obtains primitive character matrix;
The column vector chosen in primitive character matrix is subjected to statistical calculation to obtain newly-increased column vector, deletes newly-increased column vector
In with the size ranking of the Pearson correlation coefficient of each column vector in primitive character matrix be located at preset rank threshold it
Column vector afterwards is simplified rear eigenmatrix;
It will increase by a column result vector in eigenmatrix after simplification and obtain raw data set;
The row vector for extracting preset percentage by row is concentrated not mentioned as training set, and by initial data concentration in initial data
The row vector taken is as test set;And
By training set to including being trained to training pattern for multiple submodels, obtain for the anti-identification mould for cheating identification
Type.
2. Multiple recognition model building method according to claim 1, which is characterized in that described to be located in advance to source data
Reason, data after being handled, comprising:
If there are the increasing degrees of the opposite last data of current data to rise beyond preset for included time series data in source data
The value of current data delete or the value of current data is set to last data * amount of increase threshold value, obtains different by width threshold value
Data after regular data processing;
Data are taken out by preset time cycle constant duration in number after dealing of abnormal data, obtain achievement data;
Achievement data is normalized, data after being handled.
3. Multiple recognition model building method according to claim 1, which is characterized in that the statistical calculation includes summation
Difference operation, multiplying and division arithmetic are asked in operation;
It is described that the column vector chosen in primitive character matrix is subjected to statistical calculation to obtain newly-increased column vector, comprising:
Obtain preset statistical calculation species number;
Two column vectors in primitive character matrix are randomly choosed, Column vector groups identical with the statistical calculation species number are obtained
It closes;
It combines each column vector and is calculated by one type operation in statistical calculation, obtain number more than or equal to the system
Count the newly-increased column vector of operation species number.
4. Multiple recognition model building method according to claim 1, which is characterized in that it is described by training set to including
Multiple submodels are trained to training pattern, are obtained for the anti-identification model for cheating identification, comprising:
Obtain the total packet number to submodel in training pattern;
Training set is subjected to equal part according to total packet number, obtains sub- training set identical with the several numbers of total packet;
Model training is carried out using each sub- training set as the training data of corresponding grouping submodel, is obtained with submodel one by one
Corresponding identification submodel, to form multiple identification model.
5. Multiple recognition model building method according to claim 1, which is characterized in that it is described by training set to including
Multiple submodels are trained to training pattern, are obtained for after the anti-identification model for cheating identification, further includes:
Test set is inputted into the identification model and carries out model verifying, if input vector included in test set imports identification mould
The difference for the end value in result and test set that type is exported is less than default error threshold, and identification model is by verifying and saving institute
State identification model.
6. Multiple recognition model building method according to claim 1, which is characterized in that it is described by training set to including
Multiple submodels are trained to training pattern, are obtained for after the anti-identification model for cheating identification, further includes:
Data to be identified are input to identification model, are corresponded to obtain sub- recognition result according to each submodel;
Each sub- recognition result multiplied by the default weighted value of corresponding submodel and is summed, recognition result is obtained.
7. Multiple recognition model building method according to claim 1, which is characterized in that it is described by training set to including
Multiple submodels are trained to training pattern, are obtained for after the anti-identification model for cheating identification, further includes:
Data to be identified are input to identification model, are corresponded to obtain sub- recognition result according to each submodel;
Obtain the sub- recognition result that numerical value output frequency in sub- recognition result is maximum value, and the identification knot as identification model
Fruit.
8. a kind of Multiple recognition model construction device characterized by comprising
Pretreatment unit pre-processes source data for obtaining source data, data after being handled;
Original matrix acquiring unit, the data for each time point in data after handling will be handled as a row vector
In chronological sequence the corresponding row vector of sequence sequentially combines from top to bottom in data afterwards, obtains primitive character matrix;
Simplification matrix acquiring unit, for the column vector chosen in primitive character matrix to be carried out statistical calculation to be increased newly
Column vector deletes the size ranking in newly-increased column vector with the Pearson correlation coefficient of each column vector in primitive character matrix
Column vector after preset rank threshold is simplified rear eigenmatrix;
Raw data set acquiring unit obtains raw data set for increasing by a column result vector in eigenmatrix after simplifying;
Raw data set split cells extracts the row vector of preset percentage as training by row for concentrating in initial data
Collection, and concentrate non-extracted row vector as test set initial data;
Model training unit, for, to including being trained to training pattern for multiple submodels, being used for by training set
The identification model of anti-fraud identification.
9. a kind of computer equipment, including memory, processor and it is stored on the memory and can be on the processor
The computer program of operation, which is characterized in that the processor realizes such as claim 1 to 7 when executing the computer program
Any one of described in Multiple recognition model building method.
10. a kind of computer readable storage medium, which is characterized in that the computer-readable recording medium storage has computer journey
Sequence, it is as described in any one of claim 1 to 7 more that the computer program when being executed by a processor executes the processor
Weight identification model construction method.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201811601941.5A CN109784377A (en) | 2018-12-26 | 2018-12-26 | Multiple recognition model building method, device, computer equipment and storage medium |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201811601941.5A CN109784377A (en) | 2018-12-26 | 2018-12-26 | Multiple recognition model building method, device, computer equipment and storage medium |
Publications (1)
Publication Number | Publication Date |
---|---|
CN109784377A true CN109784377A (en) | 2019-05-21 |
Family
ID=66498521
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201811601941.5A Pending CN109784377A (en) | 2018-12-26 | 2018-12-26 | Multiple recognition model building method, device, computer equipment and storage medium |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN109784377A (en) |
Cited By (7)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN111582305A (en) * | 2020-03-26 | 2020-08-25 | 平安科技(深圳)有限公司 | Biological feature recognition method and device, computer equipment and storage medium |
CN111652279A (en) * | 2020-04-30 | 2020-09-11 | 中国平安财产保险股份有限公司 | Behavior evaluation method and device based on time sequence data and readable storage medium |
CN112101236A (en) * | 2020-09-17 | 2020-12-18 | 济南大学 | Intelligent error correction method and system for elderly accompanying robot |
CN112149702A (en) * | 2019-06-28 | 2020-12-29 | 北京百度网讯科技有限公司 | Feature processing method and device |
CN112308459A (en) * | 2020-11-23 | 2021-02-02 | 国网北京市电力公司 | Power grid household transformation relation identification method and identification device, and electronic equipment |
CN112732846A (en) * | 2021-01-27 | 2021-04-30 | 深圳市科荣软件股份有限公司 | Water affair operation analysis system, method, electronic equipment and storage medium |
CN113886130A (en) * | 2021-10-21 | 2022-01-04 | 深信服科技股份有限公司 | Method, device and medium for processing database fault |
-
2018
- 2018-12-26 CN CN201811601941.5A patent/CN109784377A/en active Pending
Cited By (9)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN112149702A (en) * | 2019-06-28 | 2020-12-29 | 北京百度网讯科技有限公司 | Feature processing method and device |
CN111582305A (en) * | 2020-03-26 | 2020-08-25 | 平安科技(深圳)有限公司 | Biological feature recognition method and device, computer equipment and storage medium |
CN111582305B (en) * | 2020-03-26 | 2023-08-18 | 平安科技(深圳)有限公司 | Biological feature recognition method, apparatus, computer device and storage medium |
CN111652279A (en) * | 2020-04-30 | 2020-09-11 | 中国平安财产保险股份有限公司 | Behavior evaluation method and device based on time sequence data and readable storage medium |
CN111652279B (en) * | 2020-04-30 | 2024-04-30 | 中国平安财产保险股份有限公司 | Behavior evaluation method and device based on time sequence data and readable storage medium |
CN112101236A (en) * | 2020-09-17 | 2020-12-18 | 济南大学 | Intelligent error correction method and system for elderly accompanying robot |
CN112308459A (en) * | 2020-11-23 | 2021-02-02 | 国网北京市电力公司 | Power grid household transformation relation identification method and identification device, and electronic equipment |
CN112732846A (en) * | 2021-01-27 | 2021-04-30 | 深圳市科荣软件股份有限公司 | Water affair operation analysis system, method, electronic equipment and storage medium |
CN113886130A (en) * | 2021-10-21 | 2022-01-04 | 深信服科技股份有限公司 | Method, device and medium for processing database fault |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN109784377A (en) | Multiple recognition model building method, device, computer equipment and storage medium | |
CN113627518B (en) | Method for realizing neural network brain electricity emotion recognition model by utilizing transfer learning | |
CN110031226A (en) | A kind of diagnostic method and device of bearing fault | |
CN105894039A (en) | Emotion recognition modeling method, emotion recognition method and apparatus, and intelligent device | |
CN110471913A (en) | A kind of data cleaning method and device | |
CN108446534A (en) | Select the method, apparatus and computer readable storage medium of neural network hyper parameter | |
CN106846538A (en) | Cross car record treating method and apparatus | |
US10068073B2 (en) | Verification device and control method for verifiction device, as well as computer program | |
CN109816043B (en) | Method and device for determining user identification model, electronic equipment and storage medium | |
CN113133769A (en) | Equipment control method, device and terminal based on motor imagery electroencephalogram signals | |
CN109325792A (en) | The branch mailbox method and box separation device of credit evaluation variable, equipment and storage medium | |
US20190180194A1 (en) | Computer-readable recording medium, abnormality candidate extraction method, and abnormality candidate extraction apparatus | |
CN110472268A (en) | A kind of bridge monitoring data modality recognition methods and device | |
CN112241494A (en) | Key information pushing method and device based on user behavior data | |
CN109871762A (en) | A kind of evaluation method and device of human face recognition model | |
CN108154163A (en) | Data processing method, data identification and learning method and its device | |
CN105787136A (en) | Data mining method and data mining device | |
CN113160987B (en) | Health state prediction method, apparatus, computer device and storage medium | |
CN113918433A (en) | Adaptive intelligent network equipment performance index abnormity detection device and method | |
CN111984842B (en) | Bank customer data processing method and device | |
CN116192665B (en) | Data processing method, device, computer equipment and storage medium | |
CN115758153A (en) | Target object track data processing method, device and equipment | |
CN116302870A (en) | Mechanical hard disk health assessment method, system and storage medium based on evolutionary diagram | |
EP3255593A1 (en) | Performance analysing and forecasting system | |
CN114098764A (en) | Data processing method and device, electronic equipment and storage medium |
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 |