CN109033309A - Data processing method, device, equipment and storage medium - Google Patents
Data processing method, device, equipment and storage medium Download PDFInfo
- Publication number
- CN109033309A CN109033309A CN201810783681.1A CN201810783681A CN109033309A CN 109033309 A CN109033309 A CN 109033309A CN 201810783681 A CN201810783681 A CN 201810783681A CN 109033309 A CN109033309 A CN 109033309A
- Authority
- CN
- China
- Prior art keywords
- characteristic
- dimension
- matrix
- learning model
- deep learning
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
Landscapes
- Image Analysis (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The embodiment of the present application provides a kind of data processing method, device, equipment and storage medium.This method comprises: obtaining the sample data of deep learning model；For the characteristic of each dimension in the characteristic of at least two dimensions, standardization processing is carried out respectively, obtains each dimension treated characteristic；By at least two dimensions, treated that characteristic is handled, and obtains the input data of deep learning model.In the embodiment of the present application, standardization processing is carried out by the characteristic of dimension each in the sample data to deep learning model, so that the difference between the characteristic of each dimension in above-mentioned sample data reduces as much as possible, when training deep learning model subsequently through the characteristic after above-mentioned standardization processing, the convergent speed of deep learning model can be made to get a promotion, to improve the training effectiveness and precision of deep learning model.
Description
Technical field
The invention relates to machine learning techniques field, in particular to a kind of data processing method, device, equipment and
Storage medium.
Background technique
Deep learning is a kind of based on the method for carrying out representative learning to data in machine learning.In the reality of deep learning
In, it usually needs train deep learning model in advance, then carry out subsequent prediction work by the deep learning model
Make.
Sample data used by training deep learning model generally includes the characteristic of multiple dimensions.Illustratively,
A certain deep learning model is used to carry out credit assessment to the user of application credit card, used by training the deep learning model
Sample data may include the age of applicant, gender, emolument, held credit card quantity, bank account balances, wed no, children
The characteristic of multiple dimensions such as number, the quantity that owns a house.In the training process, first to the characteristic of above-mentioned each dimension
It is integrated, the data obtained after integration is input to deep learning model later and are trained.
Since the characteristic of different dimensions differs greatly, mould is carried out using the characteristic of above-mentioned different dimensions
When type training, the convergence rate of model is relatively slow even to be difficult to restrain, cause the efficiency for training deep learning model and accuracy rate compared with
It is low.
Summary of the invention
The embodiment of the present application provides a kind of data processing method, device, equipment and storage medium, can be used for solving correlation
The lower problem of the efficiency and accuracy rate of training deep learning model in technology.
On the one hand, the embodiment of the present application provides a kind of data processing method, which comprises
The sample data of deep learning model is obtained, the sample data includes the characteristic of at least two dimensions；
For the characteristic of each dimension in the characteristic of at least two dimension, carry out at standardization respectively
Reason obtains each dimension treated characteristic；Wherein, the standardization processing is for reducing at least two dimension
Characteristic between difference；
By treated described at least two dimension, characteristic is integrated, and obtains the deep learning model
Input data.
On the other hand, the embodiment of the present application provides a kind of data processing equipment, and described device includes:
Data acquisition module, for obtaining the sample data of deep learning model, the sample data includes at least two
The characteristic of dimension；
Standardization processing module, the characteristic for each dimension in the characteristic at least two dimension
According to carrying out standardization processing respectively, obtain each dimension treated characteristic；Wherein, the standardization processing is used for
Reduce the difference between the characteristic of at least two dimension；
Processing module obtains institute for characteristic to be handled by treated described at least two dimension
State the input data of deep learning model.
In another aspect, the embodiment of the present application provides a kind of computer equipment, the computer equipment include processor and
Memory, is stored at least one instruction, at least one section of program, code set or instruction set in the memory, and described at least one
Item instruction, at least one section of program, the code set or the instruction set are loaded by the processor and are executed to realize above-mentioned side
Data processing method described in face.
Another aspect, the embodiment of the present application provide a kind of computer readable storage medium, the computer-readable storage
Be stored at least one instruction, at least one section of program, code set or instruction set in medium, at least one instruction, it is described extremely
Few one section of program, the code set or instruction set are loaded as processor and are executed to realize data processing side described in above-mentioned aspect
Method.
Another aspect provides a kind of computer program product, when the computer program product is performed, is used to hold
Data processing method described in the above-mentioned aspect of row.
Technical solution bring beneficial effect provided by the embodiments of the present application includes at least:
Standardization processing is carried out by the characteristic of each dimension in the sample data to deep learning model, so that on
The difference stated between the characteristic of each dimension in sample data reduces as much as possible, the subsequent place to above-mentioned each dimension
Characteristic after reason is handled, and the input data of deep learning model is obtained, subsequent to use the input data as depth
The training data or prediction data of learning model, due to the otherness between each dimension treated characteristic compared with
It is small, when training deep learning model, the convergent speed of deep learning model can be made to get a promotion, to improve deep learning
The training effectiveness of model, in addition, the precision of deep learning model can also get a promotion.
Detailed description of the invention
Fig. 1 is the flow chart for the data processing method that the application one embodiment provides；
Fig. 2 is the schematic diagram for the data processing method that the application one embodiment provides；
Fig. 3 is the schematic diagram for the binary conversion treatment that the application one embodiment provides；
Fig. 4 is the schematic diagram for the binary conversion treatment that another embodiment of the application provides；
Fig. 5 is the schematic diagram for the characteristic for indicating certain dimension that the relevant technologies provide；
Fig. 6 is the schematic diagram for the binary conversion treatment that another embodiment of the application provides；
Fig. 7 is the schematic diagram for the binary conversion treatment that another embodiment of the application provides；
Fig. 8 is the schematic diagram for the data processing method that another embodiment of the application provides；
Fig. 9 is the schematic diagram for the model training that the application one embodiment provides；
Figure 10 is the schematic diagram for the data processing method that another embodiment of the application provides；
Figure 11 is the schematic diagram for the model prediction that the application one embodiment provides；
Figure 12 is the schematic diagram for the application scenarios that the application one embodiment provides；
Figure 13 is the schematic diagram for the application scenarios that the application one embodiment provides；
Figure 14 is the block diagram for the data processing equipment that the application one embodiment provides；
Figure 15 is the block diagram for the data processing equipment that another embodiment of the application provides；
Figure 16 is the structural block diagram for the electronic equipment that the application one embodiment provides.
Specific embodiment
To keep the purposes, technical schemes and advantages of the application clearer, below in conjunction with attached drawing to the application embodiment party
Formula is described in further detail.
In technical solution provided by the embodiments of the present application, pass through each dimension in the sample data to deep learning model
Characteristic carry out standardization processing so that the difference between the value of the characteristic of each dimension in above-mentioned sample data
Different to reduce as much as possible, to above-mentioned each dimension, treated that characteristic is integrated later, obtains deep learning model
Input data, subsequent training data or prediction data using the input data as deep learning model, due to each
Otherness between dimension treated characteristic is smaller, when training deep learning model, can make deep learning mould
The convergence rate of type gets a promotion, so that the training effectiveness of deep learning model is improved, in addition, the precision of deep learning model
It can get a promotion.
In the application embodiment of the method, the executing subject of each step can be the computer with data-handling capacity and set
It is standby.Optionally, which, which can also be, has the function of model prediction and/or model training function.Above-mentioned computer is set
It is standby to can be PC (personal computer, personal computer) or server.
Deep learning model can obtain using such as image recognition, commending contents in a variety of different application scenarios.
In a kind of possible application scenarios, deep learning model can be used in card like games, such as in fighting landlord game, be used for root
Predict that epicycle is played a card according to the relevant information of current gambling party.
When deep learning model is used for fighting landlord project, training process and prediction process be can be such that
(1) training process
1, play card from history obtain in record training sample data (player role of every innings of game, grade, winning rate, in hand
Board, local exchange play a card record etc.)；
2, binary conversion treatment is carried out to above-mentioned training sample data and integration is handled, it is corresponding defeated to obtain training sample data
Enter data；
3, pass through the corresponding input data training deep learning model of above-mentioned training sample data.
Wherein, completing trained deep learning model can be placed in intelligent robot, and the intelligent robot has at this time
It plays card function, the fighting landlord of man-machine mode can be provided a user.
(2) process is predicted
1, during game is played a game, there is the intelligent robot for function of playing card to collect (the local exchange trip of forecast sample data
The player role of play, grade, winning rate, the board in hand, upper family play a card, the record of playing a card of local exchange game)；
2, binary conversion treatment is carried out to above-mentioned forecast sample data and integration is handled, it is corresponding defeated to obtain forecast sample data
Enter data；
3, above-mentioned forecast sample data are corresponded to input data input and complete training by the intelligent robot with function of playing card
Deep learning model, prediction result is exported by deep learning model, and determines that epicycle is played a card according to prediction result.
Referring to FIG. 1, the flow chart of the data processing method provided it illustrates the application one embodiment.This method packet
Include following steps:
Step 101, the sample data of deep learning model is obtained.
Sample data includes the characteristic of at least two dimensions.Spy included by sample data under different application scene
The dimension for levying data is usually different.By taking deep learning model is applied to fighting landlord project as an example, sample data may include
The identity information of each player, game ratings, clearing multiple, the board in hand, this play a card, local exchange is played a card record, epicycle winning rate
Etc. the characteristic of multiple dimensions.
The characteristic of each dimension can be array, discrete value, continuous real value etc., and the embodiment of the present application does not limit this
It is fixed.Array is made of multiple numerical value.Discrete value that is to say integer.Continuous real value that is to say decimal.With deep learning model
For fighting landlord project, board in hand, this is played a card and local exchange record of playing a card can be indicated using array；Identity
Information, game ratings, clearing multiple can be indicated using discrete value；Player's winning rate can be indicated using continuous real value.
The sample data of deep learning model can be training sample data, be also possible to forecast sample data, can be with
It is test sample data.For training deep learning model, forecast sample data are used for by completing training training sample data
Deep learning model predicted, test sample data be used for complete training sample data test.
In the embodiment of the present application, the acquisition modes of above-mentioned sample data can be determined according to its purposes is practical.Optionally,
Training sample data can be by log acquisition, and by taking deep learning model is applied to fighting landlord project as an example, computer equipment can
To obtain the games log of the fighting landlord project, and dissection process is carried out to above-mentioned games log, obtains training sample data.Separately
Outside, test sample data can also be obtained using above-mentioned identical mode.Optionally, forecast sample data, which can combine, works as front court
Scape obtains, by taking deep learning model is applied to fighting landlord project as an example, the related letter of the available current gambling party of computer equipment
Breath, and the relevant information of above-mentioned current gambling party is handled, obtain forecast sample data.
Step 102, it for the characteristic of each dimension in the characteristic of at least two dimensions, is standardized respectively
Change processing obtains each dimension treated characteristic.
Standardization processing is used to reduce the difference between the characteristic of above-mentioned at least two dimension.Standardization processing refers to
Standardize to the dimension and value of the characteristic of at least two dimensions.Standardization processing includes but is not limited to following any
One or more combinations: normalized, binary conversion treatment, Regularization, the embodiment of the present application are not construed as limiting this.?
In the embodiment of the present application, only it is explained so that standardization processing is binary conversion treatment as an example.
The dimension of the characteristic of certain dimension refers to the unit of the characteristic of the dimension.For example, the amount of account balance
Guiding principle is member, and the dimension of credit card quantity is.Since the dimension of the characteristic of each dimension may be identical, it is also possible to not phase
Together, it is therefore desirable to which standardization processing is carried out to the dimension of the characteristic of above-mentioned at least two dimension.In the embodiment of the present application,
Standardization processing is realized by the dimension of the characteristic of each dimension of removal.
In addition, in the embodiment of the present application, carrying out standardization to the value of characteristic and referring to diminution characteristic as far as possible
According to value between difference, subsequent to above-mentioned each dimension, treated that characteristic carries out integration processing, obtains depth
The input data of learning model, subsequent training data or prediction data using the input data as deep learning model,
Since the otherness between the value of the characteristic of each dimension is smaller, when training deep learning model, depth can be made
The convergent speed of learning model gets a promotion, so that the training effectiveness of deep learning model is improved, in addition, deep learning model
Precision can also get a promotion.
Optionally, step 102 implements are as follows: for the spy of i-th of dimension in the characteristic of at least two dimensions
Data are levied, binary conversion treatment is carried out, obtain i-th of dimension treated characteristic, i is positive integer.
Binary conversion treatment, which refers to, usually indicates characteristic using binaryzation member, and binaryzation element includes 0 and 1.Pass through
Aforesaid way, treated that characteristic only includes 0 and/or 1 for each dimension, reduces each dimension as much as possible to realize
Difference between the characteristic of degree.I-th of dimension can be any one dimension of above-mentioned at least two dimension.In addition, meter
Binary conversion treatment can also be carried out to the characteristic of each dimension of above-mentioned at least two dimension by calculating machine equipment, be obtained each
Treated characteristic that dimension is corresponding.
Treated that characteristic can be indicated using matrix or matrix stack for i-th of dimension, above-mentioned matrix or square
It only includes binaryzation element namely 0 and 1 that battle array, which is concentrated,.In addition, the ranks number of matrix can be according to the applied field of deep learning model
Scape is practical to be determined.For example, for fighting landlord project, since total board number includes the board of 15 seed types, each type of maximum quantity
It is 4, therefore can be indicated using 15 × 4 matrix.For another example for Chinese chess scene, since each party includes 7 seed types
Chess piece, the maximum quantity of each type of chess piece is 5, therefore can be indicated using 7 × 5 matrix.In addition, after for convenience
The ranks number of continuous operation, each dimension treated corresponding matrix of characteristic should be identical.
Step 103, by least two dimensions, treated that characteristic is handled, and obtains the defeated of deep learning model
Enter data.
Since the training process or prediction process of deep learning model are needed using each dimension treated feature, because
This also needs that treated that characteristic is handled to above-mentioned at least two dimension, obtains being suitable for input deep learning mould
The input data of type.
It in the embodiment of the present application, can be at least two dimensions treated mode that characteristic integrated
It is that treated that characteristic carries out stacking processing by above-mentioned at least two dimension.Optionally, computer equipment passes through matrix
Treated that characteristic carries out stacking processing to above-mentioned at least two dimension for the heap function of functions, obtains three-dimensional input data.
The above-mentioned matrix heap function of functions can be stack function.
In conjunction with reference Fig. 2, it illustrates the schematic diagram for the data handling procedure that the application one embodiment provides, computers
Equipment carries out binary conversion treatment to the characteristic of each dimension, obtains the feature of each dimension using binaryzation element representation
Data carry out integration processing to the characteristic of above-mentioned each dimension using binaryzation element representation later, obtain three-dimensional
The input data of above-mentioned three-dimensional is inputted initial deep learning model by input data later, completes the instruction of deep learning model
Practice, or the input data input of above-mentioned three-dimensional is completed to the deep learning model of training, obtains corresponding prediction result.
In conclusion technical solution provided by the embodiments of the present application, by each in the sample data to deep learning model
The dimension and/or value of the characteristic of a dimension carry out standardization processing, so that each dimension in above-mentioned sample data
Difference between characteristic reduces as much as possible, and subsequent to above-mentioned each dimension, treated that characteristic is integrated,
Obtain the input data of deep learning model, it is subsequent using the input data as the training data of deep learning model or pre-
Measured data, since the otherness between the dimension and value of each dimension treated characteristic is smaller, when training depth
When learning model, the convergent speed of deep learning model can be made to get a promotion, to improve the training effect of deep learning model
Rate, in addition, the precision of deep learning model can also get a promotion.
It was introduced in foregoing embodiments, the characteristic of each dimension can be array, is also possible to discrete value, may be used also
To be continuous real value.For the characteristic of different representations, the also not phase of method used by binary conversion treatment is carried out to it
Together.These three situations will be explained respectively below.
1, characteristic uses array representation.
When the characteristic of i-th of dimension uses array representation, binaryzation is carried out to the characteristic of i-th of dimension
Processing may include: the objective matrix be converted to the characteristic of i-th of dimension using binaryzation element representation.
Objective matrix is two-dimensional matrix, and the row of objective matrix indicates the first parameter of the characteristic of i-th of dimension, target
Matrix column indicates the second parameter of the characteristic of i-th of dimension.First parameter refers to that the characteristic of i-th of dimension is wrapped
The maximum frequency of occurrence of each type included.Second parameter refers to the sum of type included by the characteristic of i-th of dimension
Amount.For example, the characteristic of i-th of dimension is the board face amount of the board in hand, due to the board face amount of the board in hand may include 3,
4,5,6,7,8,9,10, J, Q, K, A, 2, B, R totally 15 kinds of board face amounts, wherein B represents Xiao Wang, and R represents king.Every kind of board face amount
Maximum frequency of occurrence is 4, therefore the first parameter is 4, and the second parameter is 15.
Optionally, the matrix element of the target location of the first matrix is 1, its in the first matrix in addition to target position
Matrix element at its position is 0.One matrix element of target location, in the characteristic for indicating i-th of dimension
An array element.
In conjunction with reference Fig. 3, it illustrates the schematic diagrames of the binary conversion treatment shown in one exemplary embodiment of the application.The
The characteristic of i dimension is played a card for this, this board face data played a card can be adopted with reference to part (a) in Fig. 3
It is indicated with array { 3,3,4,4,5,5 }, i-th of dimension treated characteristic can be with reference to part (b) in Fig. 3.
In conjunction with reference Fig. 4, it illustrates the schematic diagrames of the binary conversion treatment shown in another exemplary embodiment of the application.
The characteristic of i-th of dimension is the board in hand, and the board face data of the board in hand can be with reference to part (a) in Fig. 4, can be with
It is indicated using array { 8,8,9,9,10,10, J, J, K, K, K, K }, treated that characteristic can refer to for i-th of dimension
Part (b) in Fig. 4.
2, characteristic is indicated using discrete value.
When the characteristic of i-th of dimension using discrete value to indicate when, can be by using binaryzation element representation
Matrix stack indicates the characteristic of above-mentioned i-th of dimension.
In the first possible implementation, for the spy of i-th of dimension in the characteristic of at least two dimensions
Data are levied, binary conversion treatment is carried out, obtain i-th of dimension treated that characteristic may be implemented are as follows: by i-th dimension
Characteristic is converted to the first matrix stack using binaryzation element representation.
In the embodiment of the present application, the first matrix stack generally includes multiple matrixes, wherein the square for including in the first matrix stack
Battle array quantity is the value quantity of the characteristic of i-th of dimension.For example, the feature of i-th of dimension is grade, the value packet of grade
4 are included, at this time for indicating that the first matrix stack of the value of grade includes 4 the second matrixes.In addition, different in the first matrix stack
The matrix of position is used to indicate the different values of the characteristic of i-th of dimension.For example, the feature of i-th of dimension is grade, the
The matrix of third position is come in one matrix stack for indicating that grade is 3.
Optionally, include: in the first matrix stack matrix element be all 1 the first matrix and matrix element be all 0
Two matrixes.Wherein, position of first matrix in the first matrix stack is used to indicate the characteristic of i-th of dimension.In the application
In embodiment, the position of 1 the first matrix in the first matrix stack is all by matrix element to indicate the discrete value, Ye Ji
The characteristic of i dimension.
In conjunction with reference Fig. 5, it illustrates the schematic diagrames for the characteristic for indicating i-th of dimension in the related technology.I-th of dimension
The feature of degree is player levels, and corresponding characteristic is that discrete value 3 generallys use a matrix in the related art
Indicate player levels, the matrix element in the matrix is all the discrete value for indicating the characteristic of i-th of dimension, Ye Jiquan
It is 3.
In conjunction with reference Fig. 6, it illustrates the schematic diagrames of the binary conversion treatment shown in the application one embodiment.I-th of dimension
The feature of degree is player levels, and corresponding characteristic is discrete value 3, due to the value of player levels may include 1,2,3,
4 four grades, therefore the first matrix stack includes 1 the first matrix and 3 the second matrixes, and the first matrix is in the first matrix stack
Position is third position.
In other possible examples, the first matrix stack includes: the first matrix and the matrix element that matrix element is all 1
Element is all 0 the second matrix.Wherein, position of second matrix in the first matrix stack is used to indicate the characteristic of i-th of dimension
According to.In the embodiment of the present application, the position of 0 the second matrix in the first matrix stack is all by matrix element come indicate this from
Dissipate the characteristic of value namely i-th of dimension.
In the second possible implementation, for the spy of i-th of dimension in the characteristic of at least two dimensions
Data are levied, binary conversion treatment is carried out, obtain i-th of dimension treated that characteristic may be implemented are as follows: by i-th dimension
Characteristic is converted to the second matrix stack using binaryzation element representation；
The matrix quantity for including in second matrix stack is m, and the m power of target constant is more than or equal to the feature of i-th of dimension
The maximum value of data, the matrix of different location is the integer greater than 1 for indicating different numerical value, m in the second matrix stack.
Target constant can be practical determining according to the value of the characteristic of i-th of dimension.Optionally, i-th dimension
Each value of characteristic can be indicated by the integral number power of target constant.For example, the characteristic of i-th of dimension
Value be respectively 14,26,259, second constant 2.Wherein, 14 can using 23 power, 2 power and 1 power and come
It indicates namely 14=2^3+2^2+2^1,26 can be indicated namely 26=2^ using the sum of 24 power, 3 power and 1 power
4+2^3+2^1,259 can be indicated namely 259=2^8+2^1+2^0 using the sum of 28 power, 1 power and 0 power.
Optionally, the m power of target constant is greater than the smallest positive integral of the maximum value of the characteristic of i-th of dimension.
For example, the maximum value of the characteristic of i-th of dimension is 872, preset constant 2, then first constant is 1024 namely 2
10 power.
Optionally, include: in the second matrix stack matrix element be all 1 third matrix and matrix element be all 0
Four matrixes.Wherein, the sum of numerical value represented by third matrix, is used to indicate the characteristic of i-th of dimension.Implement in the application
In example, the sum of numerical value represented by 1 third matrix is all by matrix element to indicate the discrete value namely i-th of dimension
Characteristic.
In conjunction with reference Fig. 7, it illustrates the schematic diagrames of the binary conversion treatment shown in the application one embodiment.I-th of dimension
The feature of degree is the multiple of this gambling party, and corresponding characteristic is 259, preset constant 2,259=2^8+2^1+2^0,
Second matrix stack includes 3 third matrixes, is located at the 1st, the 2nd and the 9th of the second matrix stack.
It include: the third matrix and matrix that matrix element is all 1 in other possible examples, in the second matrix stack
Element is all 0 the 4th matrix.Wherein, the sum of numerical value represented by the 4th matrix, is used to indicate the characteristic of i-th of dimension
According to.In the embodiment of the present application, the sum of numerical value represented by 0 the 4th matrix is all by matrix element to indicate that this is discrete
The characteristic of value namely i-th of dimension.
3, characteristic is indicated using continuous real value.
Since continuous real value is generally difficult to carry out binary conversion treatment, to i-th of dimension using continuous real value representation
Characteristic carry out binary conversion treatment when, first conversion process is carried out to it.
Optionally, when the characteristic of i-th of dimension using continuous real value to indicate when, for the spy of at least two dimensions
The characteristic of i-th of dimension in data is levied, binary conversion treatment is carried out, obtains i-th of dimension treated characteristic
May include following two sub-steps:
Step 307, continuous real value is converted into discrete value.
Step 308, binary conversion treatment is carried out for discrete value, obtains i-th of dimension treated characteristic.
Optionally, computer equipment determines the mapping relations between continuous real value and discrete value, later according to above-mentioned mapping
Continuous real value is converted to discrete value by relationship.Illustratively, the feature of i-th of dimension is this winning rate, and corresponding data are adopted
It is indicated with continuous real value, also, existing mapping relations can be with reference table -1 between continuous real value and discrete value.
| Continuous real value | (0.00,025] | (0.25,0.50] | (0.50,0.75] | (0.75,1) |
| Discrete value | 1 | 2 | 3 | 4 |
For example, the corresponding local exchange winning rate of player A is 0.17, the discrete value obtained after converting is 1；Player B is corresponding
Local exchange winning rate be 0.53, the discrete value obtained after converting is 3；The corresponding local exchange winning rate of player C is 0.30, is passed through
The discrete value obtained after converting is 2.
In addition, the mode for carrying out binary conversion treatment to discrete value can join after continuous real value is converted to discrete value
See above-described embodiment, is not repeated herein.
The input data of above-mentioned deep learning model can be applied to model training.Below to using deep learning model
The process that input data is trained is explained.Referring to FIG. 8, it illustrates at the data shown in the application one embodiment
The flow chart of reason method.This method comprises the following steps:
Step 801, the training sample data of deep learning model are obtained.
Training sample data include the characteristic of at least two dimensions.
Step 802, it for the characteristic of each dimension in the characteristic of at least two dimensions, is standardized respectively
Change processing obtains each dimension treated characteristic.
Standardization processing is used to reduce the difference between the characteristic of at least two dimensions.
Step 803, by least two dimensions, treated that characteristic is integrated, and it is corresponding to obtain training sample data
Input data.
Step 804, it calls deep learning model to handle the corresponding input data of training sample data, is trained
The corresponding output result of sample data.
In the embodiment of the present application, the input data of above-mentioned deep learning model is training sample data, by above-mentioned training
Sample data inputs initial deep learning model, exports above-mentioned training sample data pair by above-mentioned initial deep learning model
The prediction result answered.Above-mentioned initial deep learning model can be convolutional neural networks (Convolutional Neural
Network, CNN), such as alexNet network, VGG-16 network, GoogleNet network, Deep Residual Learning
(study of depth residual error) network etc., the embodiment of the present application is not construed as limiting this.
Step 805, according to the corresponding output result of training sample data and actual result, to the ginseng in deep learning model
Number is adjusted, and obtains the deep learning model for completing training.
In the training process, prediction result and actual result are compared computer equipment, obtain loss function value, if
Loss function value is less than preset threshold, then terminates process, and obtains completing the deep learning model of training；If loss function value is big
In preset threshold, then the parameter of above-mentioned initial deep learning model is adjusted, repeats step 804 later to 805, until
Loss function value is less than preset threshold, corresponding deep learning model when loss function value is less than preset threshold by computer equipment
As the deep learning model for completing training.
In addition, above-mentioned preset threshold can be set according to actual needs, the embodiment of the present application is not construed as limiting this.In addition,
The embodiment of the present application is also not construed as limiting algorithm used by training deep learning model, can be BP (Back-
Propagation, back-propagation algorithm), faster RCNN (Regions with Convolutional Neural
Network, region convolutional neural networks) algorithm etc..
In one specifically example, in conjunction with reference Fig. 9, it is with the deep learning model that training is applied to fighting landlord project
Example, computer equipment first collect games log from game side, games log are converted to a line log later and represents a game
The form of movement is converted to game object to above-mentioned game action later, the game object include player role, player rank,
Then the information such as upper family plays a card, current player is played a card arrange data according to information filterings such as winning rate, player ranks, later basis
Above-mentioned data are converted to input channel matrix stack by binaryzation thought, later using input channel matrix stack as the input of CNN into
Row model training finally obtains the CNN model of training completion.
The input data of above-mentioned deep learning model can be applied to model prediction.Below to using deep learning model
The process that input data is predicted is explained.Referring to FIG. 10, it illustrates the data shown in the application one embodiment
The flow chart of processing method.This method comprises the following steps:
Step 1001, the forecast sample data of deep learning model are obtained.
Forecast sample data include the characteristic of at least two dimensions.
Step 1002, it for the characteristic of each dimension in the characteristic of at least two dimensions, is advised respectively
Generalized processing obtains each dimension treated characteristic.
Standardization processing, which refers to, standardizes to the dimension and/or value of the characteristic of at least two dimensions.
Step 1003, by least two dimensions, treated that characteristic is integrated, and obtains forecast sample data pair
The input data answered.
Step 1004, call training complete deep learning model to the corresponding input data of forecast sample data at
Reason, obtains the corresponding output result of forecast sample data.
In the embodiment of the present application, the input data of deep learning model is forecast sample data, by above-mentioned forecast sample
The deep learning model that data input training is completed is tied by the corresponding output of deep learning model output forecast sample data
Fruit.
It is first by taking deep learning model is applied to fighting landlord project as an example in conjunction with reference Figure 11 in one specifically example
It is first online to generate gambling party, mode input data are secondly generated according to gambling party information, binaryzation thought are then utilized, by input data
It is converted into input channel matrix stack, input channel matrix stack carries out model prediction as the input of CNN later, obtains epicycle hands
Probability distribution, finally obtain prediction result according to model output probability profile samples.
In conjunction with reference Figure 12, it illustrates the schematic diagrames of the application scenarios shown in the application one embodiment, in the application
In scene, the data such as grade, history gambling party winning rate of board, each player in intelligent robot acquisition hand, later to above-mentioned number
According to binary conversion treatment is carried out, the deep learning model for finally completing the data input training after binary conversion treatment, and then determine
Whether local exchange needs to rob landlord.
In conjunction with reference Figure 13, it illustrates the schematic diagrames of the application scenarios shown in the application one embodiment, in the application
In scene, intelligent robot obtains board in hand, upper family plays a card, the data such as the grade of each player, history gambling party winning rate, later
Binary conversion treatment is carried out to above-mentioned data, the deep learning model for finally completing the data input training after binary conversion treatment,
And then determine that local exchange needs board out.
Following is the application Installation practice, can be used for executing the application embodiment of the method.It is real for the application device
Undisclosed details in example is applied, the application embodiment of the method is please referred to.
Figure 14 is please referred to, it illustrates the block diagrams for the data processing equipment that the application one embodiment provides.Device tool
There is the function of realizing in above method example, the function can also be executed corresponding software by hardware realization by hardware
It realizes.The apparatus may include: data acquisition module 1401, standardization processing module 1402 and integrate processing module 1403.
Data acquisition module 1401, for obtaining the sample data of deep learning model, the sample data includes at least
The characteristic of two dimensions.
Standardization processing module 1402, the spy for each dimension in the characteristic at least two dimension
Data are levied, carry out standardization processing respectively, obtain each dimension treated characteristic；Wherein, the standardization processing
The difference between characteristic for reducing at least two dimension.
Processing module 1403 is obtained for by treated described at least two dimension, characteristic to be handled
To the input data of the deep learning model.
In conclusion technical solution provided by the embodiments of the present application, by the sample data to deep learning model
The characteristic of each dimension carries out standardization processing, so that between the characteristic of each dimension in above-mentioned sample data
Difference reduces as much as possible, and subsequent to above-mentioned each dimension, treated that characteristic is handled, and obtains deep learning mould
The input data of type, subsequent training data or prediction data using the input data as deep learning model, due to each
Otherness between a dimension treated characteristic is smaller, when training deep learning model, can make deep learning
The convergent speed of model gets a promotion, so that the training effectiveness of deep learning model is improved, in addition, the precision of deep learning model
Also it can get a promotion.
In the alternative embodiment provided based on embodiment illustrated in fig. 14, the standardization processing module 1402 is used
The characteristic of i-th of dimension in the characteristic at least two dimension carries out binary conversion treatment, obtains institute
State i-th of dimension treated characteristic.
Optionally, the characteristic of i-th of dimension uses array representation；The standardization processing module 1402 is used
In the characteristic of i-th of dimension to be converted to the objective matrix using binaryzation element representation；Wherein, the target square
Battle array is two-dimensional matrix, and the row of the objective matrix indicates the first parameter of the characteristic of i-th of dimension, the target square
The column of battle array indicate the second parameter of the characteristic of i-th of dimension.
Optionally, the matrix element of the target location of the objective matrix is 1, removes the target in the objective matrix
The matrix element at other positions except position is 0；Alternatively, the matrix element of the target location of the objective matrix is 0,
The matrix element at other positions in the objective matrix in addition to the target position is 1；Wherein, the target location
A matrix element, the array element in characteristic for indicating i-th of dimension.
Optionally, the characteristic of i-th of dimension is indicated using discrete value；The standardization processing module 1402,
For the characteristic of i-th of dimension to be converted to the first matrix stack using binaryzation element representation；Wherein, described
The matrix quantity for including in one matrix stack is the value quantity of the characteristic of i-th of dimension, in first matrix stack
The matrix of different location is used to indicate the different values of the characteristic of i-th of dimension.
It optionally, include: that matrix element is all 1 the first matrix and matrix element is all 0 in first matrix stack
The second matrix；Position of first matrix in first matrix stack, is used to indicate the characteristic of i-th of dimension
According to alternatively, position of second matrix in first matrix stack, is used to indicate the characteristic of i-th of dimension.
Optionally, the characteristic of i-th of dimension is indicated using discrete value；The standardization processing module 1402,
For the characteristic of i-th of dimension to be converted to the second matrix stack using binaryzation element representation；Wherein, described
The matrix quantity for including in two matrix stacks is m, and the m power of preset constant is more than or equal to the characteristic of i-th of dimension
Maximum value, the matrix of different location is the integer greater than 1 for indicating different numerical value, the m in second matrix stack.
It optionally, include: that matrix element is all 1 third matrix and matrix element is all 0 in second matrix stack
The 4th matrix；The sum of numerical value represented by the third matrix is used to indicate the characteristic of i-th of dimension；Alternatively,
The sum of numerical value represented by 4th matrix is used to indicate the characteristic of i-th of dimension.
Optionally, the standardization processing module 1402 is also used to the characteristic when i-th of dimension using continuous
When real value representation, the continuous real value is converted into the discrete value, obtains i-th of the dimension indicated using the discrete value
The characteristic of degree.
In another alternative embodiment provided based on embodiment illustrated in fig. 14, the sample data is number of training
According to；Please refer to Figure 15, described device further include: model training module 1404.
Model training module 1404, is used for:
It calls the deep learning model to handle the input data, it is corresponding to obtain the training sample data
Export result；
According to the corresponding output result of the training sample data and actual result, to the ginseng in the deep learning model
Number is adjusted, and obtains the deep learning model for completing training.
In another alternative embodiment provided based on embodiment illustrated in fig. 14, the sample data is forecast sample number
According to；Please refer to Figure 15, described device further include: model prediction module 1405.
Model prediction module 1405, the deep learning model for calling training to complete carry out the input data
Processing, obtains the corresponding output result of the forecast sample data.
Figure 16 is please referred to, it illustrates the structural schematic diagrams of computer equipment provided by one embodiment of the present invention.The meter
Calculating machine equipment can be PC or server.The computer equipment is used for the data processing method for implementing to provide in above-described embodiment.
Specifically:
Computer equipment 1600 includes 1602 He of central processing unit (CPU) 1601 including random access memory (RAM)
The system storage 1604 of read-only memory (ROM) 1603, and connection system storage 1604 and central processing unit 1601
System bus 1605.Computer equipment 1600 further includes that the substantially defeated of information is transmitted between each device helped in computer
Enter/output system (I/O system) 1606, and is used for storage program area 1613, application program 1614 and other program modules
1615 mass-memory unit 1607.
Basic input/output 1606 includes display 1608 for showing information and inputs information for user
Such as mouse, keyboard etc input equipment 1609.Wherein display 1608 and input equipment 1609 are all by being connected to
The input and output controller 1610 of system bus 1605 is connected to central processing unit 1601.Basic input/output 1606 is also
May include input and output controller 1610 with for receive and handle from keyboard, mouse or electronic touch pen etc. it is multiple its
The input of his equipment.Similarly, input and output controller 1610 also provides output to display screen, printer or other kinds of defeated
Equipment out.
Mass-memory unit 1607 is connected by being connected to the bulk memory controller (not shown) of system bus 1605
It is connected to central processing unit 1601.Mass-memory unit 1607 and its associated computer-readable medium are computer equipment
1600 provide non-volatile memories.That is, mass-memory unit 1607 may include that such as hard disk or CD-ROM drive
The computer-readable medium (not shown) of dynamic device etc.
Without loss of generality, computer-readable medium may include computer storage media and communication media.Computer storage
Medium includes any of the information such as computer readable instructions, data structure, program module or other data for storage
The volatile and non-volatile of method or technique realization, removable and irremovable medium.Computer storage medium include RAM,
ROM, EPROM, EEPROM, flash memory or other solid-state storages its technologies, CD-ROM, DVD or other optical storages, cassette, magnetic
Band, disk storage or other magnetic storage devices.Certainly, skilled person will appreciate that computer storage medium is not limited to
It states several.Above-mentioned system storage 1604 and mass-memory unit 1607 may be collectively referred to as memory.
According to various embodiments of the present invention, computer equipment 1600 can also be arrived by network connections such as internets
Remote computer operation on network.Namely computer equipment 1600 can be connect by the network being connected on system bus 1605
Mouth unit 1611 is connected to network 1612, in other words, it is other kinds of to be connected to that Network Interface Unit 1611 also can be used
Network or remote computer system (not shown).
Memory further includes one, and perhaps more than one program one or more than one program are stored in memory
In, and be configured to be executed by one or more than one processor.Said one or more than one program include for holding
The instruction of the above-mentioned data processing method of row.
In the exemplary embodiment, a kind of computer readable storage medium is additionally provided, is stored in the storage medium
At least one instruction, at least one section of program, code set or instruction set, at least one instruction, at least one section of program, institute
Code set or instruction set is stated to be loaded by the processor of electronic equipment and executed to realize the data processing in above method embodiment
Method.
Optionally, above-mentioned computer readable storage medium can be ROM, random access memory (RAM), CD-ROM, magnetic
Band, floppy disk and optical data storage devices etc..
In the exemplary embodiment, a kind of computer program product is additionally provided, when the computer program product is performed
When, it is used to execute the data processing method in above-mentioned aspect embodiment.
It should be understood that referenced herein " multiple " refer to two or more."and/or", description association
The incidence relation of object indicates may exist three kinds of relationships, for example, A and/or B, can indicate: individualism A exists simultaneously A
And B, individualism B these three situations.Character "/" typicallys represent the relationship that forward-backward correlation object is a kind of "or".Make herein
" first ", " second " and similar word are not offered as any sequence, quantity or importance, and are used only to distinguish
Different component parts.
Above-mentioned the embodiment of the present application serial number is for illustration only, does not represent the advantages or disadvantages of the embodiments.
The above is only the exemplary embodiments of the application, all in spirit herein and original not to limit the application
Within then, any modification, equivalent replacement, improvement and so on be should be included within the scope of protection of this application.
Claims (14)
1. a kind of data processing method, which is characterized in that the described method includes:
The sample data of deep learning model is obtained, the sample data includes the characteristic of at least two dimensions；
For the characteristic of each dimension in the characteristic of at least two dimension, standardization processing is carried out respectively,
Obtain each dimension treated characteristic；Wherein, the standardization processing is for reducing at least two dimension
Difference between characteristic；
By treated described at least two dimension, characteristic is handled, and obtains the defeated of the deep learning model
Enter data.
2. the method according to claim 1, wherein in the characteristic at least two dimension
Each dimension characteristic, carry out standardization processing respectively, obtain each dimension treated characteristic, comprising:
For the characteristic of i-th of dimension in the characteristic of at least two dimension, binary conversion treatment is carried out, is obtained
I-th of dimension treated characteristic.
3. according to the method described in claim 2, it is characterized in that, the characteristic of i-th of dimension uses array representation；
The characteristic of i-th of dimension in the characteristic at least two dimension carries out binary conversion treatment,
Obtain i-th of dimension treated characteristic, comprising:
The characteristic of i-th of dimension is converted to the objective matrix using binaryzation element representation；
Wherein, the objective matrix is two-dimensional matrix, and the row of the objective matrix indicates the characteristic of i-th of dimension
First parameter, the column of the objective matrix indicate the second parameter of the characteristic of i-th of dimension.
4. according to the method described in claim 3, it is characterized in that,
The matrix element of the target location of the objective matrix is 1, in the objective matrix in addition to the target position
Matrix element at other positions is 0；Alternatively, the matrix element of the target location of the objective matrix is 0, the target square
The matrix element at other positions in battle array in addition to the target position is 1；
Wherein, a matrix element of the target location, one in characteristic for indicating i-th of dimension
Array element.
5. according to the method described in claim 2, it is characterized in that, the characteristic of i-th of dimension uses discrete value table
Show；
The characteristic of i-th of dimension in the characteristic at least two dimension carries out binary conversion treatment,
Obtain i-th of dimension treated characteristic, comprising:
The characteristic of i-th of dimension is converted into the first matrix stack using binaryzation element representation；
Wherein, the matrix quantity for including in first matrix stack is the value quantity of the characteristic of i-th of dimension, institute
State the different values of characteristic of the matrix of different location in the first matrix stack for indicating i-th of dimension.
6. according to the method described in claim 5, it is characterized in that, including: that matrix element is all 1 in first matrix stack
First matrix and matrix element are all 0 the second matrix；
Position of first matrix in first matrix stack is used to indicate the characteristic of i-th of dimension, or
Person, position of second matrix in first matrix stack, is used to indicate the characteristic of i-th of dimension.
7. according to the method described in claim 2, it is characterized in that, the characteristic of i-th of dimension uses discrete value table
Show；
The characteristic of i-th of dimension in the characteristic at least two dimension carries out binary conversion treatment,
Obtain i-th of dimension treated characteristic, comprising:
The characteristic of i-th of dimension is converted into the second matrix stack using binaryzation element representation；
Wherein, the matrix quantity for including in second matrix stack is m, and the m power of target constant is more than or equal to described i-th dimension
The maximum value of the characteristic of degree, the matrix of different location is for indicating different numerical value, the m in second matrix stack
For the integer greater than 1.
8. the method according to the description of claim 7 is characterized in that including: that matrix element is all 1 in second matrix stack
Third matrix and matrix element are all 0 the 4th matrix；
The sum of numerical value represented by the third matrix is used to indicate the characteristic of i-th of dimension；Alternatively, described
The sum of numerical value represented by four matrixes is used to indicate the characteristic of i-th of dimension.
9. the method according to claim 5 or 7, which is characterized in that the characteristic at least two dimension
The characteristic of i-th of dimension in carries out binary conversion treatment, obtains i-th of dimension treated characteristic
Before, further includes:
When the characteristic of i-th of dimension uses continuous real value representation, the continuous real value is converted to described discrete
Value obtains the characteristic of i-th of the dimension indicated using the discrete value.
10. method according to any one of claims 1 to 8, which is characterized in that the sample data is number of training
According to；
Described by treated described at least two dimension, characteristic is integrated, and obtains the deep learning model
Input data after, further includes:
It calls the deep learning model to handle the input data, obtains the corresponding output of the training sample data
As a result；
According to the corresponding output result of the training sample data and actual result, to the parameter in the deep learning model into
Row adjustment obtains the deep learning model for completing training.
11. method according to any one of claims 1 to 8, which is characterized in that the sample data is forecast sample number
According to；
Described by treated described at least two dimension, characteristic is integrated, and obtains the machine learning model
Input data after, further includes:
The deep learning model for calling training to complete handles the input data, obtains the forecast sample data
Corresponding output result.
12. a kind of data processing equipment, which is characterized in that described device includes:
Data acquisition module, for obtaining the sample data of deep learning model, the sample data includes at least two dimensions
Characteristic；
Standardization processing module, for the characteristic of each dimension in the characteristic at least two dimension,
Standardization processing is carried out respectively, obtains each dimension treated characteristic；Wherein, the standardization processing is for reducing
Difference between the characteristic of at least two dimension；
Processing module obtains the depth for characteristic to be handled by treated described at least two dimension
Spend the input data of learning model.
13. a kind of computer equipment, which is characterized in that the computer equipment includes processor and memory, the memory
In be stored at least one instruction, at least one section of program, code set or instruction set, at least one instruction, described at least one
Duan Chengxu, the code set or instruction set are loaded by the processor and are executed to realize such as any one of claim 1 to 11 institute
The data processing method stated.
14. a kind of computer readable storage medium, which is characterized in that be stored at least one in the computer readable storage medium
Item instruction, at least one section of program, code set or instruction set, at least one instruction, at least one section of program, the code
Collection or instruction set are loaded by processor and are executed to realize data processing method as described in any one of claim 1 to 11.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810783681.1A CN109033309B (en) | 2018-07-17 | 2018-07-17 | Data processing method, device, equipment and storage medium |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810783681.1A CN109033309B (en) | 2018-07-17 | 2018-07-17 | Data processing method, device, equipment and storage medium |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN109033309A true CN109033309A (en) | 2018-12-18 |
| CN109033309B CN109033309B (en) | 2023-04-07 |
Family
ID=64643549
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201810783681.1A Active CN109033309B (en) | 2018-07-17 | 2018-07-17 | Data processing method, device, equipment and storage medium |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109033309B (en) |
Cited By (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110276096A (en) * | 2019-05-08 | 2019-09-24 | 北京百度网讯科技有限公司 | Improve method, electronic equipment and the storage medium of deep learning model prediction ability |
| CN110443284A (en) * | 2019-07-15 | 2019-11-12 | 超参数科技(深圳)有限公司 | Training method, call method, server and the readable storage medium storing program for executing of AI model |
| CN110602709A (en) * | 2019-09-16 | 2019-12-20 | 腾讯科技（深圳）有限公司 | Network data security method and device of wearable device and storage medium |
| CN110598855A (en) * | 2019-09-23 | 2019-12-20 | Oppo广东移动通信有限公司 | Deep learning model generation method, device, equipment and storage medium |
| CN113274738A (en) * | 2021-06-09 | 2021-08-20 | 浙江畅唐网络股份有限公司 | Card type data acquisition method of whipped egg game and related device |
| CN115130621A (en) * | 2022-08-31 | 2022-09-30 | 支付宝(杭州)信息技术有限公司 | Model training method and device, storage medium and electronic equipment |
Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20020016195A1 (en) * | 2000-08-01 | 2002-02-07 | Konami Computer Entertainment Osaka, Inc. | Game procedure control method, game system, and server |
| CN106055339A (en) * | 2016-06-08 | 2016-10-26 | 天津联众逸动科技发展有限公司 | Method for determining card playing strategy of computer player in two-against-one game |
| CN106779087A (en) * | 2016-11-30 | 2017-05-31 | 福建亿榕信息技术有限公司 | A kind of general-purpose machinery learning data analysis platform |
| CN107423442A (en) * | 2017-08-07 | 2017-12-01 | 火烈鸟网络(广州)股份有限公司 | Method and system, storage medium and computer equipment are recommended in application based on user's portrait behavioural analysis |
| CN107516135A (en) * | 2017-07-14 | 2017-12-26 | 浙江大学 | A kind of automation monitoring learning method for supporting multi-source data |
| CN107998661A (en) * | 2017-12-26 | 2018-05-08 | 苏州大学 | A kind of aid decision-making method, device and the storage medium of online battle game |
-
2018
- 2018-07-17 CN CN201810783681.1A patent/CN109033309B/en active Active
Patent Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20020016195A1 (en) * | 2000-08-01 | 2002-02-07 | Konami Computer Entertainment Osaka, Inc. | Game procedure control method, game system, and server |
| CN106055339A (en) * | 2016-06-08 | 2016-10-26 | 天津联众逸动科技发展有限公司 | Method for determining card playing strategy of computer player in two-against-one game |
| CN106779087A (en) * | 2016-11-30 | 2017-05-31 | 福建亿榕信息技术有限公司 | A kind of general-purpose machinery learning data analysis platform |
| CN107516135A (en) * | 2017-07-14 | 2017-12-26 | 浙江大学 | A kind of automation monitoring learning method for supporting multi-source data |
| CN107423442A (en) * | 2017-08-07 | 2017-12-01 | 火烈鸟网络(广州)股份有限公司 | Method and system, storage medium and computer equipment are recommended in application based on user's portrait behavioural analysis |
| CN107998661A (en) * | 2017-12-26 | 2018-05-08 | 苏州大学 | A kind of aid decision-making method, device and the storage medium of online battle game |
Cited By (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110276096A (en) * | 2019-05-08 | 2019-09-24 | 北京百度网讯科技有限公司 | Improve method, electronic equipment and the storage medium of deep learning model prediction ability |
| CN110443284A (en) * | 2019-07-15 | 2019-11-12 | 超参数科技(深圳)有限公司 | Training method, call method, server and the readable storage medium storing program for executing of AI model |
| CN110602709A (en) * | 2019-09-16 | 2019-12-20 | 腾讯科技（深圳）有限公司 | Network data security method and device of wearable device and storage medium |
| CN110602709B (en) * | 2019-09-16 | 2022-01-04 | 腾讯科技（深圳）有限公司 | Network data security method and device of wearable device and storage medium |
| CN110598855A (en) * | 2019-09-23 | 2019-12-20 | Oppo广东移动通信有限公司 | Deep learning model generation method, device, equipment and storage medium |
| CN113274738A (en) * | 2021-06-09 | 2021-08-20 | 浙江畅唐网络股份有限公司 | Card type data acquisition method of whipped egg game and related device |
| CN115130621A (en) * | 2022-08-31 | 2022-09-30 | 支付宝(杭州)信息技术有限公司 | Model training method and device, storage medium and electronic equipment |
| CN115130621B (en) * | 2022-08-31 | 2022-12-27 | 支付宝(杭州)信息技术有限公司 | Model training method and device, storage medium and electronic equipment |
Also Published As
| Publication number | Publication date |
|---|---|
| CN109033309B (en) | 2023-04-07 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109033309A (en) | Data processing method, device, equipment and storage medium | |
| CN108665159A (en) | A kind of methods of risk assessment, device, terminal device and storage medium | |
| CN104778173B (en) | Target user determination method, device and equipment | |
| CN105095964B (en) | A kind of data processing method and device | |
| CN110390561B (en) | User-financial product selection tendency high-speed prediction method and device based on momentum acceleration random gradient decline | |
| CN110443284A (en) | Training method, call method, server and the readable storage medium storing program for executing of AI model | |
| CN107547759A (en) | A kind of quality detecting method and device to contact staff's call | |
| CN107133865A (en) | A kind of acquisition of credit score, the output intent and its device of characteristic vector value | |
| CN110138627A (en) | The edge calculations lateral terminal secure accessing policy selection method of quantization | |
| CN109600344A (en) | Identify the method, apparatus and electronic equipment of risk group | |
| CN107992978A (en) | It is a kind of to net the method for prewarning risk and relevant apparatus for borrowing platform | |
| CN112183098A (en) | Session processing method and device, storage medium and electronic device | |
| CN113965313B (en) | Model training method, device, equipment and storage medium based on homomorphic encryption | |
| CN113569199B (en) | Lottery data processing method, system, device and readable storage medium | |
| CN107820677A (en) | Determine the method and its device, terminal of filter coefficient | |
| CN110428128A (en) | Performance data acquisition methods, device and storage medium | |
| CN110033092A (en) | Data label generation, model training, event recognition method and device | |
| CN109190757A (en) | Task processing method, device, equipment and computer readable storage medium | |
| CN112507104B (en) | Dialog system acquisition method, apparatus, storage medium and computer program product | |
| US20220387897A1 (en) | Systems and methods for video streaming analysis | |
| CN113230650A (en) | Data processing method and device and computer readable storage medium | |
| JP4535811B2 (en) | Data processing method and program | |
| CN107451920A (en) | Calculation server, dynamic income processing method and storage medium | |
| CN113128911A (en) | Online evaluation method and device for data resource value | |
| CN117786478B (en) | Multi-model-based user activity prediction method, system, equipment and medium |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |