CN109242165A - A kind of model training and prediction technique and device based on model training - Google Patents
A kind of model training and prediction technique and device based on model training Download PDFInfo
- Publication number
- CN109242165A CN109242165A CN201810973101.5A CN201810973101A CN109242165A CN 109242165 A CN109242165 A CN 109242165A CN 201810973101 A CN201810973101 A CN 201810973101A CN 109242165 A CN109242165 A CN 109242165A
- Authority
- CN
- China
- Prior art keywords
- training
- model
- sample data
- straton
- data
- 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
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q10/00—Administration; Management
- G06Q10/04—Forecasting or optimisation specially adapted for administrative or management purposes, e.g. linear programming or "cutting stock problem"
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- 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
- G06Q40/00—Finance; Insurance; Tax strategies; Processing of corporate or income taxes
- G06Q40/03—Credit; Loans; Processing thereof
Abstract
The invention discloses a kind of model training and prediction technique and device based on model training, which comprises be successively directed to every straton model to training pattern comprising at least two straton models, identify whether the straton model is the last layer submodel；If not, will include that each sample data of positive or negative sample label is input in the straton model in training set, which be trained；And each sample data corresponds to the confidence level of positive sample data in the straton model output training set completed based on training, it is greater than the sample data of the corresponding confidence threshold value of the straton model using the confidence level for corresponding to positive sample data in training set, the sample data in training set is updated；If so, will include that each sample data of positive or negative sample label is input in the straton model in training set, which be trained, to improve the precision of model prediction.
Description
Technical field
The present invention relates to big data science and technology field more particularly to a kind of model training and based on the prediction of model training
Method and device.
Background technique
With the economic high speed development with informationization, big data is come into being, and big data refers to that a kind of scale reaches
Obtain, storage, management, analysis etc. well beyond traditional database software means capability range data acquisition system, by right
Big data is analyzed and predicted, and can provide strong support for the business of enterprise and decision etc..In face of big data era
The data of magnanimity, tradition by manually carry out data analysis and data value excavation be not suitable for, and artificial intelligence technology go out
Now a kind of preferable solution provided is excavated for the data analysis of big data and data value.
Suitable training sample data are chosen in the application of existing artificial intelligence model, usually user, wherein training sample
It include a large amount of positive sample data and negative sample data in notebook data, by the corresponding artificial intelligence learning algorithm of model to model
It is trained, trained model can predict the data of input, predict that the data of input are positive sample data, still
Negative sample data.By taking credit card transaction data as an example, the credit card transaction data that transaction swindling occurs in training set is positive sample
Data, the credit card transaction data that transaction swindling does not occur are negative sample data, by positive sample data a large amount of in training set and
After the completion of negative sample data are trained model, after inputting credit card transaction data, the model that training is completed can be to input
Credit card transaction data predicted with the presence or absence of transaction swindling.
However existing model only has one layer, in face of complex or more comprising feature data, one layer of model is not
The feature for including in data can adequately be learnt, the accuracy of prediction result is not high.Still it is with credit card transaction data
Example, the probability that transaction swindling occurs for credit card trade is about 5/10000ths (5BP), if there are 100,000 credits in certain bank day
Card transaction, then probably with the presence of 50 credit card trade transaction swindlings, because being related to financial transaction field, need to ensure every credit card
It trades accurate, it, usually will be by model from 100,000 credit card transaction datas in order to prevent to the omission of fraudulent trading
2500 credit card transaction datas are filtered out, carry out manual review, capable of just finding out 50, there are the credit card trades of transaction swindling
Data need to expend huge manpower, therefore are badly in need of a kind of model training scheme, to improve the precision of model prediction.
Summary of the invention
The present invention provides a kind of model training and prediction technique and device based on model training, to solve the prior art
It is middle that there are the not high problems of the precision of model prediction.
In a first aspect, the invention discloses a kind of model training methods, which comprises
Successively for every straton model to training pattern, identify the straton model whether be it is described to training pattern most
Latter straton model, wherein described include at least two straton models to training pattern；
If not, will include that each sample data of positive sample or negative sample label is input to the straton mould in training set
In type, which is trained；And each sample data in the straton model output training set completed based on training
The confidence level of corresponding positive sample data is greater than the straton model pair using the confidence level for corresponding to positive sample data in the training set
The sample data for the confidence threshold value answered is updated the sample data in the training set；
If so, will include that each sample data of positive sample or negative sample label is input to the straton mould in training set
In type, which is trained.
Further, described will include that each sample data of positive sample or negative sample label is input to this in training set
In straton model, before being trained to the straton model, the method also includes:
Judge whether the quantity of sample data in the training set is greater than the amount threshold of setting；
If so, carrying out subsequent step；
If not, issuing warning information.
Second aspect, the invention discloses a kind of prediction techniques based on above-mentioned model training method, which comprises
Data to be tested are input in the model of training completion；
Based on the model that the training is completed, output predicts whether the data to be tested are positive the result of sample data.
Further, the sample data if it is predicted that data to be tested are positive, the method also includes:
Issue warning information.
The third aspect, the invention discloses a kind of model training apparatus, described device includes:
Identification module, for successively for every straton model to training pattern, identifying whether the straton model is described
To the last layer submodel of training pattern, wherein described include at least two straton models to training pattern；
Training module will be in training set if being the non-the last layer submodel to training pattern for submodel
Include that each sample data of positive sample or negative sample label is input in the straton model, which has been instructed
Practice；And each sample data corresponds to the confidence level of positive sample data in the straton model output training set completed based on training,
It is greater than the sample number of the corresponding confidence threshold value of the straton model using the confidence level for corresponding to positive sample data in the training set
According to being updated to the sample data in the training set；
Training module will be in training set if being also used to submodel is the last layer submodel to training pattern
Include that each sample data of positive sample or negative sample label is input in the straton model, which has been instructed
Practice.
Further, described device further include:
Alarm module is judged, for judging whether the quantity of sample data in the training set is greater than the quantity threshold of setting
Value, and when the judgment result is yes, training module is triggered, when the judgment result is No, issues warning information.
Fourth aspect, the invention discloses a kind of prediction meanss based on above-mentioned model training apparatus, described device includes:
Input module, for data to be tested to be input in the model of training completion；
Output module, the model for being completed based on the training, output predict whether the data to be tested are positive sample
The result of notebook data.
Further, described device further include:
Alarm module issues warning information for the sample data if it is predicted that data to be tested are positive.
Due in embodiments of the present invention, to include at least two straton models in training pattern, and successively treating training
When every straton model of model is trained, for every straton model to the last layer non-in training pattern, passing through instruction
It is every in the straton model output training set based on training completion after the completion of practicing the sample data concentrated to straton model training
A sample data corresponds to the confidence level of positive sample data, and is greater than the layer using the confidence level for corresponding to positive sample data in training set
The sample data of the corresponding confidence threshold value of submodel, is updated the sample data in training set, guarantees successively be directed to
When every straton model of training pattern is trained, the interference of layer-by-layer exclusive segment negative sample data, so as to training pattern
In, for training the sample data of every straton model to have differences, every straton model can extract the different characteristic of sample data,
It ensure that the abundant study to training pattern to sample data feature, to improve the precision of prediction of model.
Detailed description of the invention
In order to more clearly explain the embodiment of the invention or the technical proposal in the existing technology, to embodiment or will show below
There is attached drawing needed in technical description to be briefly described, it should be apparent that, the accompanying drawings in the following description is only this
Some embodiments of invention for those of ordinary skill in the art without creative efforts, can be with
It obtains other drawings based on these drawings.
Fig. 1 is a kind of one of model training process schematic provided in an embodiment of the present invention；
Fig. 2 is the two of a kind of model training process schematic provided in an embodiment of the present invention；
Fig. 3 is a kind of prediction process schematic provided in an embodiment of the present invention；
Fig. 4 is a kind of model training apparatus structural schematic diagram provided in an embodiment of the present invention；
Fig. 5 is a kind of prediction meanss structural schematic diagram provided in an embodiment of the present invention.
Specific embodiment
To make the objectives, technical solutions, and advantages of the present invention clearer, make below in conjunction with the attached drawing present invention into one
Step ground detailed description, it is clear that described embodiment is only a part of the embodiments of the present invention, rather than whole implementation
Example.Based on the embodiments of the present invention, obtained by those of ordinary skill in the art without making creative efforts
Every other embodiment, shall fall within the protection scope of the present invention.
Embodiment 1:
Fig. 1 is a kind of model training process schematic provided in an embodiment of the present invention, and the process includes:
S101: successively for every straton model to training pattern, identify whether the straton model is the mould to be trained
The last layer submodel of type, if not, S102 is carried out, if so, carrying out S103.
Data quality checking method provided in an embodiment of the present invention is applied to electronic equipment, and the electronic equipment can be hand
The equipment such as machine, PC (PC), tablet computer, are also possible to the equipment such as server, server cluster.
It in embodiments of the present invention, include at least two straton models to training pattern, wherein the corresponding calculation of every straton model
Method can be the same or different, such as: it include three straton models to training pattern, every straton model can correspond to convolution mind
Through network algorithm；It is also possible to the first straton model and corresponds to convolutional neural networks algorithm, the second straton model and third layer submodule
Type counterlogic regression algorithm, without specifically limiting.
In addition, electronic equipment when treating training pattern and being trained, is every straton mould of the successively model for training
What type was trained, such as: include three straton models to training pattern, then first the first straton model is trained, first
After the completion of straton model training, the second straton model is trained, after the completion of the training of the second straton model, to third straton
Model is trained.
S102: will include that each sample data of positive sample or negative sample label is input to the straton model in training set
In, which is trained；And each sample data pair in the straton model output training set completed based on training
The confidence level for answering positive sample data, it is corresponding greater than the straton model using the confidence level for corresponding to positive sample data in the training set
Confidence threshold value sample data, the sample data in the training set is updated.
S103: will include that each sample data of positive sample or negative sample label is input to the straton model in training set
In, which is trained.
Specifically, comprising being largely known to be the sample data of positive sample or negative sample, and training set in training set
In include each sample data include positive sample label or negative sample label.In addition, in embodiments of the present invention, electronics is set
The corresponding confidence threshold value of every straton model is also preserved in standby, preferably, the corresponding number of plies of submodel is bigger, corresponding confidence
It is bigger to spend threshold value.Such as: it include three straton models to training pattern, the corresponding confidence threshold value of the first straton model is 0.5, the
The corresponding confidence threshold value of two straton models is 0.7, the corresponding confidence threshold value of third straton model is 0.9.
It will include positive sample or negative sample in training set successively for every straton model to training pattern in training
Each sample data of this label is input in the straton model, is trained to the straton model, the submodel that training is completed
It can determine that the detection data corresponds to the confidence level of positive sample data, and can be according to the detection according to the detection data of input
The confidence level of the corresponding positive sample data of data, if be greater than the corresponding confidence threshold value of the straton model, export the straton mould
Whether type is positive the prediction result of sample data to the detection data.
In addition, if the straton model is the non-the last layer submodel to training pattern, to straton model training
After the completion, before being trained to next straton model of the straton model, the straton model completed based on training exports training
Each sample data is concentrated to correspond to the confidence level of positive sample data, using the confidence level for corresponding to positive sample data in the training set
The sample data of confidence threshold value corresponding greater than the straton model, is updated the sample data in the training set, from
And the negative sample data of training concentrated part are removed, the interference of exclusive segment negative sample data, so that being used for in training pattern
The sample data of the every straton model of training has differences, and guarantees the abundant study to training pattern to positive sample data characteristics, from
And improve the precision of prediction for the model that training is completed.
Fig. 2 is a kind of model training process schematic provided in an embodiment of the present invention, as shown in Fig. 2, treating trained mould
When type is trained, each sample data comprising positive sample or negative sample label in training set 1 is input to first layer first
In submodel (F1 (x)), to F1 (x) training, each sample data is corresponding in F1 (x) the output training set 1 completed based on training
The confidence level of positive sample data filters out the confidence level that the corresponding positive sample data of condition 1 are unsatisfactory in training set 1 no more than F1 (x)
The sample data of corresponding confidence threshold value, i.e., it is corresponding greater than F1 (x) using the confidence level for corresponding to positive sample data in training set 1
Confidence threshold value sample data, the sample data in training set 1 is updated, training set 2 is obtained；It will be in training set 2
Each sample data comprising positive sample or negative sample label, is input in the second straton model (F2 (x)), trains to F2 (x),
Each sample data corresponds to the confidence level of positive sample data in F2 (x) the output training set 2 completed based on training, filters out training set
The confidence level that the corresponding positive sample data of condition 2 are unsatisfactory in 2 is not more than the sample data of the corresponding confidence threshold value of F2 (x), i.e.,
It is greater than the sample data of the corresponding confidence threshold value of F1 (x) using the confidence level for corresponding to positive sample data in training set 2, to training
Sample data in collection 2 is updated, and obtains training set 3, will include each sample of positive sample or negative sample label in training set 3
Notebook data is input in third straton model (F3 (x)) ... until treating n-th layer submodel (Fn (the x)) training of training pattern
It completes.Wherein, the training set 1 is to input for treating the initial training collection that training pattern is trained, the training set
In include whole sample datas.
Due in embodiments of the present invention, to include at least two straton models in training pattern, and successively treating training
When every straton model of model is trained, for every straton model to the last layer non-in training pattern, passing through instruction
It is every in the straton model output training set based on training completion after the completion of practicing the sample data concentrated to straton model training
A sample data corresponds to the confidence level of positive sample data, and is greater than the layer using the confidence level for corresponding to positive sample data in training set
The sample data of the corresponding confidence threshold value of submodel, is updated the sample data in training set, guarantees successively be directed to
When every straton model of training pattern is trained, the interference of layer-by-layer exclusive segment negative sample data, so as to training pattern
In, for training the sample data of every straton model to have differences, every straton model can extract the different characteristic of sample data,
It ensure that the abundant study to training pattern to sample data feature, to improve the precision of prediction of model.
Embodiment 2:
In order to guarantee the effect to the training of every straton model, on the basis of the above embodiments, in embodiments of the present invention,
Described will include that each sample data of positive sample or negative sample label is input in the straton model in training set, to the layer
Before submodel is trained, the method also includes:
Judge whether the quantity of sample data in the training set is greater than the amount threshold of setting；
If so, carrying out subsequent step；
If not, issuing warning information.
Specifically, the accuracy of submodel can be reduced if the quantity for the sample data being trained to submodel is very few
It reduces, in embodiments of the present invention, the quantity for the sample data being trained in order to prevent to submodel is very few, to submodel
Before being trained, whether the quantity of training of judgement concentration sample data is greater than the amount threshold of setting；If so, carrying out subsequent
To the process that submodel is trained, if it is not, then issuing warning information, prompt user to the sample data volume of sub- model training
It is very few.
Embodiment 3:
Fig. 3 is a kind of prediction process schematic based on above-mentioned model training process provided in an embodiment of the present invention, the mistake
Journey includes:
S301: data to be tested are input in the model of training completion.
S302: the model completed based on the training, output predict whether the data to be tested are positive sample data
As a result.
Specifically, after the completion of the training set by the inclusion of a large amount of positive sample data and negative sample data is to model training,
Data to be tested are input in training pattern, every straton model that the model that training is completed is completed based on training, successively basis
Every straton model that training is completed determines that data to be tested correspond to the confidence level of positive sample data, and submodel determination to
Detection data corresponds to the confidence level of positive sample data, when being greater than the corresponding confidence threshold value of the submodel, data to be tested are defeated
Enter into next straton model of the straton model, until data to be tested to be input to the last layer of the model of training completion
Submodel stops.If data to be tested be input to training completion model the last layer submodel, and it is described last
Straton model determines that data to be detected correspond to the confidence level of positive sample data, corresponding greater than the last layer submodel to set
Confidence threshold, output prediction data to be tested are positive the result of sample data；If data to be tested are not input to trained completion
Model the last layer submodel or the last layer submodel determine that data to be detected correspond to setting for positive sample data
Reliability, is not more than the corresponding confidence threshold value of the last layer submodel, and output prediction data to be tested are negative sample data
Result.
Preferably, knowing for the ease of user to prediction result, the sample data if prediction data to be tested are positive, electricity
Sub- equipment can be sent out warning information, and user is prompted to pay attention to.Such as: the model that training is completed is to be to credit card transaction data
The no model for the credit card transaction data prediction of transaction swindling occurs, if the model prediction credit to be detected that training is completed
Card transaction data is positive sample data, that is, the credit card transaction data of transaction swindling occurs, and issues warning information, reminds user's note
Meaning.
It is predicted based on above-mentioned model training method, is applied to from credit card transaction data and searches generation transaction swindling
Credit card transaction data, the model that training is completed can filter out not more than 1000 from 100,000 credit card transaction datas
Credit card transaction data carries out manual review, can find out the credit card transaction data of existing 50 transactions fraud, greatly
Manpower is saved.
Embodiment 4:
Fig. 4 is a kind of model training apparatus structural schematic diagram provided in an embodiment of the present invention, which includes:
Identification module 41, for successively for every straton model to training pattern, identifying whether the straton model is institute
The last layer submodel to training pattern is stated, wherein described include at least two straton models to training pattern；
Training module 42, if being the non-the last layer submodel to training pattern for submodel, by training set
In included that each sample data of positive sample or negative sample label is input in the straton model, which is instructed
Practice；And each sample data corresponds to the confidence level of positive sample data in the straton model output training set completed based on training,
It is greater than the sample number of the corresponding confidence threshold value of the straton model using the confidence level for corresponding to positive sample data in the training set
According to being updated to the sample data in the training set；
Training module 42, if being also used to submodel is the last layer submodel to training pattern, by training set
In included that each sample data of positive sample or negative sample label is input in the straton model, which is instructed
Practice.
Described device further include:
Alarm module 43 is judged, for judging whether the quantity of sample data in the training set is greater than the quantity threshold of setting
Value, and when the judgment result is yes, training module is triggered, when the judgment result is No, issues warning information.
Embodiment 5:
Fig. 5 is a kind of prediction meanss structure based on model training apparatus as shown in Figure 4 provided in an embodiment of the present invention
Schematic diagram, the device include:
Input module 51, for data to be tested to be input in the model of training completion；
Output module 52, the model for being completed based on the training, output predict whether the data to be tested are positive
The result of sample data.
Described device further include:
Alarm module 53 issues warning information for the sample data if it is predicted that data to be tested are positive.
The invention discloses a kind of model training and prediction technique and device based on model training, which comprises
Successively for every straton model to training pattern, identify whether the straton model is described last straton to training pattern
Model, wherein described include at least two straton models to training pattern；If not, will include positive sample or negative sample in training set
Each sample data of this label is input in the straton model, is trained to the straton model；And completed based on training
Each sample data corresponds to the confidence level of positive sample data in straton model output training set, using corresponding in the training set
The confidence level of positive sample data is greater than the sample data of the corresponding confidence threshold value of the straton model, to the sample in the training set
Notebook data is updated；If so, will include that each sample data of positive sample or negative sample label is input in training set
In the straton model, which is trained.Due in embodiments of the present invention, to include at least two in training pattern
Straton model, and when the every straton model for successively treating training pattern is trained, for in training pattern it is non-last
Every straton model of layer is completed after the completion of by the sample data in training set to straton model training based on training
Straton model output training set in each sample data correspond to the confidence levels of positive sample data, and using correspondence in training set
The confidence level of positive sample data is greater than the sample data of the corresponding confidence threshold value of the straton model, to the sample number in training set
According to being updated, guarantee successively be directed to layer-by-layer exclusive segment negative sample when every straton model of training pattern is trained
The interference of data, so that, for training the sample data of every straton model to have differences, every straton model can in training pattern
To extract the different characteristic of sample data, the abundant study to training pattern to sample data feature ensure that, to improve
The precision of prediction of model.
For systems/devices embodiment, since it is substantially similar to the method embodiment, so the comparison of description is simple
Single, the relevent part can refer to the partial explaination of embodiments of method.
It should 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 reality combining software and hardware aspects can be used in the application
Apply the form of example.Moreover, it wherein includes the computer of computer usable program code that the application, 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 application is referring to method, the process of equipment (system) and computer program product according to the embodiment of the present application
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.
Although the preferred embodiment of the application has been described, it is created once a person skilled in the art knows basic
Property concept, then additional changes and modifications can be made to these embodiments.So it includes excellent that the following claims are intended to be interpreted as
It selects embodiment and falls into all change and modification of the application range.
Obviously, various changes and modifications can be made to the invention without departing from essence of the invention by those skilled in the art
Mind and range.In this way, if these modifications and changes of the present invention belongs to the range of the claims in the present invention and its equivalent technologies
Within, then the present invention is also intended to include these modifications and variations.
Claims (8)
1. a kind of model training method, which is characterized in that the described method includes:
Successively for every straton model to training pattern, identify whether the straton model is last to training pattern
Straton model, wherein described include at least two straton models to training pattern；
If not, will include that each sample data of positive sample or negative sample label is input to the straton model in training set
In, which is trained；And each sample data pair in the straton model output training set completed based on training
The confidence level for answering positive sample data, it is corresponding greater than the straton model using the confidence level for corresponding to positive sample data in the training set
Confidence threshold value sample data, the sample data in the training set is updated；
If so, will include that each sample data of positive sample or negative sample label is input to the straton model in training set
In, which is trained.
2. the method as described in claim 1, which is characterized in that described will include positive sample or negative sample label in training set
Each sample data be input in the straton model, before being trained to the straton model, the method also includes:
Judge whether the quantity of sample data in the training set is greater than the amount threshold of setting；
If so, carrying out subsequent step；
If not, issuing warning information.
3. a kind of prediction technique based on the described in any item model training methods of claim 1-2, which is characterized in that the side
Method includes:
Data to be tested are input in the model of training completion；
Based on the model that the training is completed, output predicts whether the data to be tested are positive the result of sample data.
4. method as claimed in claim 3, which is characterized in that the sample data if it is predicted that data to be tested are positive, institute
State method further include:
Issue warning information.
5. a kind of model training apparatus, which is characterized in that described device includes:
Identification module, for successively for every straton model to training pattern, identifying whether the straton model is described wait instruct
Practice the last layer submodel of model, wherein described include at least two straton models to training pattern；
Training module will wrap if being the non-the last layer submodel to training pattern for submodel in training set
Each sample data containing positive sample or negative sample label is input in the straton model, is trained to the straton model；And
Each sample data corresponds to the confidence level of positive sample data in the straton model output training set completed based on training, using institute
It states and corresponds to sample data of the confidence level of positive sample data greater than the corresponding confidence threshold value of the straton model in training set, to institute
The sample data stated in training set is updated；
Training module will wrap if being also used to submodel is the last layer submodel to training pattern in training set
Each sample data containing positive sample or negative sample label is input in the straton model, is trained to the straton model.
6. device as claimed in claim 5, which is characterized in that described device further include:
Judge alarm module, for judging whether the quantity of sample data in the training set is greater than the amount threshold of setting, and
When the judgment result is yes, training module is triggered, when the judgment result is No, issues warning information.
7. a kind of prediction meanss based on the described in any item model training apparatus of claim 5-6, which is characterized in that the dress
It sets and includes:
Input module, for data to be tested to be input in the model of training completion；
Output module, the model for being completed based on the training, output predict whether the data to be tested are positive sample number
According to result.
8. device as claimed in claim 7, which is characterized in that described device further include:
Alarm module issues warning information for the sample data if it is predicted that data to be tested are positive.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810973101.5A CN109242165A (en) | 2018-08-24 | 2018-08-24 | A kind of model training and prediction technique and device based on model training |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810973101.5A CN109242165A (en) | 2018-08-24 | 2018-08-24 | A kind of model training and prediction technique and device based on model training |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN109242165A true CN109242165A (en) | 2019-01-18 |
Family
ID=65067882
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201810973101.5A Pending CN109242165A (en) | 2018-08-24 | 2018-08-24 | A kind of model training and prediction technique and device based on model training |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109242165A (en) |
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110263824A (en) * | 2019-05-29 | 2019-09-20 | 阿里巴巴集团控股有限公司 | The training method of model, calculates equipment and computer readable storage medium at device |
| CN111461164A (en) * | 2020-02-25 | 2020-07-28 | 清华大学 | Sample data set capacity expansion method and model training method |
| CN111477219A (en) * | 2020-05-08 | 2020-07-31 | 合肥讯飞数码科技有限公司 | Keyword distinguishing method and device, electronic equipment and readable storage medium |
| CN113804982A (en) * | 2021-09-18 | 2021-12-17 | 山东大学 | Solar radio burst real-time detection method and system based on digital filtering |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102208009A (en) * | 2010-03-31 | 2011-10-05 | 索尼公司 | Classifier and classification method |
| CN107133628A (en) * | 2016-02-26 | 2017-09-05 | 阿里巴巴集团控股有限公司 | A kind of method and device for setting up data identification model |
| CN108228469A (en) * | 2018-02-23 | 2018-06-29 | 科大讯飞股份有限公司 | test case selection method and device |
| CN108288161A (en) * | 2017-01-10 | 2018-07-17 | 第四范式（北京）技术有限公司 | The method and system of prediction result are provided based on machine learning |
| CN108416250A (en) * | 2017-02-10 | 2018-08-17 | 浙江宇视科技有限公司 | Demographic method and device |
-
2018
- 2018-08-24 CN CN201810973101.5A patent/CN109242165A/en active Pending
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102208009A (en) * | 2010-03-31 | 2011-10-05 | 索尼公司 | Classifier and classification method |
| CN107133628A (en) * | 2016-02-26 | 2017-09-05 | 阿里巴巴集团控股有限公司 | A kind of method and device for setting up data identification model |
| CN108288161A (en) * | 2017-01-10 | 2018-07-17 | 第四范式（北京）技术有限公司 | The method and system of prediction result are provided based on machine learning |
| CN108416250A (en) * | 2017-02-10 | 2018-08-17 | 浙江宇视科技有限公司 | Demographic method and device |
| CN108228469A (en) * | 2018-02-23 | 2018-06-29 | 科大讯飞股份有限公司 | test case selection method and device |
Non-Patent Citations (2)
| Title |
|---|
| 张辰: "《复杂环境中运动目标检测与跟踪研究》", 31 August 2014 * |
| 黄宏伟 等: "《枫林学苑21》", 30 April 2018 * |
Cited By (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110263824A (en) * | 2019-05-29 | 2019-09-20 | 阿里巴巴集团控股有限公司 | The training method of model, calculates equipment and computer readable storage medium at device |
| CN110263824B (en) * | 2019-05-29 | 2023-09-05 | 创新先进技术有限公司 | Model training method, device, computing equipment and computer readable storage medium |
| CN111461164A (en) * | 2020-02-25 | 2020-07-28 | 清华大学 | Sample data set capacity expansion method and model training method |
| CN111461164B (en) * | 2020-02-25 | 2024-04-12 | 清华大学 | Sample data set capacity expansion method and model training method |
| CN111477219A (en) * | 2020-05-08 | 2020-07-31 | 合肥讯飞数码科技有限公司 | Keyword distinguishing method and device, electronic equipment and readable storage medium |
| CN113804982A (en) * | 2021-09-18 | 2021-12-17 | 山东大学 | Solar radio burst real-time detection method and system based on digital filtering |
| CN113804982B (en) * | 2021-09-18 | 2022-06-21 | 山东大学 | Solar radio burst real-time detection method and system based on digital filtering |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109242165A (en) | A kind of model training and prediction technique and device based on model training | |
| CN110751557B (en) | Abnormal fund transaction behavior analysis method and system based on sequence model | |
| CN110111113B (en) | Abnormal transaction node detection method and device | |
| CN110738564A (en) | Post-loan risk assessment method and device and storage medium | |
| CN111177250A (en) | Abnormal transaction monitoring method, system and storage medium | |
| CN110287316A (en) | A kind of Alarm Classification method, apparatus, electronic equipment and storage medium | |
| CN110852881B (en) | Risk account identification method and device, electronic equipment and medium | |
| CN111931047B (en) | Artificial intelligence-based black product account detection method and related device | |
| CN110782349A (en) | Model training method and system | |
| CN113657896A (en) | Block chain transaction topological graph analysis method and device based on graph neural network | |
| CN117235608B (en) | Risk detection method, risk detection device, electronic equipment and storage medium | |
| CN113449753B (en) | Service risk prediction method, device and system | |
| CN116611911A (en) | Credit risk prediction method and device based on support vector machine | |
| CN110544166A (en) | Sample generation method, device and storage medium | |
| CN112346995B (en) | Banking industry-based test risk prediction model construction method and device | |
| CN113987351A (en) | Artificial intelligence based intelligent recommendation method and device, electronic equipment and medium | |
| CN113781056A (en) | Method and device for predicting user fraud behavior | |
| Lawrencia et al. | Fraud detection decision support system for Indonesian financial institution | |
| CN112396513B (en) | Data processing method and device | |
| CN110415099A (en) | A kind of credit financing swindle recognition methods, system and electronic equipment | |
| CN114565030B (en) | Feature screening method and device, electronic equipment and storage medium | |
| CN112651172B (en) | Rainfall peak type dividing method, device, equipment and storage medium | |
| Sapozhnikova et al. | Distributed infrastructure for big data processing in the transaction monitoring systems | |
| CN115187377A (en) | Credit card risk information detection method and device | |
| CN116188179A (en) | Abnormality detection method and device based on artificial intelligence, computer equipment and 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 | ||
| TA01 | Transfer of patent application right | ||
| TA01 | Transfer of patent application right |
Effective date of registration: 20200825 Address after: 501, 5 / F, block B, No. 28, xinjiekouwei street, Xicheng District, Beijing 100032 Applicant after: Joint digital technology (Beijing) Co., Ltd Address before: 100082 No. 508, 5th floor, Block B, 28 Xinjiekouwai Street, Xicheng District, Beijing Applicant before: MIXIAOFENG WISDOM (BEIJING) TECHNOLOGY Co.,Ltd. |
|
| RJ01 | Rejection of invention patent application after publication | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20190118 |