CN105786830A - Method, device and system for self-adaptively adjusting models in computer systems - Google Patents
Method, device and system for self-adaptively adjusting models in computer systems Download PDFInfo
- Publication number
- CN105786830A CN105786830A CN201410805857.0A CN201410805857A CN105786830A CN 105786830 A CN105786830 A CN 105786830A CN 201410805857 A CN201410805857 A CN 201410805857A CN 105786830 A CN105786830 A CN 105786830A
- Authority
- CN
- China
- Prior art keywords
- outcome
- module
- packet
- tuning function
- predicting
- 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
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The invention relates to a method, device and system for self-adaptively adjusting models in computer systems. The method comprises the following steps: obtaining behavior data of the current observation point in a first performing stage; obtaining a prediction result output by a target model at the moment of a first observation point, wherein the first observation point is a starting time point of the first performing stage, and the prediction result is a probability estimation value, predicted by the target model, of a preset event occurring in the first performing stage; carrying out adjustment function training by utilizing the obtained behavior data and prediction result so as to obtain an adjustment function of the target model; and adjusting the prediction result output by the target model at the moment of the current observation point by utilizing the adjustment function of the target model. According to the method, device and system, self-adaptive adjustment can be carried out on the models in systems, so that adjustment response can be rapidly and efficiently carried out on the change of user behavior characteristics and the properties of the models of the systems is ensured.
Description
Technical field
The application relates to field of computer technology, particularly relates to model adaptation method of adjustment, Apparatus and system in a kind of computer system.
Background technology
At present, in the computer system of every field, increasing computer system is assisted carry out decision-making by being disposed corresponding model, and model provides decision-making foundation with its excellent precision of prediction and ranking function for all types of computer systems.
But, owing to the model in computer system obtains according to historical information training, As time goes on, sample can change to migrate and cause that declining occurs in the precision of model.
In order to solve the problem that this model performance declines, the method commonly used at present is reconstruction model over time, become, and variable parameter is adjusted, or a brand-new model of redeveloping out, then in systems original model is updated or replaces.
No matter it is reconstruction model or develops brand-new model, be required for a lot of resource input, and, from being reconstructed into, redeploying reaches the standard grade needs the regular hour to model, causes that the renewal of model also exists certain hysteresis quality.And in some applications, due to user behavior changing features quickly, it is necessary to model can be adjusted response fast and efficiently, and the method for Model Reconstruction is to be unable to reach this requirement.
Summary of the invention
The purpose of the application is, there is provided model adaptation method of adjustment, Apparatus and system in a kind of computer system, it is possible to the model in system is carried out self-adaptative adjustment, save cost, realize the change of user behavior feature is adjusted response fast and efficiently, it is ensured that the performance of system model.
This application provides model adaptation method of adjustment in a kind of computer system, described method includes:
Behavioral data in first performance phase before acquisition Current observation point；
Obtain object module predicting the outcome of exporting of the first observation station moment, described first observation station is the start time point of described first performance phase, described in predict the outcome, for the prediction of described object module, the probabilistic estimated value of described predeterminable event occurred within the described first performance phase；
Utilize the described behavioral data obtained and be adjusted function training with predicting the outcome, obtain the Tuning function of described object module；
Utilize the Tuning function of described object module, described object module is adjusted predicting the outcome of exporting of described Current observation point moment.
Another aspect, present invention also provides model adaptation adjusting apparatus in a kind of computer system, and described device includes:
First acquiring unit, the behavioral data in the first performance phase before acquisition Current observation point；
Second acquisition unit, obtain object module predicting the outcome of exporting of the first observation station moment, described first observation station is the start time point of described first performance phase, described in predict the outcome, for the prediction of described object module, the probabilistic estimated value of described predeterminable event occurred within the described first performance phase；
Training unit, utilize described first acquiring unit obtain described behavioral data and described second acquisition unit obtain described in predict the outcome be adjusted function training, obtain the Tuning function of described object module；
Adjustment unit, utilizes the Tuning function of the described object module that described training unit obtains, and described object module is adjusted predicting the outcome of exporting of described Current observation point moment.
Another aspect, present invention also provides a kind of model adaptation system, for the object module in goal systems is carried out self-adaptative adjustment, described goal systems includes data memory module, model module and decision-making module, described object module is the model that described model module calls, and described model adaptation system includes: data acquisition module, Tuning function training module and adjusting module；
Described data acquisition module, is connected with the data memory module of described goal systems, the behavioral data in the first performance phase before acquisition Current observation point from the data memory module of described goal systems；And, obtain described object module predicting the outcome of exporting of the first observation station moment, described first observation station is the start time point of described first performance phase, described in predict the outcome, for the prediction of described object module, the probabilistic estimated value of described predeterminable event occurred within the described first performance phase；
Described Tuning function training module, utilizes the described behavioral data that described data acquisition module obtains to be adjusted function training with predicting the outcome, obtains the Tuning function of described object module；
Described adjusting module, it is connected with described decision-making module, utilize the Tuning function of the described object module that described Tuning function training module obtains, described object module is adjusted predicting the outcome of exporting of described Current observation point moment, and by the output that predicts the outcome after adjusting to described decision-making module.
Model adaptation method of adjustment, Apparatus and system in the computer system that the embodiment of the present application provides, the agenda data in the performance phase are utilized to be adjusted function training with predicting the outcome of this performance phase initial time model output, self-adaptative adjustment is carried out with the estimated performance to the model in system, save cost, realize the change of user behavior feature is adjusted response fast and efficiently, it is ensured that the performance of system model.
Accompanying drawing explanation
In order to be illustrated more clearly that the embodiment of the present invention or technical scheme of the prior art, the accompanying drawing used required in embodiment or description of the prior art will be briefly described below, apparently, accompanying drawing in the following describes is only some embodiments of the present invention, for those of ordinary skill in the art, under the premise not paying creative work, it is also possible to obtain other accompanying drawing according to these accompanying drawings.
Fig. 1 is the structural representation of a kind of typical automated decision-making system；
The structural representation of model adaptation system in a kind of computer system that Fig. 2 provides for the embodiment of the present application；
Model adaptation method of adjustment flow chart in a kind of computer system that Fig. 3 provides for the embodiment of the present application；
The schematic diagram showing phase and observation station that Fig. 4 provides for the embodiment of the present application；
The flow chart of a kind of Tuning function training method that Fig. 5 provides for the embodiment of the present application；
Fig. 6 truly exceeds the time limit for the sample that the embodiment of the present application provides a kind of relation curve schematic diagram of rate and average PD scoring；
Model adaptation adjusting apparatus schematic diagram in a kind of computer system that Fig. 7 provides for the embodiment of the present application；
The structural representation of the training unit that Fig. 8 provides for the embodiment of the present application.
Detailed description of the invention
For making present invention purpose, feature, the advantage can be more obvious and understandable, below in conjunction with the accompanying drawing in the embodiment of the present application, the technical scheme in the embodiment of the present application is described, it is clear that, described embodiment is only some embodiments of the present application, and not all embodiments.Based on the embodiment in the application, the every other embodiment that those of ordinary skill in the art obtain under not making creative work premise, broadly fall into the scope of the application protection.
In the computer system that the embodiment of the present application provides, model adaptation method of adjustment, device and model adaptation adjust system, can be used in the computer system of every field, in order to utilize machine model to carry out decision-making or management, such as, in marketing system, it is possible to the prediction that the marketing activity whether user is provided businessman responds；In stock market, it is possible to the Future price ups and downs of certain stock are made prediction；In e-commerce field, whether user can be bought certain commodity by model within following a period of time is made prediction, etc..The decision system in these fields can be disposed the various models having developed and being proved to certain effect.Various models can provide good decision-making foundation for system with its excellent precision of prediction and ranking function, and particularly at big data age, the model in computer system can provide excellent reference frame to enterprise or individual in operational decision making.The self-adaptative adjustment of various models is possible not only to save the resource putting into reconstruction model or development model again, it is also possible in time model is carried out self-adaptative adjustment, it is achieved adjustment response rapidly and efficiently, it is ensured that the performance of system model.
Fig. 1 is the structural representation of a kind of typical automated decision-making system, as it is shown in figure 1, this decision system 1 includes: model module 11, data memory module 12 and decision-making module 13.
Wherein, data memory module 12 is used for storing behavioral data (the transaction details data etc. such as user), base values that model needs and the data such as output data (predicting the outcome) at each observation station model.
Model module 11 is at fixing Preset Time point (such as first day every month), the base values that object module from the data memory module 12 acquisition system of decision system 1 is corresponding, object module in operation system, give a mark to all users, obtain predicting the outcome that this observation station exports, and will predict the outcome and store to data memory module 12.
Decision-making module 13 when needs decision system makes decisions, can extract model in corresponding the predicting the outcome of user sometime from data memory module 12, to make certain decision-making to this user, such as refusal user's access or calculating user's loan limit etc..
Fig. 2 is the structural representation of model adaptation system in a kind of computer system that the embodiment of the present application provides, as shown in Figure 2, after the data memory module 12 that this model adaptation system 2 is deployed in decision system 1 and before decision-making module 13, make decision-making module 13 directly can not obtain predicting the outcome of object module output from data memory module 12, for the object module in this decision system 1 is carried out self-adaptative adjustment, wherein, object module is the model that model module 11 runs.
This model adaptation system 2 includes data acquisition module 21, Tuning function training module 22 and adjusting module 23.Data acquisition module 21 is connected with the data memory module 12 of decision system 1, and adjusting module 23 is connected with the decision-making module 13 of decision system 1.
Specifically, data acquisition module 21 obtains the behavioral data in a period of time from data memory module 12, and obtains the predicting the outcome of initial time model output of this period of time.Tuning function training module 22 utilizes the behavioral data that data acquisition module 21 obtains to be adjusted function training with predicting the outcome, and obtains the Tuning function of object module.Predicting the outcome that adjusting module 23 utilizes that object module exports by the Tuning function of the object module that Tuning function training module 22 obtains is adjusted, and by the output that predicts the outcome after adjusting to decision-making module 13.When decision system 1 needs the scoring utilizing object module to carry out decision-making, decision system 1 will trigger model Adaptable System 2.Adjusting module 23 first obtains the object module output result at Current observation point from data memory module 12, described output result is adjusted by the Tuning function utilizing the object module that Tuning function training module 22 obtains, the scoring of the object module after being adjusted, it is provided that use to decision-making module 13.Certainly, the Tuning function of the object module that Tuning function training module 22 obtains can also be stored to data memory module 12 by adjusting module 23, in order to follow-up use.
With model adaptation system in Fig. 2 accordingly, Fig. 3 is model adaptation method of adjustment flow chart in the computer system that the embodiment of the present application provides, as it is shown on figure 3, the method for adjustment of the embodiment of the present application includes:
Behavioral data in first performance phase before S101, acquisition Current observation point.
Observation station refers to the time point of object module periodic operation.Described object module can include the models such as forecast model, decision model, administrative model.
Data in the previous performance phase utilizing observation station, as the input data of object module, by running the object module reserved in advance, are predicted the outcome.Wherein, the performance phase is a period of time set in advance, it is possible to is adjusted according to actual needs and arranges, such as half a year or 3 months etc..
In general, object module only point (i.e. each observation station) operation at a fixed time, for instance every month is run once, then corresponding the predicting the outcome of output.The server of decision system can utilize data memory module predicting the outcome of each observation station to be stored to data base.
Current observation point refers to the nearest observation station of front distance current point in time of current point in time or current point in time.The described first performance phase is the interval from described first observation station to Current observation point.As shown in Figure 4, if current time is in JIUYUE, 2014, a performance phase is 6 months, runs an object module every month, namely has an observation station corresponding every month.So, Current observation point is JIUYUE in 2014 1, and the first observation station that distance Current observation point is a performance phase is on March 1st, 2014, and the first performance phase was in August, 2014 in March, 2014 to.
Behavioral data corresponding in the first performance phase before obtaining Current observation point from the data base of system, for instance, Current observation point is in JIUYUE, 2014, then obtain the behavioral data between in August, 2014 in March, 2014 to.For the accuracy calculated, the behavioral data of acquisition includes multiple user or the behavioral data of whole user or whole validated user.
S102, obtain object module predicting the outcome of exporting of the first observation station moment.
Described first observation station is the start time point of described first performance phase, described in predict the outcome, for the prediction of described object module, the probabilistic estimated value of described predeterminable event occurred within the described first performance phase.
Such as, object module predicting the outcome in March, 2014 output of the first observation station is obtained.
Accordingly, for the accuracy calculated, the predicting the outcome of acquisition also includes multiple user or all user or all the predicting the outcome of validated user.
It is noted that S102 can perform before S101, it is also possible to perform with S101 simultaneously.
S103, utilize the described behavioral data and predicting the outcome obtained to be adjusted function training, obtain the Tuning function of described object module.
Utilize predicting the outcome of acquisition in the behavioral data obtained in S101 and S102 to be analyzed as data sample, be adjusted function.
Alternatively, S103 can be, but not limited to adopt the method shown in Fig. 5 to be analyzed and Function Fitting, specifically includes:
S1031, the plurality of user being grouped, form N number of packet, N is positive integer.
Specifically, it is possible to the size of probabilistic estimated value according to the predicting the outcome of the plurality of user, each user described is ranked up packet.
First it is ranked up from high in the end according to the predicting the outcome of model output obtained in S102, is divided into N number of packet according to total number of users amount according to the mode of the crowd of grade.Wherein, the needs that arrange of N determine a suitable numerical value according to total number of users, it is ensured that the number of each packet can not be very little.
S1032, according to described behavioral data, calculate the actual event incidence rate of predeterminable event described in described N number of packet.
Predeterminable event is the event that object module to be predicted, for instance, it was predicted that the model of (loan) rate of exceeding the time limit, then predeterminable event is user's overdue loan event；The model of prediction buying rate, then predeterminable event is the event etc. that user buys certain commodity.
Specifically, it is possible to according to the behavioral data of each user in described packet, it is judged that whether described user, within the described first performance phase, described predeterminable event occurs；There is the number of users of described predeterminable event and the ratio of the total number of users amount in described packet in statistics, as the actual event incidence rate of predeterminable event described in described packet within the described first performance phase.
Predict the outcome described in S1033, basis, calculate the average probability estimated value of predeterminable event described in described N number of packet.
According to the predicting the outcome of each user in the model output grouping obtained in S102, adopt the mode that summation is averaged to calculate and obtain the average probability estimated value of predeterminable event in each packet.
S1034, according to the actual event incidence rate in described N number of packet and average probabilistic estimated value, matching obtains the Tuning function of described object module.
According to the actual event incidence rate in described N number of packet and average probabilistic estimated value, matching obtains the plurality of Tuning function, and calculates the goodness of fit of the plurality of Tuning function；The goodness of fit according to described Tuning function, chooses the Tuning function obtaining described object module.
If the actual event incidence rate obtained in S1032 is designated as event_rate, the average probability estimated value obtained in S1033 is designated as predict_avg, then can obtain N group data: (event_rate (i), predict_avg (i)), 1≤i≤N.
With predict_avg for independent variable, event_rate is dependent variable, and the absolute value sum to obtain the difference of described actual event incidence rate event_rate and average probability estimated value predict_avg minimizesFor target, adopt different Function Fitting algorithms, for instance, method of least square, EM algorithm etc., simulate a plurality of curve (such as straight line, exponential curve, quadratic function curve, loaarithmic curve etc.), obtain multiple Tuning function, and calculate the goodness of fit of each Tuning function.Generally, for linear function, can automatically adopt least square fitting, and for nonlinear function, then EM algorithm can be adopted to be fitted.
The goodness of fit can adopt determines that coefficients R-square characterizes.Determine that coefficients R-square is the quality that the change by data characterizes a fitting result, determine that the normal span of coefficient is for [0,1], it is determined that coefficient is closer to 1, show that the variable of equation is more strong to the interpretability of function, also better to data fitting of this model.Determining that coefficients R-square is determined by SSR and SST, R-square=SSR+SST, wherein SSR is the quadratic sum of prediction data and the difference of initial data average, and SST is the quadratic sum of the difference of initial data and average.
From multiple Tuning function, choose the Tuning function that the goodness of fit is minimum, as the functional form best embodying the true corresponding relation of data, and the parameter of functional form and estimation is stored in data base.
S104, utilize the Tuning function of described object module, described object module is adjusted predicting the outcome of exporting of described Current observation point moment.
Specifically include: for object module predicting the outcome of exporting of Current observation point moment, utilize the Tuning function that S103 obtains, predicting the outcome of Current observation point is adjusted.
Alternatively, after S104, it is also possible to including: carry out decision-making and management according to predicting the outcome after described adjustment.Specifically, it is possible to by the output that predicts the outcome after adjustment to the decision-making module of system, to carry out decision-making and management according to the model score after adjusting.
Give an example, the prediction loan user run in the decision system of some mechanism will show the rate appraising model (being called PD model) that exceeds the time limit that whether can exceed the time limit in the phase half a year in future, when current point in time, it is necessary to the situation that this model carries out self-adaptative adjustment illustrates.The model adaptation that the embodiment of the present application provides adjusts system carry in the rear end of the decision system of this mechanism.PD model in decision system runs once beginning of the month every month, and 6 months futures that every month is run are a performance phase.
If current time is in March, 2014, the first performance phase was then in JIUYUE, 2013 in February, 2014, and wherein the first observation station is in JIUYUE, 2013, then first obtain in this decision system the loan user behavior data in JIUYUE, 2013 in February, 2014.Obtain PD model predicting the outcome in JIUYUE, 2013 output of the first observation station, utilize in JIUYUE, 2013 in the first performance phase obtained to be adjusted function training to data and the PD model in February, 2014 predicting the outcome of the first observation station in JIUYUE, 2013 output, obtain the Tuning function of this PD model.
Specifically, first the loan user in JIUYUE, 2013 in February, 2014 is ranked up according to PD model predicting the outcome of exporting of in JIUYUE, 2013 and is grouped, for instance be divided equally into 20 packets that number of users is identical.The loan user behavior data of each packet in being grouped for 20, it is judged that whether the user in each packet there occurs overdue loan event (predeterminable event), calculates the rate of exceeding the time limit that each packet is actual.Rate of exceeding the time limit is the number of users that overdue loan event occurs and the ratio of total number of users in packet.The PD score (predicting the outcome) utilizing the PD model of each loan user obtained export in JIUYUE, 2013, calculates the average PD score of user during each is grouped.Concrete outcome is as shown in table 1 below:
Table 1
| Packet | Number | Rate of exceeding the time limit (event_rate) | Average PD score (predict_avg) |
| 1 | 394 | 0.269035533 | 0.37467808 |
| 2 | 394 | 0.154822335 | 0.26577529 |
| 3 | 394 | 0.116751269 | 0.23224488 |
| 4 | 394 | 0.086294416 | 0.17125118 |
| 5 | 394 | 0.068527919 | 0.12541541 |
| 6 | 394 | 0.07106599 | 0.14438414 |
| …… | …… | …… | …… |
Utilize exceed the time limit rate and these 20 groups of data of average PD score of each packet in 20 packets, using average PD score as independent variable, rate of exceeding the time limit as dependent variable, a plurality of function of matching adaptively, and calculate the R-square judging quota as the goodness of fit.
After each packet fitting function of obtaining of matching is contrasted, showing that the best fitting function of goodness of fit is as the Tuning function of described PD model, is illustrated in figure 6 a linear function y=0.89x-0.0144, the goodness of fit is 0.9853.From fig. 6 it can be seen that utilize the rate of truly exceeding the time limit that the loan user behavior data in JIUYUE, 2013 in February, 2014 obtains linear with average PD score, functional form is trained adaptively by the approximating method of method of least square and is obtained.Finally, then to PD model in March, 2014 predicting the outcome of exporting be adjusted.
Further, in order to verify the effect of model adaptation method of adjustment in the computer system that the embodiment of the present application provides, it is possible to be verified by the data in above-mentioned example.Specifically, using in JIUYUE, 2013 to the data in November, 2013 as training dataset, in December, 2013 to the data in February, 2014 as test data set.The basic condition of the behavioral data obtained is as shown in table 2 below:
Table 2
| Month | Colony | Number of users | Exceed the time limit rate |
| 2013/09-2013/11 | Training set | 56,000 | 6.30% |
| 2013/12-2014/02 | Test set | 36,000 | 6.80% |
Utilizing the data in training set to be adjusted the matching of function, the result for matching adopts the data of test set to be verified.
Similarly, first the loan user in JIUYUE, 2013 in February, 2014 is ranked up packet according to predicting the outcome of PD model, then for the loan user behavior data in each packet, whether the user judged in each packet there occurs overdue loan event (predeterminable event), calculates the rate of exceeding the time limit that each packet is actual.Utilize the PD score (predicting the outcome) of the PD model output of each loan user obtained, calculate the average PD score of user in each packet.And then, recycling exceed the time limit rate and the average PD score of each packet, matching adaptively obtains the Tuning function of described PD model.
Then, the Tuning function obtained being applied to test data set, and rate of truly exceeding the time limit with it contrasts, point estimation meter comparing result such as table 3 below, rate of truly exceeding the time limit and average PD diversity of values drop to 0.4% from 3.4%.Contrast before and after adjusting in test set is as shown in table 3 below:
Table 3
Therefore, by efficiently utilizing user behavior data, it is possible to predicting the outcome of PD model output is carried out self-adaptative adjustment so that predicting the outcome of the PD model after adjustment is more accurate, and effect is notable.
Model adaptation method of adjustment in the computer system that the embodiment of the present application provides, the agenda data effectively utilized in the performance phase are adjusted function training with predicting the outcome of this performance phase initial time model output, self-adaptative adjustment is carried out with the estimated performance to the model in system, save cost, realize the change of user behavior feature is adjusted response fast and efficiently, it is ensured that the performance of system model.
Being above the detailed description that in the computer system that the embodiment of the present application is provided, model adaptation method of adjustment carries out, in the computer system below the application provided, model adaptation adjusting apparatus is described in detail.
Fig. 7 is model adaptation adjusting apparatus schematic diagram in the computer system that the embodiment of the present application provides, as it is shown in fig. 7, the device of the application includes: the first acquiring unit 601, second acquisition unit 602, training unit 603 and adjustment unit 604.
First acquiring unit 601 obtains the behavioral data before Current observation point in the first performance phase.
Second acquisition unit 602 obtains object module predicting the outcome of exporting of the first observation station moment.
Described first observation station is the start time point of described first performance phase, described in predict the outcome, for the prediction of described object module, the probabilistic estimated value of described predeterminable event occurred within the described first performance phase.
Training unit 603 utilize first acquiring unit 601 obtain described behavioral data and second acquisition unit 602 obtain described in predict the outcome be adjusted function training, obtain the Tuning function of described object module.
Adjustment unit 604 utilizes the Tuning function of the described object module that training unit 603 obtains, and described object module is adjusted predicting the outcome of exporting of described Current observation point moment.
Wherein, the described behavioral data that the first acquiring unit 601 obtains includes the behavioral data of multiple user.
Predict the outcome described in second acquisition unit 602 acquisition and include predicting the outcome of the plurality of user.
Alternatively, as shown in Figure 8, training unit 603 specifically includes: packet subelement the 6031, first computation subunit the 6032, second computation subunit 6033 and the 3rd computation subunit 6034.
The plurality of user is grouped by packet subelement 6031, forms N number of packet, and N is positive integer.
The described behavioral data that first computation subunit 6032 obtains according to the first acquiring unit 601, calculates the actual event incidence rate of predeterminable event described in described N number of packet.
Second computation subunit 6033 predicts the outcome according to second acquisition unit 602 acquisition, calculates the average probability estimated value of predeterminable event described in described N number of packet.
Actual event incidence rate in described N number of packet that 3rd computation subunit 6034 obtains according to the first computation subunit 6032 and the second computation subunit 6033 and average probabilistic estimated value, matching obtains the Tuning function of described object module.
Alternatively, described in the predicting the outcome of the packet the plurality of user that obtains according to second acquisition unit 602 of subelement 6031, the size of probabilistic estimated value, is ranked up each user described being grouped.
First computation subunit 6032 specifically includes judgment sub-unit and statistics subelement.
Described judgment sub-unit is according to the behavioral data of each user in described packet, it is judged that whether described user, in the performance phase in described very first time interval, described predeterminable event occurs.
There is the number of users of described predeterminable event and the ratio of the total number of users amount in described packet in described statistics subelement statistics, as the actual event incidence rate of predeterminable event described in described packet in the performance phase in described very first time interval.
Alternatively, the 3rd computation subunit 6034 specifically includes matching subelement and chooses subelement.
Described matching subelement is according to the actual event incidence rate in described N number of packet and average probabilistic estimated value, and matching obtains multiple Tuning function, and calculates the goodness of fit of the plurality of Tuning function.
The described goodness of fit choosing the described Tuning function that subelement obtains according to described matching subelement, chooses the Tuning function obtaining described object module.
Alternatively, described device also includes: decision package (not shown).Described decision package carries out decision-making and management according to predicting the outcome after described adjustment unit adjustment.
The self-adaptative adjustment of the said method that the function of above-mentioned each unit may correspond to Fig. 3 detailed description processes step, repeats no more in this.
Model adaptation method of adjustment and device in the computer system that the embodiment of the present application provides, the agenda data effectively utilized in the performance phase are adjusted function training with predicting the outcome of this performance phase initial time model output, self-adaptative adjustment is carried out with the estimated performance to the model in system, save cost, realize the change of user behavior feature is adjusted response fast and efficiently, it is ensured that the performance of system model.
Professional should further appreciate that, the unit of each example described in conjunction with the embodiments described herein and algorithm steps, can with electronic hardware, computer software or the two be implemented in combination in, in order to clearly demonstrate the interchangeability of hardware and software, generally describe composition and the step of each example in the above description according to function.These functions perform with hardware or software mode actually, depend on application-specific and the design constraint of technical scheme.Professional and technical personnel specifically can should be used for using different methods to realize described function to each, but this realization is it is not considered that exceed scope of the present application.
The method described in conjunction with the embodiments described herein or the step of algorithm can use the software module that hardware, processor perform, or the combination of the two is implemented.Software module can be placed in any other form of storage medium known in random access memory (RAM), internal memory, read only memory (ROM), electrically programmable ROM, electrically erasable ROM, depositor, hard disk, moveable magnetic disc, CD-ROM or technical field.
Above-described detailed description of the invention; the purpose of the application, technical scheme and beneficial effect have been further described; it is it should be understood that; the foregoing is only the detailed description of the invention of the application; it is not used to limit the protection domain of the application; all within spirit herein and principle, any amendment of making, equivalent replacement, improvement etc., should be included within the protection domain of the application.
Claims (17)
1. model adaptation method of adjustment in a computer system, it is characterised in that described method includes:
Behavioral data in first performance phase before acquisition Current observation point；
Obtain object module predicting the outcome of exporting of the first observation station moment, described first observation station is the start time point of described first performance phase, described in predict the outcome, for the prediction of described object module, the probabilistic estimated value of described predeterminable event occurred within the described first performance phase；
Utilize the described behavioral data obtained and be adjusted function training with predicting the outcome, obtain the Tuning function of described object module；
Utilize the Tuning function of described object module, described object module is adjusted predicting the outcome of exporting of described Current observation point moment.
2. method according to claim 1, it is characterised in that described behavioral data includes the behavioral data of multiple user；Described predicting the outcome includes predicting the outcome of the plurality of user；
The described described behavioral data utilizing acquisition is adjusted function training with predicting the outcome, and obtains the Tuning function of described object module, specifically includes:
The plurality of user being grouped, forms N number of packet, N is positive integer；
According to described behavioral data, calculate the actual event incidence rate of predeterminable event described in described N number of packet；
Predict the outcome according to described, calculate the average probability estimated value of predeterminable event described in described N number of packet；
According to the actual event incidence rate in described N number of packet and average probabilistic estimated value, matching obtains the Tuning function of described object module.
3. method according to claim 2, it is characterised in that described the plurality of user is grouped, specifically includes:
The size of probabilistic estimated value according to the predicting the outcome of the plurality of user, is ranked up packet to each user described.
4. method according to claim 2, it is characterised in that according to described behavioral data, calculates the actual event incidence rate of predeterminable event described in described N number of packet, specifically includes:
According to the behavioral data of each user in described packet, it is judged that whether described user, in the performance phase in described very first time interval, described predeterminable event occurs；
There is the number of users of described predeterminable event and the ratio of the total number of users amount in described packet in statistics, as the actual event incidence rate of predeterminable event described in described packet in the performance phase in described very first time interval.
5. method according to claim 2, it is characterised in that according to the actual event incidence rate in described N number of packet and average probabilistic estimated value, matching obtains the Tuning function of described object module, specifically includes:
According to the actual event incidence rate in described N number of packet and average probabilistic estimated value, matching obtains the plurality of Tuning function, and calculates the goodness of fit of the plurality of Tuning function；
The goodness of fit according to described Tuning function, chooses the Tuning function obtaining described object module.
6. method according to claim 1, it is characterised in that utilize the Tuning function of described object module, to described object module after the described Current observation point moment, predicting the outcome of exporting was adjusted, also includes:
Decision-making and management is carried out according to predicting the outcome after described adjustment.
7. model adaptation adjusting apparatus in a computer system, it is characterised in that described device includes:
First acquiring unit, the behavioral data in the first performance phase before acquisition Current observation point；
Second acquisition unit, obtain object module predicting the outcome of exporting of the first observation station moment, described first observation station is the start time point of described first performance phase, described in predict the outcome, for the prediction of described object module, the probabilistic estimated value of described predeterminable event occurred within the described first performance phase；
Training unit, utilize described first acquiring unit obtain described behavioral data and described second acquisition unit obtain described in predict the outcome be adjusted function training, obtain the Tuning function of described object module；
Adjustment unit, utilizes the Tuning function of the described object module that described training unit obtains, and described object module is adjusted predicting the outcome of exporting of described Current observation point moment.
8. device according to claim 7, it is characterised in that the described behavioral data that described first acquiring unit obtains includes the behavioral data of multiple user；Predict the outcome described in the acquisition of described second acquisition unit and include predicting the outcome of the plurality of user；
Described training unit specifically includes:
Packet subelement, is grouped the plurality of user, forms N number of packet, and N is positive integer；
First computation subunit, according to the described behavioral data that described first acquiring unit obtains, calculates the actual event incidence rate of predeterminable event described in described N number of packet；
Second computation subunit, according to described second acquisition unit obtain described in predict the outcome, calculate the average probability estimated value of predeterminable event described in described N number of packet；
3rd computation subunit, according to the actual event incidence rate in described N number of packet that described first computation subunit and described second computation subunit obtain and average probabilistic estimated value, matching obtains the Tuning function of described object module.
9. device according to claim 8, it is characterised in that the size of probabilistic estimated value described in the predicting the outcome of the plurality of user that described packet subelement obtains according to described second acquisition unit, is ranked up packet to each user described.
10. device according to claim 8, it is characterised in that described first computation subunit specifically includes:
Judgment sub-unit, according to the behavioral data of each user in described packet, it is judged that whether described user, in the performance phase in described very first time interval, described predeterminable event occurs；
Statistics subelement, there is the number of users of described predeterminable event and the ratio of the total number of users amount in described packet in statistics, as the actual event incidence rate of predeterminable event described in described packet in the performance phase in described very first time interval.
11. device according to claim 8, it is characterised in that described 3rd computation subunit specifically includes:
Matching subelement, according to the actual event incidence rate in described N number of packet and average probabilistic estimated value, matching obtains multiple Tuning function, and calculates the goodness of fit of the plurality of Tuning function；
Choose subelement, the goodness of fit according to the described Tuning function that described matching subelement obtains, choose the Tuning function obtaining described object module.
12. device according to claim 7, it is characterised in that described device also includes:
Decision package, carries out decision-making and management according to predicting the outcome after described adjustment unit adjustment.
13. a model adaptation system, for the object module in goal systems is carried out self-adaptative adjustment, described goal systems includes data memory module, model module and decision-making module, described object module is the model that described model module runs, it is characterized in that, described model adaptation system includes: data acquisition module, Tuning function training module and adjusting module；
Described data acquisition module, is connected with the data memory module of described goal systems, the behavioral data in the first performance phase before acquisition Current observation point from the data memory module of described goal systems；And, obtain described object module predicting the outcome of exporting of the first observation station moment, described first observation station is the start time point of described first performance phase, described in predict the outcome, for the prediction of described object module, the probabilistic estimated value of described predeterminable event occurred within the described first performance phase；
Described Tuning function training module, utilizes the described behavioral data that described data acquisition module obtains to be adjusted function training with predicting the outcome, obtains the Tuning function of described object module；
Described adjusting module, it is connected with described decision-making module, utilize the Tuning function of the described object module that described Tuning function training module obtains, described object module is adjusted predicting the outcome of exporting of described Current observation point moment, and by the output that predicts the outcome after adjusting to described decision-making module.
14. system according to claim 13, it is characterised in that the described behavioral data that described data acquisition module obtains includes the behavioral data of multiple user；Predict the outcome described in the acquisition of described data acquisition module and include predicting the outcome of the plurality of user；
The plurality of user is grouped by described Tuning function training module, forms N number of packet, and N is positive integer；
Described Tuning function training module, according to described behavioral data, calculates the actual event incidence rate of predeterminable event described in described N number of packet；
Described Tuning function training module according to described in predict the outcome, calculate the average probability estimated value of predeterminable event described in described N number of packet；
Described Tuning function training module is according to the actual event incidence rate in described N number of packet and average probabilistic estimated value, and matching obtains the Tuning function of described object module.
15. system according to claim 14, it is characterised in that described Tuning function training module is the size of probabilistic estimated value according to the predicting the outcome of the plurality of user, each user described is ranked up packet.
16. system according to claim 14, it is characterised in that described Tuning function training module is according to the behavioral data of each user in described packet, it is judged that whether described user, in the performance phase in described very first time interval, described predeterminable event occurs；And in the performance phase that statistics is in described very first time interval, there is the number of users of described predeterminable event and the ratio of the total number of users amount in described packet, as the actual event incidence rate of predeterminable event described in described packet.
17. system according to claim 14, it is characterized in that, described Tuning function training module is according to the actual event incidence rate in described N number of packet and average probabilistic estimated value, and matching obtains multiple Tuning function, and calculates the goodness of fit of the plurality of Tuning function；
The goodness of fit according to described Tuning function of the described Tuning function training module, chooses the Tuning function obtaining described object module.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201410805857.0A CN105786830A (en) | 2014-12-19 | 2014-12-19 | Method, device and system for self-adaptively adjusting models in computer systems |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201410805857.0A CN105786830A (en) | 2014-12-19 | 2014-12-19 | Method, device and system for self-adaptively adjusting models in computer systems |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN105786830A true CN105786830A (en) | 2016-07-20 |
Family
ID=56385221
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201410805857.0A Pending CN105786830A (en) | 2014-12-19 | 2014-12-19 | Method, device and system for self-adaptively adjusting models in computer systems |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN105786830A (en) |
Cited By (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2018018520A1 (en) * | 2016-07-28 | 2018-02-01 | 孙秋实 | Grouping revocation method and grouping system |
| CN108461152A (en) * | 2018-01-12 | 2018-08-28 | 平安科技（深圳）有限公司 | Medical model training method, medical recognition methods, device, equipment and medium |
| WO2019056499A1 (en) * | 2017-09-20 | 2019-03-28 | 平安科技（深圳）有限公司 | Prediction model training method, data monitoring method, apparatuses, device and medium |
| CN111786802A (en) * | 2019-04-03 | 2020-10-16 | 北京嘀嘀无限科技发展有限公司 | Event detection method and device |
| WO2020228655A1 (en) * | 2019-05-10 | 2020-11-19 | 腾讯科技（深圳）有限公司 | Method, apparatus, electronic device, and computer storage medium for optimizing quantization model |
| CN113574327A (en) * | 2019-03-15 | 2021-10-29 | 3M创新有限公司 | Determining causal models for a control environment |
-
2014
- 2014-12-19 CN CN201410805857.0A patent/CN105786830A/en active Pending
Cited By (12)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2018018520A1 (en) * | 2016-07-28 | 2018-02-01 | 孙秋实 | Grouping revocation method and grouping system |
| WO2019056499A1 (en) * | 2017-09-20 | 2019-03-28 | 平安科技（深圳）有限公司 | Prediction model training method, data monitoring method, apparatuses, device and medium |
| CN108461152A (en) * | 2018-01-12 | 2018-08-28 | 平安科技（深圳）有限公司 | Medical model training method, medical recognition methods, device, equipment and medium |
| WO2019136806A1 (en) * | 2018-01-12 | 2019-07-18 | 平安科技（深圳）有限公司 | Medical model training method and apparatus, medical identification method and apparatus, device, and medium |
| CN113574327A (en) * | 2019-03-15 | 2021-10-29 | 3M创新有限公司 | Determining causal models for a control environment |
| CN113574327B (en) * | 2019-03-15 | 2023-01-10 | 3M创新有限公司 | Method and system for controlling an environment by selecting a control setting |
| US11853018B2 (en) | 2019-03-15 | 2023-12-26 | 3M Innovative Properties Company | Determining causal models for controlling environments |
| CN111786802A (en) * | 2019-04-03 | 2020-10-16 | 北京嘀嘀无限科技发展有限公司 | Event detection method and device |
| CN111786802B (en) * | 2019-04-03 | 2023-07-04 | 北京嘀嘀无限科技发展有限公司 | Event detection method and device |
| WO2020228655A1 (en) * | 2019-05-10 | 2020-11-19 | 腾讯科技（深圳）有限公司 | Method, apparatus, electronic device, and computer storage medium for optimizing quantization model |
| JP2022531641A (en) * | 2019-05-10 | 2022-07-08 | テンセント・テクノロジー・（シェンジェン）・カンパニー・リミテッド | Quantization model optimization method, device, information recommendation method, device, neural network model optimization method, device, electronic device and computer program |
| JP7194284B2 (en) | 2019-05-10 | 2022-12-21 | テンセント・テクノロジー・（シェンジェン）・カンパニー・リミテッド | Quantization model optimization method, device, information recommendation method, device, neural network model optimization method, device, electronic device, and computer program |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN105786830A (en) | Method, device and system for self-adaptively adjusting models in computer systems | |
| US10248528B2 (en) | System monitoring method and apparatus | |
| US9002774B2 (en) | Systems and methods for generating a forecasting model and forecasting future values | |
| CN103024762B (en) | Service feature based communication service forecasting method | |
| US7765123B2 (en) | Indicating which of forecasting models at different aggregation levels has a better forecast quality | |
| WO2019105226A1 (en) | Method and device used for predicting effect of marketing activity, and electronic device | |
| CN111162925B (en) | Network flow prediction method and device, electronic equipment and storage medium | |
| EP1565856A2 (en) | Portfolio rebalancing by means of resampled efficient frontiers | |
| CN109583729B (en) | Data processing method and device for platform online model | |
| Santos et al. | Forecasting value-at-risk with a duration-based POT method | |
| Miguéis et al. | Enhanced decision support in credit scoring using Bayesian binary quantile regression | |
| CN103617146B (en) | A kind of machine learning method and device based on hardware resource consumption | |
| CN110880127A (en) | Consumption level prediction method and device, electronic equipment and storage medium | |
| Lizares et al. | Corporate financial distress: The case of publicly listed firms in an emerging market economy | |
| CN111340606A (en) | Full-process income auditing method and device | |
| CN110866698A (en) | Device for assessing service score of service provider | |
| CN112488496A (en) | Financial index prediction method and device | |
| CN111524001A (en) | Method and device for predicting user credit line and related equipment | |
| CN112398670A (en) | Flow prediction method, flow prediction device, computing equipment and computer storage medium | |
| CN114022221A (en) | Sales prediction method, acquisition method and device of model thereof, and electronic equipment | |
| CN111784173B (en) | AB experiment data processing method, device, server and medium | |
| CN113283673A (en) | Model performance attenuation evaluation method, model training method and device | |
| Du et al. | Backtesting portfolio value‐at‐risk with estimated portfolio weights | |
| Volk-Makarewicz et al. | Assessing the impact of jumps in an option pricing model: A gradient estimation approach | |
| CN107480165B (en) | A kind of method and apparatus of page ranking |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| RJ01 | Rejection of invention patent application after publication | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20160720 |