CN110909878A - Training method and device of neural network model for estimating resource usage share - Google Patents
Training method and device of neural network model for estimating resource usage share Download PDFInfo
- Publication number
- CN110909878A CN110909878A CN201911214674.0A CN201911214674A CN110909878A CN 110909878 A CN110909878 A CN 110909878A CN 201911214674 A CN201911214674 A CN 201911214674A CN 110909878 A CN110909878 A CN 110909878A
- Authority
- CN
- China
- Prior art keywords
- resource usage
- user
- label
- time period
- share
- 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
Images
Classifications
-
- 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
- G06N3/08—Learning methods
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F17/00—Digital computing or data processing equipment or methods, specially adapted for specific functions
- G06F17/10—Complex mathematical operations
- G06F17/18—Complex mathematical operations for evaluating statistical data, e.g. average values, frequency distributions, probability functions, regression analysis
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N20/00—Machine learning
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q10/00—Administration; Management
- G06Q10/06—Resources, workflows, human or project management; Enterprise or organisation planning; Enterprise or organisation modelling
- G06Q10/063—Operations research, analysis or management
- G06Q10/0631—Resource planning, allocation, distributing or scheduling for enterprises or organisations
-
- 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
- G06Q40/12—Accounting
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Business, Economics & Management (AREA)
- Theoretical Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- Mathematical Physics (AREA)
- Human Resources & Organizations (AREA)
- Software Systems (AREA)
- Strategic Management (AREA)
- Economics (AREA)
- General Engineering & Computer Science (AREA)
- Development Economics (AREA)
- Computing Systems (AREA)
- Finance (AREA)
- Accounting & Taxation (AREA)
- General Business, Economics & Management (AREA)
- Operations Research (AREA)
- Mathematical Optimization (AREA)
- Mathematical Analysis (AREA)
- Marketing (AREA)
- Artificial Intelligence (AREA)
- Life Sciences & Earth Sciences (AREA)
- Entrepreneurship & Innovation (AREA)
- Computational Mathematics (AREA)
- Evolutionary Computation (AREA)
- Pure & Applied Mathematics (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Biophysics (AREA)
- General Health & Medical Sciences (AREA)
- Medical Informatics (AREA)
- Evolutionary Biology (AREA)
- Computational Linguistics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Educational Administration (AREA)
- Molecular Biology (AREA)
- Game Theory and Decision Science (AREA)
- Biomedical Technology (AREA)
- Health & Medical Sciences (AREA)
Abstract
The embodiment of the specification provides a method and a device for training a neural network model for estimating resource usage share, wherein the method comprises the following steps: acquiring a plurality of groups of training samples, wherein the training samples comprise characteristic information of a user, a first label and a second label, the first label is used for indicating the resource usage share of the user in a first time period, and the second label is used for indicating whether the user uses resources in the first time period; aiming at the characteristic information of users in each group of training samples, a resource usage share pre-estimated value and a resource usage probability pre-estimated value are obtained by utilizing a neural network model; determining a first loss according to the resource usage share estimated value, the first label and a first loss function; determining a second loss according to the resource use probability estimated value, the second label and a second loss function; weighting and summing the first loss and the second loss to obtain a total loss; the neural network model is trained with the goal of minimizing the total loss. The accuracy of estimating the resource use share can be improved.
Description
Technical Field
One or more embodiments of the present disclosure relate to the field of computers, and more particularly, to a method and apparatus for training a neural network model for estimating resource usage.
Background
Currently, with the continuous development of artificial intelligence, the resource usage share of a user can be estimated through a machine learning technology, so that the resource allocation, regulation and control of the user and other matters are facilitated. For example, in the application scenario of internet finance, machine learning is increasingly playing a role in tasks such as wind control, money forecast, and the like. For the estimation of the amount, for financial consumption scenes such as bank or network loan and the like, the payment amount condition of the user in the next period of time is reasonably and accurately estimated, so that reasonable fund storage and use of an organization are facilitated, and more reasonable use of the fund is achieved.
In the prior art, when the machine learning technology is applied to the task of estimating the resource usage share, the task is usually solved by formalizing the task into a typical regression task. However, the accuracy of estimating the resource usage is not high.
Accordingly, improved solutions are desired that improve the accuracy of machine learning for predicting resource usage.
Disclosure of Invention
One or more embodiments of the present specification describe a method and an apparatus for training a neural network model for estimating resource usage, which can improve the accuracy of machine learning for estimating resource usage.
In a first aspect, a method for training a neural network model for estimating resource usage shares is provided, and the method includes:
acquiring a plurality of groups of training samples, wherein the training samples comprise characteristic information of a user, and a first label and a second label corresponding to the characteristic information, the first label is used for indicating a resource usage share of the user in a first time period, and the second label is used for indicating whether the user uses resources in the first time period;
aiming at the characteristic information of users in each group of training samples, a resource usage share pre-estimated value and a resource usage probability pre-estimated value are obtained by utilizing a neural network model;
determining a first loss according to the resource usage share estimated value, the first label and a first loss function; determining a second loss according to the resource usage probability estimated value, the second label and a second loss function; obtaining a total loss by performing weighted summation on the first loss and the second loss;
training the neural network model with the goal of minimizing the total loss.
In one possible embodiment, the obtaining of the plurality of sets of training samples includes:
acquiring a plurality of groups of original training data, wherein the original training data comprise characteristic information of a user and resource usage share of the user in the first time period;
performing a first operation on the resource usage share of the user in the first time period to obtain the first label, wherein the numerical value of the first label is smaller than the numerical value of the corresponding resource usage share;
and obtaining the second label according to whether the resource usage share of the user in the first time period is 0.
Further, the first operation includes: and (4) carrying out logarithmic operation.
In one possible embodiment, the first loss function is a squared loss function.
In one possible implementation, the second loss function is a cross-entropy loss function.
In one possible embodiment, after the training of the neural network model, the method further includes:
inputting feature information of a target user into a trained neural network model, and outputting a first predicted value of the target user for a resource usage share in the first time period and a second predicted value of the target user for a resource usage probability in the first time period through the neural network model;
and comprehensively determining the target resource usage share of the target user in the first time period according to the first predicted value and the second predicted value.
Further, the integrating determining the target user's target resource usage share during the first time period comprises:
performing second operation on the first predicted value to obtain a first resource usage share;
and adjusting the first resource usage share according to the second predicted value to obtain a target resource usage share of the target user in the first time period.
In one possible embodiment, the first time period is from the beginning of the user qualifying for the use of the resource to the end of a preset time point.
Further, the preset time point includes any one of:
the last day of the month, the last day of the quarter, and the last day of the year.
In a second aspect, an apparatus for training a neural network model to predict resource usage shares is provided, the apparatus comprising:
an obtaining unit, configured to obtain multiple sets of training samples, where each training sample includes feature information of a user, and a first label and a second label corresponding to the feature information, where the first label is used to indicate a resource usage share of the user in a first time period, and the second label is used to indicate whether the user uses a resource in the first time period;
the estimating unit is used for obtaining a resource usage share estimated value and a resource usage probability estimated value by utilizing a neural network model according to the characteristic information of the users in each group of training samples obtained by the obtaining unit;
a determining unit, configured to determine a first loss according to the resource usage share estimated value obtained by the estimating unit, the first label, and a first loss function; determining a second loss according to the resource utilization probability estimated value obtained by the estimating unit, the second label and a second loss function; obtaining a total loss by performing weighted summation on the first loss and the second loss;
and the training unit is used for training the neural network model by taking the total loss determined by the determining unit as a target.
In a third aspect, there is provided a computer readable storage medium having stored thereon a computer program which, when executed in a computer, causes the computer to perform the method of the first aspect.
In a fourth aspect, there is provided a computing device comprising a memory having stored therein executable code and a processor that, when executing the executable code, implements the method of the first aspect.
By the method and the device provided by the embodiment of the specification, firstly, a plurality of groups of training samples are obtained, wherein each training sample comprises the characteristic information of a user, and a first label and a second label corresponding to the characteristic information, the first label is used for indicating the resource usage share of the user in a first time period, and the second label is used for indicating whether the user uses resources in the first time period; then, aiming at the characteristic information of the users in each group of training samples, a resource usage share pre-estimated value and a resource usage probability pre-estimated value are obtained by utilizing a neural network model; then, determining a first loss according to the resource usage share estimated value, the first label and a first loss function; determining a second loss according to the resource usage probability estimated value, the second label and a second loss function; obtaining a total loss by performing weighted summation on the first loss and the second loss; and finally, training the neural network model by taking the minimization of the total loss as a target. As can be seen from the above, in the embodiment of the present specification, the task of estimating the resource usage share is split into two subtasks, namely, the "estimated resource usage probability" and the "estimated resource usage share", the two subtasks are simultaneously learned by using a multi-task learning method, and the output of the two subtasks is finally considered as the finally estimated resource usage share, so that the accuracy of machine learning for estimating the resource usage share can be improved, and the estimation of the resource usage probability of the user is also beneficial to the matters of allocation regulation and control of the user, and the like.
Drawings
In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings needed to be used in the description of the embodiments are briefly introduced below, and it is obvious that the drawings in the following description are only some embodiments of the present invention, and it is obvious for those skilled in the art to obtain other drawings based on these drawings without creative efforts.
FIG. 1 is a schematic diagram illustrating an implementation scenario of an embodiment disclosed herein;
FIG. 2 illustrates a flow diagram of a method for training a neural network model to predict resource usage shares, according to one embodiment;
FIG. 3 illustrates a schematic structural diagram of a neural network model for predicting resource usage shares, according to one embodiment;
FIG. 4 shows a schematic block diagram of a training apparatus for a neural network model to predict resource usage shares, according to one embodiment.
Detailed Description
The scheme provided by the specification is described below with reference to the accompanying drawings.
Fig. 1 is a schematic view of an implementation scenario of an embodiment disclosed in this specification. The implementation scenario involves training of a neural network model that predicts resource usage shares. Referring to fig. 1, in this implementation scenario, there may be multiple resource providers providing resources for users, where a specific resource provider provides a resource to a target user, or how the resource provider reserves and uses the resource may be determined according to the estimated resource usage share of each user. For example, when the resource usage share of the estimated user belongs to a first interval range, the resource provider 1 provides the resource for the user; when the estimated resource usage share of the user belongs to the second interval range, the resource provider 2 provides the resource for the user; when the resource usage share of the estimated user belongs to the third interval range, the resource provider 3 provides the resource for the user. It can be understood that the accurate estimation of the resource usage share is beneficial to resource allocation and resource regulation of the user, and many possible scenarios exist, and is not limited to selecting a resource provider that provides resources to the user.
The resources can be currency resources, computer processing resources, computer storage resources, network bandwidth resources, virtual article resources in games, and the like.
In the embodiment of the present specification, only the money resource is taken as an example for specific description, and the user payment estimation is taken as an example to estimate the payment amount of the user in the next period (for example, one month) from the perspective of machine learning through the basic information and the historical information of the user, and for an organization such as a bank, corresponding funds can be prepared based on the estimation result to obtain reasonable use of the funds.
Fig. 2 shows a flowchart of a method for training a neural network model for predicting resource usage shares according to an embodiment, which may be based on the implementation scenario shown in fig. 1. As shown in fig. 2, the training method of the neural network model for estimating resource usage in this embodiment includes the following steps: step 21, obtaining a plurality of sets of training samples, where each training sample includes feature information of a user, and a first label and a second label corresponding to the feature information, where the first label is used to indicate a resource usage share of the user in a first time period, and the second label is used to indicate whether the user uses a resource in the first time period; step 22, aiming at the characteristic information of the users in each group of training samples, obtaining a resource usage share pre-estimated value and a resource usage probability pre-estimated value by utilizing a neural network model; step 23, determining a first loss according to the resource usage share estimated value, the first label and a first loss function; determining a second loss according to the resource usage probability estimated value, the second label and a second loss function; obtaining a total loss by performing weighted summation on the first loss and the second loss; and 24, training the neural network model by taking the total loss as a target to be minimized. Specific execution modes of the above steps are described below.
First, in step 21, a plurality of sets of training samples are obtained, where each training sample includes feature information of a user, and a first label and a second label corresponding to the feature information, where the first label is used to indicate a resource usage share of the user in a first time period, and the second label is used to indicate whether the user uses a resource in the first time period. It is understood that when the resource usage share of the user in the first time period is 0, it means that the user does not use the resource in the first time period.
In one example, a plurality of sets of original training data are obtained, wherein the original training data comprise characteristic information of a user and resource usage share of the user in the first time period; then, performing a first operation on the resource usage share of the user in the first time period to obtain the first label, wherein the numerical value of the first label is smaller than the numerical value of the corresponding resource usage share; and obtaining the second label according to whether the resource usage share of the user in the first time period is 0.
Further, the first operation includes: and (4) carrying out logarithmic operation.
In one example, the first time period is from the beginning of the user qualifying for resource usage to the end of a preset time point.
Further, the preset time point includes any one of:
the last day of the month, the last day of the quarter, and the last day of the year.
For example, the process of obtaining a new sample representation from the original training data, i.e. the initial sample, may comprise: giving an initial sample (x, y), wherein x represents the characteristic information of a user, the characteristic information is preprocessed, y represents the sum of payment amounts of the user passing through the end of the month, the initial sample is preprocessed, and a new sample representation is obtained
Wherein, the sum of the payment sum is logarithmized,thereby reducing the scale thereof to obtain
Obtaining information whether the user supports, if y > 0
If y is 0
It will be appreciated that the above-described,
in correspondence with the aforementioned first label,
corresponding to the aforementioned second label.
Then, in step 22, for the feature information of the users in each training sample set, a resource usage share pre-estimated value and a resource usage probability pre-estimated value are obtained by using a neural network model. It is understood that training of the neural network model is in supervised learning, and in multitask learning.
Wherein, the supervised learning: one area of research in machine learning is predicting new test samples by learning a machine learning model from given data samples, and it is worth pointing out that samples for supervised learning are represented as feature vectors describing their features and labeled information representing their attributes. In the embodiment of the present specification, the label information is the aforementioned first label and second label.
Multi-task learning: in a research field of machine learning, a plurality of tasks are learned simultaneously, and the performance of each task is improved by using useful information (data and models) of other tasks, so that a better effect is expected compared with a model trained by single-task data. Typically, given m learning tasks, where all or a portion of the tasks are related but not identical, the goal of multi-task learning is to help improve the performance of the individual tasks by using the knowledge contained in the m tasks. In the embodiment of the specification, two tasks of predicting resource usage share and predicting resource usage probability are included.
Then, in step 23, determining a first loss according to the resource usage share pre-estimated value, the first label, and a first loss function; determining a second loss according to the resource usage probability estimated value, the second label and a second loss function; and obtaining the total loss by carrying out weighted summation on the first loss and the second loss. It will be appreciated that the first and second losses have different effects on the model training, and therefore the total loss is obtained by means of a weighted sum.
In one example, the first loss function is a squared loss function.
In one example, the second loss function is a cross entropy loss (cross entropy loss) function.
FIG. 3 illustrates a schematic structural diagram of a neural network model for predicting resource usage shares, according to one embodiment. Referring to fig. 3, the neural network model includes a sharing layer, a first prediction layer, and a second prediction layer. The sharing layer can be, but is not limited to, two layers and is used for receiving characteristic information of a user; the first prediction layer is used for predicting the resource usage share; the second prediction layer is used for predicting the resource utilization probability.
And respectively determining respective losses for a first estimation layer and a second estimation layer, wherein for the first estimation layer, since the label is a real value obtained by logarithm of the real resource share, the loss can be determined by adopting a square loss function and is marked as L1, and for the second estimation layer, since the label is marked by 0 and 1, the loss can be determined by adopting a cross entropy loss function and is marked as L2. The total loss is the weighted average of the two, i.e., L ═ w1 × L1+ w2 × L2, w1 and w2 are hyper-parameters that need to be optimized.
Finally, at step 24, the neural network model is trained with the goal of minimizing the total loss. It will be appreciated that training the model involves adjusting the parameters of the aforementioned shared layer.
In one example, after step 24, further comprising:
inputting feature information of a target user into a trained neural network model, and outputting a first predicted value of the target user for a resource usage share in the first time period and a second predicted value of the target user for a resource usage probability in the first time period through the neural network model;
and comprehensively determining the target resource usage share of the target user in the first time period according to the first predicted value and the second predicted value.
Further, performing a second operation on the first predicted value to obtain a first resource usage share; and adjusting the first resource usage share according to the second predicted value to obtain a target resource usage share of the target user in the first time period.
It will be appreciated that the second operation is an inverse of the first operation, for example the first operation comprises a logarithmic operation and correspondingly the second operation comprises an exponential operation.
For example, for sample input, the model predicts its probability of branching
And amount of money spent
(taking the value after logarithm), introducing the payment probability value into the final estimation of the payment amount value of the user, namely calculating the payment amount of the user as
The method provided by the embodiment of the specification takes the learning of the user resource usage share (such as the fund payment amount) as an important learning task, and also takes the learning of the user resource usage probability (such as the fund payment probability) as another important learning task, and since the two tasks are strongly related tasks, the two tasks are simultaneously learned, so that the model achieves better practical effect. In addition, in a scene with more new users, the estimation of the resource use probability is introduced, so that the final prediction of the resource use share has more reference information, and the whole prediction is more accurate.
According to another aspect of embodiments, a device for training a neural network model of estimated resource usage is also provided, and the device is used for executing the method for training the neural network model of estimated resource usage provided by the embodiments of the present specification. FIG. 4 shows a schematic block diagram of a training apparatus for a neural network model to predict resource usage shares, according to one embodiment. As shown in fig. 4, the apparatus 400 includes:
an obtaining unit 41, configured to obtain multiple sets of training samples, where each training sample includes feature information of a user, and a first label and a second label corresponding to the feature information, where the first label is used to indicate a resource usage share of the user in a first time period, and the second label is used to indicate whether the user uses a resource in the first time period;
the estimating unit 42 is configured to obtain a resource usage share estimated value and a resource usage probability estimated value by using a neural network model for the feature information of the user in each group of training samples acquired by the acquiring unit 41;
a determining unit 43, configured to determine a first loss according to the resource usage share estimated value obtained by the estimating unit 42, the first label, and a first loss function; determining a second loss according to the resource usage probability estimated value obtained by the estimating unit 42, the second label, and a second loss function; obtaining a total loss by performing weighted summation on the first loss and the second loss;
a training unit 44, configured to train the neural network model with a goal of minimizing the total loss determined by the determining unit 43.
Optionally, as an embodiment, the obtaining unit 41 is specifically configured to:
acquiring a plurality of groups of original training data, wherein the original training data comprise characteristic information of a user and resource usage share of the user in the first time period;
performing a first operation on the resource usage share of the user in the first time period to obtain the first label, wherein the numerical value of the first label is smaller than the numerical value of the corresponding resource usage share;
and obtaining the second label according to whether the resource usage share of the user in the first time period is 0.
Further, the first operation includes: and (4) carrying out logarithmic operation.
Optionally, as an embodiment, the first loss function is a square loss function.
Optionally, as an embodiment, the second loss function is a cross-entropy loss function.
Optionally, as an embodiment, the apparatus further includes:
a prediction unit, configured to, after the training unit 44 trains the neural network model, input feature information of a target user into the trained neural network model, output a first prediction value of the target user for a resource usage share in the first time period and a second prediction value of the target user for a resource usage probability in the first time period through the neural network model;
and the comprehensive unit is used for comprehensively determining the target resource usage share of the target user in the first time period according to the first predicted value and the second predicted value obtained by the prediction unit.
Further, the synthesis unit is specifically configured to:
performing second operation on the first predicted value to obtain a first resource usage share;
and adjusting the first resource usage share according to the second predicted value to obtain a target resource usage share of the target user in the first time period.
Optionally, as an embodiment, the first time period is from the beginning when the user qualifies for the resource to the end of a preset time point.
Further, the preset time point includes any one of:
the last day of the month, the last day of the quarter, and the last day of the year.
With the apparatus provided in this specification, first, the obtaining unit 41 obtains a plurality of sets of training samples, where the training samples include feature information of a user, and a first label and a second label corresponding to the feature information, where the first label is used to indicate a resource usage share of the user in a first time period, and the second label is used to indicate whether the user uses a resource in the first time period; then, the pre-estimation unit 42 obtains a resource usage share pre-estimation value and a resource usage probability pre-estimation value by using a neural network model according to the characteristic information of the users in each group of training samples; then, the determining unit 43 determines a first loss according to the resource usage share estimation value, the first label, and a first loss function; determining a second loss according to the resource usage probability estimated value, the second label and a second loss function; obtaining a total loss by performing weighted summation on the first loss and the second loss; finally, a training unit 44 trains the neural network model with the goal of minimizing the total loss. As can be seen from the above, in the embodiment of the present specification, the task of estimating the resource usage share is split into two subtasks, namely, the "estimated resource usage probability" and the "estimated resource usage share", the two subtasks are simultaneously learned by using a multi-task learning method, and the output of the two subtasks is finally considered as the finally estimated resource usage share, so that the accuracy of machine learning for estimating the resource usage share can be improved, and the estimation of the resource usage probability of the user is also beneficial to the matters of allocation regulation and control of the user, and the like.
According to an embodiment of another aspect, there is also provided a computer-readable storage medium having stored thereon a computer program which, when executed in a computer, causes the computer to perform the method described in connection with fig. 2.
According to an embodiment of yet another aspect, there is also provided a computing device comprising a memory having stored therein executable code, and a processor that, when executing the executable code, implements the method described in connection with fig. 2.
Those skilled in the art will recognize that, in one or more of the examples described above, the functions described in this invention may be implemented in hardware, software, firmware, or any combination thereof. When implemented in software, the functions may be stored on or transmitted over as one or more instructions or code on a computer-readable medium.
The above-mentioned embodiments, objects, technical solutions and advantages of the present invention are further described in detail, it should be understood that the above-mentioned embodiments are only exemplary embodiments of the present invention, and are not intended to limit the scope of the present invention, and any modifications, equivalent substitutions, improvements and the like made on the basis of the technical solutions of the present invention should be included in the scope of the present invention.
Claims (20)
1. A method of training a neural network model to predict resource usage shares, the method comprising:
acquiring a plurality of groups of training samples, wherein the training samples comprise characteristic information of a user, and a first label and a second label corresponding to the characteristic information, the first label is used for indicating a resource usage share of the user in a first time period, and the second label is used for indicating whether the user uses resources in the first time period;
aiming at the characteristic information of users in each group of training samples, a resource usage share pre-estimated value and a resource usage probability pre-estimated value are obtained by utilizing a neural network model;
determining a first loss according to the resource usage share estimated value, the first label and a first loss function; determining a second loss according to the resource usage probability estimated value, the second label and a second loss function; obtaining a total loss by performing weighted summation on the first loss and the second loss;
training the neural network model with the goal of minimizing the total loss.
2. The method of claim 1, wherein the obtaining a plurality of sets of training samples comprises:
acquiring a plurality of groups of original training data, wherein the original training data comprise characteristic information of a user and resource usage share of the user in the first time period;
performing a first operation on the resource usage share of the user in the first time period to obtain the first label, wherein the numerical value of the first label is smaller than the numerical value of the corresponding resource usage share;
and obtaining the second label according to whether the resource usage share of the user in the first time period is 0.
3. The method of claim 2, wherein the first operation comprises: and (4) carrying out logarithmic operation.
4. The method of claim 1, wherein the first loss function is a squared loss function.
5. The method of claim 1, wherein the second loss function is a cross-entropy loss function.
6. The method of claim 1, wherein after the training of the neural network model, the method further comprises:
inputting feature information of a target user into a trained neural network model, and outputting a first predicted value of the target user for a resource usage share in the first time period and a second predicted value of the target user for a resource usage probability in the first time period through the neural network model;
and comprehensively determining the target resource usage share of the target user in the first time period according to the first predicted value and the second predicted value.
7. The method of claim 6, wherein the synthesizing determines a target resource usage share of the target user over the first time period, comprising:
performing second operation on the first predicted value to obtain a first resource usage share;
and adjusting the first resource usage share according to the second predicted value to obtain a target resource usage share of the target user in the first time period.
8. The method of claim 1, wherein the first time period is from a start when a user qualifies for resource usage to an end of a preset time point.
9. The method of claim 8, wherein the preset time point comprises any one of:
the last day of the month, the last day of the quarter, and the last day of the year.
10. An apparatus for training a neural network model to predict resource usage shares, the apparatus comprising:
an obtaining unit, configured to obtain multiple sets of training samples, where each training sample includes feature information of a user, and a first label and a second label corresponding to the feature information, where the first label is used to indicate a resource usage share of the user in a first time period, and the second label is used to indicate whether the user uses a resource in the first time period;
the estimating unit is used for obtaining a resource usage share estimated value and a resource usage probability estimated value by utilizing a neural network model according to the characteristic information of the users in each group of training samples obtained by the obtaining unit;
a determining unit, configured to determine a first loss according to the resource usage share estimated value obtained by the estimating unit, the first label, and a first loss function; determining a second loss according to the resource utilization probability estimated value obtained by the estimating unit, the second label and a second loss function; obtaining a total loss by performing weighted summation on the first loss and the second loss;
and the training unit is used for training the neural network model by taking the total loss determined by the determining unit as a target.
11. The apparatus of claim 10, wherein the obtaining unit is specifically configured to:
acquiring a plurality of groups of original training data, wherein the original training data comprise characteristic information of a user and resource usage share of the user in the first time period;
performing a first operation on the resource usage share of the user in the first time period to obtain the first label, wherein the numerical value of the first label is smaller than the numerical value of the corresponding resource usage share;
and obtaining the second label according to whether the resource usage share of the user in the first time period is 0.
12. The apparatus of claim 11, wherein the first operation comprises: and (4) carrying out logarithmic operation.
13. The apparatus of claim 10, wherein the first loss function is a squared loss function.
14. The apparatus of claim 10, wherein the second loss function is a cross-entropy loss function.
15. The apparatus of claim 10, wherein the apparatus further comprises:
the prediction unit is used for inputting the characteristic information of the target user into the trained neural network model after the training unit trains the neural network model, and outputting a first predicted value of the target user for the resource usage share in the first time period and a second predicted value of the target user for the resource usage probability in the first time period through the neural network model;
and the comprehensive unit is used for comprehensively determining the target resource usage share of the target user in the first time period according to the first predicted value and the second predicted value obtained by the prediction unit.
16. The apparatus of claim 15, wherein the synthesis unit is specifically configured to:
performing second operation on the first predicted value to obtain a first resource usage share;
and adjusting the first resource usage share according to the second predicted value to obtain a target resource usage share of the target user in the first time period.
17. The apparatus of claim 10, wherein the first time period is from a start of a user qualifying for resource usage to an end of a preset time point.
18. The apparatus of claim 17, wherein the preset time point comprises any one of:
the last day of the month, the last day of the quarter, and the last day of the year.
19. A computer-readable storage medium, on which a computer program is stored which, when executed in a computer, causes the computer to carry out the method of any one of claims 1-9.
20. A computing device comprising a memory having stored therein executable code and a processor that, when executing the executable code, implements the method of any of claims 1-9.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201911214674.0A CN110909878A (en) | 2019-12-02 | 2019-12-02 | Training method and device of neural network model for estimating resource usage share |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201911214674.0A CN110909878A (en) | 2019-12-02 | 2019-12-02 | Training method and device of neural network model for estimating resource usage share |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN110909878A true CN110909878A (en) | 2020-03-24 |
Family
ID=69821645
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201911214674.0A Pending CN110909878A (en) | 2019-12-02 | 2019-12-02 | Training method and device of neural network model for estimating resource usage share |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN110909878A (en) |
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112288436A (en) * | 2020-10-27 | 2021-01-29 | 上海淇馥信息技术有限公司 | Triggered resource quota adjusting method, device and system |
| CN113988254A (en) * | 2020-07-27 | 2022-01-28 | 腾讯科技（深圳）有限公司 | Method and device for determining neural network model for multiple environments |
| CN113988291A (en) * | 2021-10-26 | 2022-01-28 | 支付宝(杭州)信息技术有限公司 | Training method and device for user representation network |
| US11438373B2 (en) * | 2020-01-09 | 2022-09-06 | Cymulate Ltd. | Monitoring for security threats from lateral movements |
-
2019
- 2019-12-02 CN CN201911214674.0A patent/CN110909878A/en active Pending
Cited By (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US11438373B2 (en) * | 2020-01-09 | 2022-09-06 | Cymulate Ltd. | Monitoring for security threats from lateral movements |
| US11818168B2 (en) | 2020-01-09 | 2023-11-14 | Cymulate Ltd. | Monitoring for lateral movements-related security threats |
| CN113988254A (en) * | 2020-07-27 | 2022-01-28 | 腾讯科技（深圳）有限公司 | Method and device for determining neural network model for multiple environments |
| CN113988254B (en) * | 2020-07-27 | 2023-07-14 | 腾讯科技（深圳）有限公司 | Method and device for determining neural network model for multiple environments |
| CN112288436A (en) * | 2020-10-27 | 2021-01-29 | 上海淇馥信息技术有限公司 | Triggered resource quota adjusting method, device and system |
| CN113988291A (en) * | 2021-10-26 | 2022-01-28 | 支付宝(杭州)信息技术有限公司 | Training method and device for user representation network |
| CN113988291B (en) * | 2021-10-26 | 2024-06-04 | 支付宝(杭州)信息技术有限公司 | Training method and device for user characterization network |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110909878A (en) | Training method and device of neural network model for estimating resource usage share | |
| EP3446260B1 (en) | Memory-efficient backpropagation through time | |
| US11861474B2 (en) | Dynamic placement of computation sub-graphs | |
| US11586919B2 (en) | Task-oriented machine learning and a configurable tool thereof on a computing environment | |
| EP3593289A1 (en) | Training action selection neural networks using a differentiable credit function | |
| CN110520871A (en) | Training machine learning model | |
| CN111340244B (en) | Prediction method, training method, device, server and medium | |
| CN113064700A (en) | Container deployment method and device | |
| CN108898221B (en) | Joint learning method of characteristics and strategies based on state characteristics and subsequent characteristics | |
| US11620493B2 (en) | Intelligent selection of time series models | |
| Chiorrini et al. | A preliminary study on the application of reinforcement learning for predictive process monitoring | |
| CN113256335B (en) | Data screening method, multimedia data delivery effect prediction method and device | |
| Lin et al. | A deep reinforcement learning framework for optimal trade execution | |
| US11880765B2 (en) | State-augmented reinforcement learning | |
| CN114742644A (en) | Method and device for training multi-scene wind control system and predicting business object risk | |
| CN113592263A (en) | Resource return increment prediction method and device based on dynamic resource return increase ratio | |
| Giannikis et al. | Modelling nonlinearities and heavy tails via threshold normal mixture GARCH models | |
| CN116820730B (en) | Task scheduling method, device and storage medium of multi-engine computing system | |
| Candar et al. | Equity Portfolio Optimization Using Reinforcement Learning: Emerging Market Case | |
| Wicaksono et al. | Prediction Land Value Using Geographically Weighted Extreme Learning Machine | |
| CN112288123B (en) | Resource attribute prediction method and system, training method and related equipment | |
| Na et al. | An adaptive replanning mechanism for dependable service-based systems | |
| Boskovic et al. | Helping the Oracle: Vector Sign Constraints for Model Shrinkage Methodologies | |
| Lind Amigo et al. | Integrating Customer Behavior Analysisfor Cost Prediction and ResourceUtilization in Mobile Networks: A Machine Learning Approach to Azure Server Analysis | |
| CN118057371A (en) | Method, apparatus, electronic device and storage medium for managing multitasking model |
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 | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20200324 |
|
| RJ01 | Rejection of invention patent application after publication |