CN108734330A - Data processing method and device - Google Patents
Data processing method and device Download PDFInfo
- Publication number
- CN108734330A CN108734330A CN201710272081.4A CN201710272081A CN108734330A CN 108734330 A CN108734330 A CN 108734330A CN 201710272081 A CN201710272081 A CN 201710272081A CN 108734330 A CN108734330 A CN 108734330A
- Authority
- CN
- China
- Prior art keywords
- data
- prediction
- prediction model
- algorithm
- model
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q10/00—Administration; Management
- G06Q10/04—Forecasting or optimisation specially adapted for administrative or management purposes, e.g. linear programming or "cutting stock problem"
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- 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
- G06Q30/00—Commerce
- G06Q30/02—Marketing; Price estimation or determination; Fundraising
- G06Q30/0201—Market modelling; Market analysis; Collecting market data
- G06Q30/0202—Market predictions or forecasting for commercial activities
Landscapes
- Business, Economics & Management (AREA)
- Engineering & Computer Science (AREA)
- Strategic Management (AREA)
- Development Economics (AREA)
- Economics (AREA)
- Finance (AREA)
- Accounting & Taxation (AREA)
- Entrepreneurship & Innovation (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Business, Economics & Management (AREA)
- General Physics & Mathematics (AREA)
- Marketing (AREA)
- Human Resources & Organizations (AREA)
- Game Theory and Decision Science (AREA)
- Operations Research (AREA)
- Quality & Reliability (AREA)
- Tourism & Hospitality (AREA)
- Data Mining & Analysis (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The present invention provides a kind of data processing method and device.The data processing method of the present invention includes the following steps:With machine learning method come training data cleaning rule to carry out data cleansing, and data cleansing judgement is carried out to prediction model training data using the data cleansing discrimination model trained;Selection participates in the prediction model of prediction model training operation from predictive model algorithm library;The specific prediction model of operation is trained to carry out arameter optimization to participating in prediction model in prediction model trains calculating process.
Description
Technical field
The present invention relates to computer realm more particularly to a kind of data processing method and device, electronic equipment and readable deposit
Storage media.
Background technology
Method for Sales Forecast is a kind of method according to commodity history sales volume quantitative forecast commodity future sales volume data, machine learning
It is widely used in Method for Sales Forecast, machine learning prediction model needs to carry out data before being trained a large amount of
Data cleansing work;Multiple prediction models, each prediction model are often selected to need individually to carry out tuning during prediction,
Then the prediction result of multiple prediction models is subjected to the comprehensive Method for Sales Forecast value last as commodity.
Generally, Method for Sales Forecast is carried out by machine learning and generally includes following steps:
Cleaning rule is formulated before training prediction model.
Each commodity are predicted using multiple prediction techniques, the prediction technique for then taking last-period forecast effect best
Prediction result as predicted value.
Wherein, when carrying out prediction model training, there are two types of methods for the fine tuning of model optimized parameter, and one is every one section
One suboptimum training parameter of time search, another method search for optimal training parameter before being each training.
In training data some data may due to the system failure, or due to cause specific can embody some spy
Sign, for example, it is apparent high or minimum.This partial data can make prediction model deviation occur in training process, and it is accurate to reduce prediction
Degree, it is therefore desirable to processing are filtered to this all partial data, i.e., so-called data cleansing.
In realizing process of the present invention, inventor has found that at least there are the following problems in the prior art:
First, data cleansing is a critically important job, generally requires artificially to formulate a large amount of cleaning rules, and clean
Rule needs at any time and business is adjusted, time-consuming and laborious.And many data cannot be cleaned in time, therefore can distort
Prediction model.
Secondly as predicted using multiple prediction techniques each commodity, then take last-period forecast effect best
Prediction technique prediction result as predicted value, this just needs all to carry out all prediction algorithms a time calculating, if number
Huger according to measuring, computing resource can become bottleneck.
In addition, the fine tuning of model optimized parameter is often also required to algorithm and is run multiple times when model training, to find most
Excellent parameter, this can equally sell a large amount of computing resources of consumption.
Invention content
In view of this, a kind of data processing method and device of offer of the embodiment of the present invention, electronic equipment and readable storage medium
Matter, thus, it is possible to extract previous prediction experience, (these experiences include how progress data screening, how to carry out model
How selection carries out arameter optimization to concrete model), it is used in prediction afterwards, to improve predictablity rate
While reduce calculation amount.
To achieve the above object, one side according to the ... of the embodiment of the present invention provides a kind of data processing method.
According to a kind of preferred embodiment of the present invention, data processing method of the invention includes the following steps:
With machine learning method come training data cleaning rule to carry out data cleansing, and it is clear using the data trained
It washes discrimination model and data cleansing judgement is carried out to prediction model training data;
Selection participates in the prediction model of prediction model training operation from predictive model algorithm library;
The specific prediction model of operation is trained to carry out parameter to participating in prediction model in prediction model trains calculating process
Tuning.
In the present invention, it is preferred that in order to which training data cleans discrimination model, provide and clean library, in the cleaning library
Store the feature of abnormal data, wherein the source of abnormal data includes at least one of the following:History abnormal data,
The newfound abnormal data of business side's feedback, the data for being unanimously judged as by all data cleansing discrimination models abnormal data,
Or be judged as abnormal data through one or more data cleansing discrimination models and be confirmed as abnormal data after manual identified can
Doubt data.
It may further be preferable that using the data cleansing discrimination model that trains to prediction model training data into line number
When judging according to cleaning, the feature that is extracted to prediction model training data using the data cleansing discrimination model trained is into line number
Judge according to cleaning, wherein put the abnormal data that all data cleansing discrimination models are adjudicated into suspicious data library, wherein:
If all data cleansing discrimination model court verdicts are abnormal data, this data is directly added into cleaning
Library, the empirical data as the training of next data cleansing discrimination model;
If multiple data cleansing discrimination model court verdicts are inconsistent, selected respective counts after carrying out manual identified
It library or is not processed according to cleaning is added.
According to a kind of preferred embodiment of the present invention, the data cleansing discrimination model that can be used include in the following terms extremely
One item missing:SVM, random forest, logistic regression, Bayes classifier.Certainly other suitable disaggregated models can also be used.
It is gathered around from predictive model algorithm library according to a kind of preferred embodiment of the present invention when carrying out prediction model selection
N1 minimum prediction algorithm of prediction error rate is selected in the N number of predictive model algorithm having and participates in prediction model training operation, so
Remaining N2 prediction algorithm is participated in into prediction model by probability afterwards and trains operation.
It may further be preferable that participating in the select probability Pi of the prediction algorithm of prediction model training operation by probability by as follows
Formula obtains:
Pi=(1/ Δ i)/(sumN2 (1/ Δ i)),
Wherein, i represents the number of candidate algorithm, i=1,2 ... ..., N2;Δ i is the Algorithm Error of i-th of candidate algorithm
Rate;SumN2 () is summing function, by the 1/ Δ i summations of selection coefficient of N2 candidate algorithm.
Furthermore it is preferred that being predicted by selected prediction algorithm, of the quantity with selected prediction algorithm is obtained
The corresponding predicted values of number select in the prediction algorithm selected by these minimum algorithm of average forecasting error in scheduled time slot
Prediction result is exported as prediction, after the generation of true sales volume, according to the selected predicted value for carrying out the algorithm of prediction output
Comparison with true sales volume updates modelling effect.
According to a kind of preferred embodiment of the present invention, the specific prediction mould of each of operation is trained for prediction model is participated in
Type is taken out from prediction model training parameter library for optimized parameter known to the prediction model, then to known to being taken out
Optimized parameter is soundd out at random.
If parameter testing space is huger, and optimized parameter changes at any time, if training is all right every time
Parameter space is traversed, and needs to expend more computing resource, it is possible to which each only selectivity looks for some most possible
Trial operation is carried out as the parameter of optimized parameter, sees operational effect, to decide whether to carry out parameter update.That is it adopts
Take certain Probe Strategy.
Thus, the known preferred parameter preferably taken out from prediction model training parameter library in the present invention pertains only to most have
It is likely to become the parameter of optimized parameter.
Furthermore it is preferred that in random sound out, for each taken out known preferred parameter, according to adjusting step-length and adjusting
Step number generates new probe parameters.
It may further be preferable that by probe parameters assign it is corresponding be selected prediction model and combine once purged prediction mould
Type training data carries out prediction model training, obtains different errors, by the corresponding parameter value update of minimal error to prediction mould
Type training parameter library.
Above-mentioned purpose to realize the present invention, another aspect according to the ... of the embodiment of the present invention provide a kind of data processing
Device.
According to a kind of preferred embodiment of the present invention, data processing equipment of the invention includes following modules:
Data cleansing rule training module, the data cleansing rule training module can train number with machine learning method
According to cleaning rule to carry out data cleansing, and using the data cleansing discrimination model trained to prediction model training data into
Row data cleansing judges;
Prediction model selecting module, the prediction model selecting module can select to participate in from predictive model algorithm library pre-
Survey the prediction model of model training operation;
Prediction model parameters tuning module, the prediction model parameters tuning module can train operation in prediction model
The specific prediction model of operation is trained to carry out arameter optimization to participating in prediction model in journey.
In the present invention, it is preferred that data cleansing rule training module cleans discrimination model for training data, provide
Have a cleaning library, the feature of abnormal data stored in the cleaning library, wherein the source of abnormal data include in the following terms extremely
One item missing:History abnormal data, the newfound abnormal data of business side's feedback are unanimously sentenced by all data cleansing discrimination models
Break as the data of abnormal data or is judged as abnormal data and after manual identified through one or more data cleansing discrimination models
It is confirmed as the suspicious data of abnormal data.
It may further be preferable that using the data cleansing discrimination model that trains to prediction model training data into line number
When judging according to cleaning, data cleansing rule training module can utilize the data cleansing discrimination model trained to train prediction model
The feature that data pick-up goes out carries out data cleansing judgement, and is provided with suspicious data library, wherein can adjudicate all data cleansings
The abnormal data that model is adjudicated puts suspicious data library into, wherein:
If all data cleansing discrimination model court verdicts are abnormal data, this data is directly added into cleaning
Library, the empirical data as the training of next data cleansing discrimination model;
If multiple data cleansing discrimination model court verdicts are inconsistent, selected respective counts after carrying out manual identified
It library or is not processed according to cleaning is added.
According to a kind of preferred embodiment of the present invention, the data cleansing discrimination model that can be used include in the following terms extremely
One item missing:SVM, random forest, logistic regression, Bayes classifier.Certainly other suitable disaggregated models can also be used.
According to a kind of preferred embodiment of the present invention, when carrying out prediction model selection, prediction model selecting module can be from
N1 minimum prediction algorithm of prediction error rate is selected in N number of predictive model algorithm that predictive model algorithm library is possessed to participate in advance
Model training operation is surveyed, remaining N2 prediction algorithm is then participated in prediction model by probability trains operation.
It may further be preferable that participating in the select probability Pi of the prediction algorithm of prediction model training operation by probability by as follows
Formula obtains:
Pi=(1/ Δ i)/(sumN2 (1/ Δ i)),
Wherein, i represents the number of candidate algorithm, i=1,2 ... ..., N2;Δ i is the Algorithm Error of i-th of candidate algorithm
Rate;SumN2 () is summing function, by the 1/ Δ i summations of selection coefficient of N2 candidate algorithm.
In addition, prediction model selecting module can preferably be predicted by selected prediction algorithm, quantity and institute are obtained
The corresponding predicted value of number for the prediction algorithm selected selects average pre- in scheduled time slot in the prediction algorithm selected by these
The prediction result for surveying the minimum algorithm of error is exported as prediction, after the generation of true sales volume, according to selected for being predicted
The comparison of the predicted value of the algorithm of output and true sales volume updates modelling effect.
According to a kind of preferred embodiment of the present invention, prediction model parameters tuning module, which can be directed to, participates in prediction model training
The specific prediction model of each of operation is taken out from prediction model training parameter library for optimal ginseng known to the prediction model
Number, then sounds out the known preferred parameter taken out at random.
If parameter testing space is huger, and optimized parameter changes at any time, if training is all right every time
Parameter space is traversed, and needs to expend more computing resource, it is possible to which each only selectivity looks for some most possible
Trial operation is carried out as the parameter of optimized parameter, sees operational effect, to decide whether to carry out parameter update.That is it adopts
Take certain Probe Strategy.
Thus, the known preferred parameter preferably taken out from prediction model training parameter library in embodiments of the present invention is only
It is related to most possibly becoming the parameter of optimized parameter.
Furthermore it is preferred that in random sound out, for each taken out known preferred parameter, according to adjusting step-length and adjusting
Step number generates new probe parameters.
It may further be preferable that probe parameters can be assigned and corresponding be selected prediction model by prediction model parameters tuning module
And once purged prediction model training data is combined to carry out prediction model training, different errors is obtained, by minimal error pair
The parameter value answered is updated to prediction model training parameter library.
Above-mentioned purpose to realize the present invention, it is according to the ... of the embodiment of the present invention in another aspect, providing a kind of can execute number
According to the electronic equipment of processing method.
The a kind of electronic equipment of the embodiment of the present invention includes:
One or more processors;
Storage device, for storing one or more programs,
When one or more of programs are executed by one or more of processors so that one or more of processing
Device realizes the data processing method of the embodiment of the present invention.
Above-mentioned purpose to realize the present invention, another aspect according to the ... of the embodiment of the present invention, providing a kind of computer can
Read storage medium.
A kind of computer-readable recording medium storage of the embodiment of the present invention has computer program, and described program is by processor
The data processing method of the embodiment of the present invention is realized when execution.
One embodiment in foregoing invention has the following advantages that or advantageous effect:
Because the method using machine learning replaces the method to lay down a regulation by hand to carry out data cleansing, overcome
Need artificially to formulate a large amount of cleaning rules and cleaning rule need at any time and business be adjusted thus time-consuming and laborious technology
Problem, and then reach and automatically record cleaning data and empirically data and enhance the technique effect of cleaning performance;
Because determining be combined using which algorithm according to Probe Strategy, adjusts and sound out further according to forecast result
Strategy is as next exploration experience, so computing resource bottleneck problem when overcoming traversal prediction algorithm, saves computing resource;
Because with the optimal training parameter of certain resource stochastic searching prediction model, more new historical in each training process
Optimized parameter is saved so the fine tuning for overcoming model optimized parameter when model training can sell the problem of consuming a large amount of computing resources
Computing resource simultaneously makes the prediction model most of time operate in optimum state.
It can be seen that the present invention uses machine learning using the prediction experience in the method extraction Method for Sales Forecast for souning out accumulation
Method replace the method that lays down a regulation by hand to carry out data cleansing, judge which data to training pattern be abnormal data or
Extreme value data are filtered processing when prediction model is trained, and empirically data enhance cleaning performance to record cleaning data;Root
It determines be combined using which algorithm according to Probe Strategy, to save computing resource, adjusts and try further according to forecast result
Strategy is visited as next exploration experience;Joined with the optimal training of certain resource stochastic searching prediction model in each training process
Number, more new historical optimized parameter make prediction model most of time operate in optimum state.The Method for Sales Forecast system of the present invention can
It the more runs the more intelligent in a manner of by experience accumulation, environmental change can be adapted to automatically, ensure higher predictablity rate.
Further effect possessed by above-mentioned non-usual optional mode adds hereinafter in conjunction with specific implementation mode
With explanation.
Description of the drawings
Attached drawing does not constitute inappropriate limitation of the present invention for more fully understanding the present invention.Wherein:
Fig. 1 is a kind of method flow diagram of embodiment of data processing method according to the present invention;
Fig. 2 is a kind of schematic diagram of the main modular of embodiment of data processing equipment according to the present invention;
Fig. 3 is adapted for the structural schematic diagram of the computer system of the electronic equipment for realizing the embodiment of the present invention.
Specific implementation mode
It explains to the exemplary embodiment of the present invention below in conjunction with attached drawing, including the various of the embodiment of the present invention
Details should think them only exemplary to help understanding.Therefore, those of ordinary skill in the art should recognize
It arrives, various changes and modifications can be made to the embodiments described herein, without departing from scope and spirit of the present invention.Together
The description to known function and structure is omitted for clarity and conciseness in sample in following description.
Fig. 1 is a kind of method flow diagram of preferred embodiment of data processing method according to the present invention.In the embodiment
In, the data processing method in Method for Sales Forecast of the invention includes the following steps:
1) preparation model training data includes mainly:Newfound abnormal data, history normal data, training data.
Wherein:
Newfound abnormal data is mainly derived from the feedback of business side;
Normal sales volume data in January before history normal data refers to, wherein being free of abnormal data;
Training data is mainly the sales volume data in the previous day to preceding January, wherein may the abnormal number in need that be filtered
According to.
2) feature extraction is carried out to model training data.Feature includes mainly:First M weeks average sales volume, preceding M days sales volumes, when
Its sales volume, unit price, inventory status, the date, festivals or holidays, from festivals or holidays number of days, sales promotion information, etc..Feedback from business side
Abnormal data is in the deposit cleaning library after feature extraction.
3) balanced sample is mainly used for training data washing moulding or says discrimination model.It is adopted to be balanced
Sample, the feature for taking out problematic data abnormal data in other words from cleaning library take as positive sample from normal historical data
Go out the feature of normal historical data as negative sample, determines positive and negative sample proportion (such as 1:3), wherein can be according to training
Discrimination model test data is made decisions after accuracy rate and recall rate adjust the positive and negative sample proportion of balanced sample, to seek
Seek rational sampling balance.
4) feature selecting, for selecting to participate in the feature of discrimination model training.It is preferably examined in the present embodiment using card side
Proved recipe method is advantageously selected for carrying out the feature of positive and negative example classification (i.e. positive and negative sample classification).It is of course also possible to use other alternatives
Method replaces Chi-square method, such as:Information gain method (mutual information, dropout) etc..
5) discrimination model is trained, and discrimination model is cleaned according to data training data of the balanced sample after feature selecting.
Multiple disaggregated models can be used in data cleansing discrimination model:Such as SVM, random forest, logistic regression, Bayes classifier, etc..Root
Carry out training data cleaning judgement mould according to the result (such as accuracy rate and recall rate) after data cleansing discrimination model testing results data
Type and the parameter for adjusting data cleansing discrimination model obtain suitable data cleansing discrimination model in other words to train.
6) abnormal judgement, that is, carry out data cleansing judgement.At this point, the characteristic that prediction model training data is extracted
Input the data cleansing discrimination model that is obtained after discrimination model is trained, obtain training sample whether be abnormal data judgement knot
Fruit.Wherein, multiple data cleansing discrimination models can obtain multiple court verdicts.
7) filtration treatment, including:The abnormal data that all data cleansing discrimination models are adjudicated is put into suspicious
Database;If all judgement grader court verdicts are abnormal data, this data is directly added into cleaning library, as under
The empirical data of secondary cleaning training;If multiple judgement grader court verdicts are inconsistent, add again after manual identified can be carried out
Into cleaning library or it is not processed.
It is as follows that library data structure is cleaned in the present embodiment:
Field name | Data type | Explanation |
DCID | string | Predict unit ID, such as warehouse number |
SKUID | string | Commodity ID |
DATE | string | Date |
FEATURE | string | Data characteristics |
RAWDATA | string | Original training data (before feature extraction) |
8) prediction model selection is carried out.Wherein, it is integrated from prediction algorithm and takes out prediction algorithm and prediction error rate in library
(mapd), prediction error rate here is the prediction error rate mean value for all commodity, to weigh the accurate of prediction algorithm
Degree.Such as it can be calculated according to following formula:
Mapd=abs (predicted value-actual value)/actual value,
Wherein, abs () is the function that takes absolute value.Mapd shows that more greatly the accuracy of prediction algorithm is lower.
It is assumed that prediction algorithm, which integrates, shares algorithm quantity for N, according to the sufficient situation of computing resource, from algorithms library in library
It selects N1 minimum algorithm of prediction error rate and participates in operation, then by remaining N2 algorithm (can be described as probability candidate algorithm)
Operation is participated in by probability, wherein N1+N2=N.The select probability Pi of probability candidate algorithm can for example be obtained by following formula:
Pi=(1/ Δ i)/(sumN2 (1/ Δ i)),
Wherein, i represents the number of candidate algorithm, i=1,2 ... ..., N2;Δ i is the Algorithm Error of i-th of candidate algorithm
Rate;SumN2 () is summing function, by the 1/ Δ i summations of selection coefficient of N2 candidate algorithm.
The above is a kind of specific embodiment of " model selection " frame in flow chart.The determination of select probability should make
Model smaller mapd obtains more selected chances, so that it is selected probability in other words bigger;And mapd is higher, then corresponds to mould
Type is selected that chance is smaller, i.e., it is smaller to be selected probability for it.Therefore mapd theoretically can be used for example and carry out computational algorithm participation fortune
The probability of calculation.
Prediction algorithm library data structure is as follows in the present embodiment:
Field name | Data type | Explanation |
ALGOID | string | Algorithm ID |
MAPD | float | Algorithm Error rate Δ i |
MODIFYDATE | date | The Last modification Algorithm Error rate date |
ONLINEDATE | date | Algorithm is reached the standard grade the date |
RUNNINGTIMES | int | Algorithm participates in calculation times |
9) optimized parameter is chosen, the specific prediction model selected by model in library is integrated from prediction algorithm to each
Algorithm is taken out from prediction model training parameter library for optimized parameter known to the model.These known optimized parameters be through
The optimized parameter tested, and may not be optimized parameter to current environment, for example same prediction model is in different month meetings
There is different optimized parameters.These parameters include the regular coefficient of data, classification tree regression tree depth, Missing Data Filling, loss
Type function, penalty coefficient, iterations etc..Then selected known preferred parameter is soundd out at random.
Prediction model training parameter library data structure is as follows in the present embodiment:
Field name | Data type | Explanation |
ID | string | Parameter ID |
NAME | string | Parameter name |
CATE | string | Merchandise classification corresponding to parameter |
OPTIMAL | float | The current optimal value of parameter |
TYPE | int | Sound out type (unique step, index step-length, arbitrary width) |
STEP | float | Parameter regulation step-length |
STEPCOUNT | int | Parameter regulation step number |
In random sound out, for each parameter, new probe parameters can be generated according to adjusting step-length and adjusting step number.It lifts
For example, for example parameter optimal value is 1.5, step-length 0.1, step number 1, then can generate 1.4 for unique step (is equal to parameter
Optimal value subtracts to be grown step by step, i.e. 1.5-0.1=1.4), 1.6 (equal to parameter optimal value plus step by step growing, i.e. 1.5+0.1=
1.6) two probe parameters;(0.1) two 1.5+exp (0.1), 1.5-exp probe parameters can be generated for index step-length,
In, exp () is exponential function;For arbitrary width, can generate with 1.5 as mean value, using 0.1*1 as two random numbers of variance
As probe parameters.
10) cross validation, by the probe parameters that selected optimized parameter obtains after random sound out assign it is corresponding be selected it is pre-
It surveys model and once purged prediction model training data is combined to carry out prediction model training.These different probe parameters in this way
Different training errors can be obtained, by the corresponding parameter value update of minimum training error to prediction model training parameter library.Error
Including training error and true error.Training error be, for example, use probe parameters prediction model historical forecast value with go through
The comparison result of the true sales volume of history;True error be, for example, use probe parameters prediction model predicted value and future it is true
The comparison result of sales volume.It theoretically, can also be after obtaining the following true sales volume, more by the corresponding parameter value of minimum true error
Newly arrive prediction model training parameter library.
11) prediction output, can be a randomly selected by the lower algorithm of N1 prediction error rate and N2 ' for each commodity
Algorithm is predicted (wherein, N2 '≤N2), obtains a predicted values of N1+N2 ', select in these prediction algorithms predetermined period (such as
Nearly one week) prediction result of the minimum algorithm of interior average forecasting error exports as prediction.
12) modelling effect updates, after the generation of true sales volume, according to the selected prediction for carrying out the algorithm of prediction output
Value and true sales volume update modelling effect by (true sales volume-prediction sales volume)/true sales volume, are equivalent to and are updated to mapd.
The present invention is using the prediction experience in the method extraction Method for Sales Forecast for souning out accumulation as a result, with the side of machine learning
Method replaces the method to lay down a regulation by hand to carry out data cleansing, judges which data is abnormal data or extreme value to training pattern
Data are filtered processing when prediction model is trained, and empirically data enhance cleaning performance to record cleaning data;According to examination
It visits strategy decision to be combined using which algorithm, to save computing resource, is adjusted further according to forecast result and sound out plan
Slightly it is next exploration experience;With the optimal training parameter of certain resource stochastic searching prediction model in each training process,
More new historical optimized parameter makes prediction model most of time operate in optimum state.
Fig. 2 is a kind of schematic diagram of the main modular of preferred embodiment of data processing equipment according to the present invention.
As shown in Fig. 2, in one preferred embodiment of the invention, data processing equipment 10 includes mainly:
Data cleansing rule training module 101, data cleansing rule training module 101 can be trained with machine learning method
Data cleansing rule utilizes the data cleansing discrimination model trained to prediction model training data to carry out data cleansing
Carry out data cleansing judgement;
Prediction model selecting module 102, prediction model selecting module 102 can select to participate in from predictive model algorithm library
Prediction model trains the prediction model of operation;
Prediction model parameters tuning module 103, prediction model parameters tuning module energy 103 train operation in prediction model
The specific prediction model of operation is trained to carry out arameter optimization to participating in prediction model in the process.
In the other unshowned embodiments of the present invention, according to specific computing resource, data processing equipment 10 can have
There is one or more of above three module 101~103.
Data cleansing rule training module 101 cleans discrimination model for training data, is provided with cleaning library, cleans in library
Store the feature of abnormal data, wherein the source of abnormal data includes at least one of the following:History abnormal data,
The newfound abnormal data of business side's feedback, the data for being unanimously judged as by all data cleansing discrimination models abnormal data,
Or be judged as abnormal data through one or more data cleansing discrimination models and be confirmed as abnormal data after manual identified can
Doubt data.
For this purpose, carrying out data cleansing judgement to prediction model training data using the data cleansing discrimination model trained
When, data cleansing rule training module 101 can also utilize the data cleansing discrimination model trained to prediction model training data
The feature extracted carries out data cleansing judgement, and is provided with suspicious data library, wherein can be by all data cleansing discrimination models
It adjudicates obtained abnormal data and puts suspicious data library into, wherein:
If all data cleansing discrimination model court verdicts are abnormal data, this data is directly added into cleaning
Library, the empirical data as the training of next data cleansing discrimination model;
If multiple data cleansing discrimination model court verdicts are inconsistent, selected respective counts after carrying out manual identified
It library or is not processed according to cleaning is added.
When carrying out prediction model selection, prediction model selecting module 102 can be possessed N number of from predictive model algorithm library
N1 minimum prediction algorithm of prediction error rate is selected in predictive model algorithm and participates in prediction model training operation, then will be left
N2 prediction algorithm by probability participate in prediction model train operation.
Wherein, participating in prediction model by probability trains the select probability Pi of the prediction algorithm of operation to be obtained by following formula:
Pi=(1/ Δ i)/(sumN2 (1/ Δ i)),
Wherein, i represents the number of candidate algorithm, i=1,2 ... ..., N2;Δ i is the Algorithm Error of i-th of candidate algorithm
Rate;SumN2 () is summing function, by the 1/ Δ i summations of selection coefficient of N2 candidate algorithm.
After choosing prediction algorithm, prediction model selecting module 102 can be predicted by selected prediction algorithm,
Quantity predicted value corresponding with the number of selected prediction algorithm is obtained, is selected in the prediction algorithm selected by these predetermined
The prediction result of the minimum algorithm of average forecasting error is exported as prediction in period, after the generation of true sales volume, according to selected
The comparison of predicted value and the true sales volume of algorithm for carrying out prediction output updates modelling effect.
Prediction model parameters tuning module 103, which can be directed to, participates in the specific prediction model of each of prediction model training operation,
It is taken out from prediction model training parameter library and is directed to optimized parameter known to the prediction model, then the known preferred to being taken out
Parameter is soundd out at random.Specific random the step of souning out, may refer to corresponding content in above method step.
Prediction model parameters tuning module 103, which in addition can also assign probe parameters, corresponding to be selected prediction model and combines
Once purged prediction model training data carries out prediction model training, different errors is obtained, by the corresponding ginseng of minimal error
Numerical value is updated to prediction model training parameter library.
According to an embodiment of the invention, the present invention also provides a kind of electronic equipment and a kind of readable storage medium storing program for executing.
The present invention electronic equipment include:
One or more processors;
Storage device, for storing one or more programs,
When one or more of programs are executed by one or more of processors so that one or more of processing
Device realizes the data processing method of the embodiment of the present invention.
The computer-readable recording medium storage of the present invention has computer program, is realized when described program is executed by processor
The data processing method of the embodiment of the present invention.
Below with reference to the knot of the computer system 300 of Fig. 3 electronic equipments for illustrating to be suitable for being used for realizing the embodiment of the present invention
Structure schematic diagram.Electronic equipment shown in Fig. 3 is only an example, should not be to the function and use scope band of the embodiment of the present invention
Carry out any restrictions.
As shown in figure 3, computer system 300 includes central processing unit (CPU) 301, it can be read-only according to being stored in
Program in memory (ROM) 302 or be loaded into the program in random access storage device (RAM) 303 from storage section 308 and
Execute various actions appropriate and processing.In RAM 303, also it is stored with system 300 and operates required various programs and data.
CPU 301, ROM 302 and RAM 303 are connected with each other by bus 304.Input/output (I/O) interface 305 is also connected to always
Line 304.
It is connected to I/O interfaces 305 with lower component:Importation 306 including keyboard, mouse etc.;It is penetrated including such as cathode
The output par, c 307 of spool (CRT), liquid crystal display (LCD) etc. and loud speaker etc.;Storage section 308 including hard disk etc.;
And the communications portion 309 of the network interface card including LAN card, modem etc..Communications portion 309 via such as because
The network of spy's net executes communication process.Driver 310 is also according to needing to be connected to I/O interfaces 305.Detachable media 311, such as
Disk, CD, magneto-optic disk, semiconductor memory etc. are mounted on driver 310, as needed in order to be read from thereon
Computer program be mounted into storage section 308 as needed.
Particularly, it according to embodiment disclosed by the invention, may be implemented as counting above with reference to the process of flow chart description
Calculation machine software program.For example, embodiment disclosed by the invention includes a kind of computer program product comprising be carried on computer
Computer program on readable medium, the computer program include the program code for method shown in execution flow chart.?
In such embodiment, which can be downloaded and installed by communications portion 309 from network, and/or from can
Medium 311 is dismantled to be mounted.When the computer program is executed by central processing unit (CPU) 301, system of the invention is executed
The above-mentioned function of middle restriction.
It should be noted that computer-readable medium shown in the present invention can be computer-readable signal media or meter
Calculation machine readable storage medium storing program for executing either the two arbitrarily combines.Computer readable storage medium for example can be --- but not
Be limited to --- electricity, magnetic, optical, electromagnetic, infrared ray or semiconductor system, device or device, or arbitrary above combination.Meter
The more specific example of calculation machine readable storage medium storing program for executing can include but is not limited to:Electrical connection with one or more conducting wires, just
It takes formula computer disk, hard disk, random access storage device (RAM), read-only memory (ROM), erasable type and may be programmed read-only storage
Device (EPROM or flash memory), optical fiber, portable compact disc read-only memory (CD-ROM), light storage device, magnetic memory device,
Or above-mentioned any appropriate combination.In the present invention, can be any include computer readable storage medium or storage journey
The tangible medium of sequence, the program can be commanded the either device use or in connection of execution system, device.And at this
In invention, computer-readable signal media may include in a base band or as the data-signal that a carrier wave part is propagated,
Wherein carry computer-readable program code.Diversified forms may be used in the data-signal of this propagation, including but unlimited
In electromagnetic signal, optical signal or above-mentioned any appropriate combination.Computer-readable signal media can also be that computer can
Any computer-readable medium other than storage medium is read, which can send, propagates or transmit and be used for
By instruction execution system, device either device use or program in connection.Include on computer-readable medium
Program code can transmit with any suitable medium, including but not limited to:Wirelessly, electric wire, optical cable, RF etc. or above-mentioned
Any appropriate combination.
Flow chart in attached drawing and block diagram, it is illustrated that according to the system of various embodiments of the invention, method and computer journey
The architecture, function and operation in the cards of sequence product.In this regard, each box in flowchart or block diagram can generation
A part for a part for one module, program segment, or code of table, above-mentioned module, program segment, or code includes one or more
Executable instruction for implementing the specified logical function.It should also be noted that in some implementations as replacements, institute in box
The function of mark can also occur in a different order than that indicated in the drawings.For example, two boxes succeedingly indicated are practical
On can be basically executed in parallel, they can also be executed in the opposite order sometimes, this is depended on the functions involved.Also it wants
It is noted that the combination of each box in block diagram or flow chart and the box in block diagram or flow chart, can use and execute rule
The dedicated hardware based systems of fixed functions or operations is realized, or can use the group of specialized hardware and computer instruction
It closes to realize.
Being described in module involved in the embodiment of the present invention can be realized by way of software, can also be by hard
The mode of part is realized.Described module can also be arranged in the processor, for example, can be described as:A kind of processor packet
Include data cleansing rule training module, prediction model selecting module, prediction model parameters tuning module.Wherein, these modules
Title does not constitute the restriction to the module itself under certain conditions, for example, data cleansing rule training module can also quilt
It is described as " with machine learning method come training data cleaning rule to carry out the module of data cleansing ".
As on the other hand, the present invention also provides a kind of computer-readable medium, which can be
Included in equipment described in above-described embodiment;Can also be individualism, and without be incorporated the equipment in.Above-mentioned calculating
Machine readable medium carries one or more program, when said one or multiple programs are executed by an equipment so that
The equipment executes the method including at least following steps:
With machine learning method come training data cleaning rule to carry out data cleansing, and it is clear using the data trained
It washes discrimination model and data cleansing judgement is carried out to prediction model training data;
Selection participates in the prediction model of prediction model training operation from predictive model algorithm library;
The specific prediction model of operation is trained to carry out parameter to participating in prediction model in prediction model trains calculating process
Tuning.
The said goods can perform the method that the embodiment of the present invention is provided, and has the corresponding function module of execution method and has
Beneficial effect.The not technical detail of detailed description in the present embodiment, reference can be made to the method that the embodiment of the present invention is provided.
Technical solution according to the ... of the embodiment of the present invention is passed through using the prediction in the method extraction Method for Sales Forecast for souning out accumulation
It tests, the method to lay down a regulation by hand is replaced to carry out data cleansing with the method for machine learning, judge which data to training mould
Type is abnormal data or extreme value data, and processing is filtered when prediction model is trained, and record cleans data empirically data
Enhance cleaning performance;Determine be combined using which algorithm according to Probe Strategy, to save computing resource, further according to reality
Prediction effect adjusts Probe Strategy as next exploration experience;In each training process mould is predicted with certain resource stochastic searching
The optimal training parameter of type, more new historical optimized parameter, make prediction model most of time operate in optimum state.The present invention's
Method for Sales Forecast system the more can run the more intelligent by way of experience accumulation, can adapt to environmental change automatically, ensure higher
Predictablity rate.
Above-mentioned specific implementation mode, does not constitute limiting the scope of the invention.Those skilled in the art should be bright
It is white, design requirement and other factors are depended on, various modifications, combination, sub-portfolio and replacement can occur.It is any
Modifications, equivalent substitutions and improvements made by within the spirit and principles in the present invention etc., should be included in the scope of the present invention
Within.
Claims (24)
1. a kind of data processing method, which is characterized in that the data processing method includes:
It is adjudicated with the data cleansing for carrying out data cleansing, and utilizing to train come training data cleaning rule with machine learning method
Prediction model training data carries out data cleansing judgement;
Selection participates in the prediction model of prediction model training operation from predictive model algorithm library;
The specific prediction model of operation is trained to carry out arameter optimization to participating in prediction model in prediction model trains calculating process.
2. according to the method described in claim 1, it is characterized in that, in order to which training data cleans discrimination model, cleaning library is provided,
The feature of abnormal data is stored in the cleaning library, wherein the source of abnormal data includes at least one of the following:It goes through
History abnormal data, the newfound abnormal data of business side's feedback are unanimously judged as exception by all data cleansing discrimination models
The data of data are judged as abnormal data through one or more data cleansing discrimination models and are confirmed as after manual identified different
The suspicious data of regular data.
3. according to the method described in claim 2, it is characterized in that, described utilize the data cleansing discrimination model trained to pre-
Surveying the progress data cleansing judgement of model training data includes:Prediction model is trained using the data cleansing discrimination model trained
The feature that data pick-up goes out carries out data cleansing judgement, wherein the abnormal number for adjudicating all data cleansing discrimination models
According to putting suspicious data library into, wherein:
If all data cleansing discrimination model court verdicts are abnormal data, this data is directly added into cleaning library, is made
For the empirical data of next data cleansing discrimination model training;
If multiple data cleansing discrimination model court verdicts are inconsistent, select to add corresponding data after carrying out manual identified
Into cleaning library or it is not processed.
4. according to the method in any one of claims 1 to 3, which is characterized in that the data cleansing discrimination model that can be used
Including at least one of the following:SVM, random forest, logistic regression, Bayes classifier.
5. according to the method in any one of claims 1 to 3, which is characterized in that when carrying out prediction model selection, from pre-
It surveys and selects the minimum N1 prediction algorithm participation prediction of prediction error rate in N number of predictive model algorithm that model algorithm library is possessed
Then remaining N2 prediction algorithm is participated in prediction model by probability and trains operation by model training operation.
6. according to the method described in claim 5, it is characterized in that, the probability P i is obtained by following formula:
Pi=(1/ Δ i)/(sumN2 (1/ Δ i)),
Wherein, i represents the number of candidate algorithm, i=1,2 ... ..., N2;Δ i is the Algorithm Error rate of i-th of candidate algorithm;
SumN2 () is summing function, by the 1/ Δ i summations of selection coefficient of N2 candidate algorithm.
7. according to the method described in claim 5, it is characterized in that, predicted by selected prediction algorithm, counted
Amount predicted value corresponding with the number of selected prediction algorithm, selects in the prediction algorithm selected by these in scheduled time slot
The prediction result of the minimum algorithm of average forecasting error is exported as prediction, after the generation of true sales volume, according to it is selected for into
The predicted value of the algorithm of row prediction output updates modelling effect with the comparison of true sales volume.
8. according to the method in any one of claims 1 to 3, which is characterized in that train operation for prediction model is participated in
Each of specific prediction model, taken out for optimized parameter known to the prediction model, so from prediction model training parameter library
The known preferred parameter taken out is soundd out at random afterwards.
9. according to the method described in claim 8, it is characterized in that, the known preferred parameter taken out pertain only to most possibly at
For the parameter of optimized parameter.
10. according to the method described in claim 8, it is characterized in that, in random sound out, for it is each it is taken out known to most
Excellent parameter generates new probe parameters according to adjusting step-length and adjusting step number.
11. according to the method described in claim 10, corresponding being selected prediction model simultaneously it is characterized in that, probe parameters are assigned
Prediction model training is carried out in conjunction with once purged prediction model training data, different errors is obtained, minimal error is corresponded to
Parameter value update to prediction model training parameter library.
12. a kind of data processing equipment, which is characterized in that the data processing equipment includes:
Data cleansing rule training module, the data cleansing rule training module can be cleaned with machine learning method come training data
Rule carries out data using the data cleansing discrimination model trained to carry out data cleansing to prediction model training data
Cleaning judges;
Prediction model selecting module, the prediction model selecting module can select to participate in prediction model from predictive model algorithm library
The prediction model of training operation;
Prediction model parameters tuning module, the prediction model parameters tuning module can be right in prediction model trains calculating process
The specific prediction model for participating in prediction model training operation carries out arameter optimization.
13. device according to claim 12, which is characterized in that the data cleansing rule training module is in order to train number
According to cleaning discrimination model, it is provided with cleaning library, the feature of abnormal data is stored in the cleaning library, wherein abnormal data comes
Source includes at least one of the following:History abnormal data, business side feedback newfound abnormal data, by all numbers
Unanimously it is judged as the data of abnormal data according to cleaning discrimination model or is judged as through one or more data cleansing discrimination models different
Regular data and the suspicious data that abnormal data is confirmed as after manual identified.
14. device according to claim 13, which is characterized in that in the data cleansing discrimination model that utilization trains to pre-
When surveying the progress data cleansing judgement of model training data, the data cleansing rule training module can utilize the data trained clear
It washes discrimination model and data cleansing judgement is carried out to the feature that prediction model training data extracts, and be provided with suspicious data library,
Wherein, the abnormal data that all data cleansing discrimination models are adjudicated can be put into suspicious data library, wherein:
If all data cleansing discrimination model court verdicts are abnormal data, this data is directly added into cleaning library, is made
For the empirical data of next data cleansing discrimination model training;
If multiple data cleansing discrimination model court verdicts are inconsistent, select to add corresponding data after carrying out manual identified
Into cleaning library or it is not processed.
15. the device according to any one of claim 12 to 14, which is characterized in that the data cleansing judgement mould that can be used
Type includes at least one of the following:SVM, random forest, logistic regression, Bayes classifier.
16. the device according to any one of claim 12 to 14, which is characterized in that when carrying out prediction model selection,
Prediction error rate is selected in N number of predictive model algorithm that the prediction model selecting module can be possessed from predictive model algorithm library
N1 minimum prediction algorithm participates in prediction model and trains operation, and remaining N2 prediction algorithm is then participated in prediction by probability
Model training operation.
17. device according to claim 16, which is characterized in that the probability P i is obtained by following formula:
Pi=(1/ Δ i)/(sumN2 (1/ Δ i)),
Wherein, i represents the number of candidate algorithm, i=1,2 ... ..., N2;Δ i is the Algorithm Error rate of i-th of candidate algorithm;
SumN2 () is summing function, by the 1/ Δ i summations of selection coefficient of N2 candidate algorithm.
18. device according to claim 16, which is characterized in that the prediction model selecting module can be by selected
Prediction algorithm is predicted, is obtained quantity predicted value corresponding with the number of selected prediction algorithm, is selected selected by these
Prediction algorithm in scheduled time slot the minimum algorithm of average forecasting error prediction result as prediction export, wait for really selling
After amount generates, imitated come more new model with the comparison of true sales volume according to selected for carrying out the predicted value of the algorithm of prediction output
Fruit.
19. the device according to any one of claim 12 to 14, which is characterized in that the prediction model parameters tuning mould
Block, which can be directed to, participates in the specific prediction model of each of prediction model training operation, takes out and is directed to from prediction model training parameter library
Then optimized parameter known to the prediction model is soundd out the known preferred parameter taken out at random.
20. device according to claim 19, which is characterized in that the known preferred parameter taken out pertains only to most possibly
As the parameter of optimized parameter.
21. device according to claim 19, which is characterized in that in random sound out, for it is each it is taken out known to
Optimized parameter generates new probe parameters according to adjusting step-length and adjusting step number.
22. device according to claim 21, which is characterized in that the prediction model parameters tuning module can join souning out
Number imparting is corresponding to be selected prediction model and once purged prediction model training data is combined to carry out prediction model training, obtains
Different error, by the corresponding parameter value update of minimal error to prediction model training parameter library.
23. a kind of electronic equipment, which is characterized in that including:
One or more processors;
Storage device, for storing one or more programs,
When one or more of programs are executed by one or more of processors so that one or more of processors are real
The now method as described in any in claim 1-11.
24. a kind of computer readable storage medium, is stored thereon with computer program, which is characterized in that described program is handled
The method as described in any in claim 1-11 is realized when device executes.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201710272081.4A CN108734330A (en) | 2017-04-24 | 2017-04-24 | Data processing method and device |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201710272081.4A CN108734330A (en) | 2017-04-24 | 2017-04-24 | Data processing method and device |
Publications (1)
Publication Number | Publication Date |
---|---|
CN108734330A true CN108734330A (en) | 2018-11-02 |
Family
ID=63934391
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201710272081.4A Pending CN108734330A (en) | 2017-04-24 | 2017-04-24 | Data processing method and device |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN108734330A (en) |
Cited By (10)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN109685160A (en) * | 2019-01-18 | 2019-04-26 | 创新奇智(合肥)科技有限公司 | A kind of on-time model trained and dispositions method and system automatically |
CN111639798A (en) * | 2020-05-26 | 2020-09-08 | 华青融天(北京)软件股份有限公司 | Intelligent prediction model selection method and device |
CN111724211A (en) * | 2020-06-30 | 2020-09-29 | 名创优品(横琴)企业管理有限公司 | Offline store commodity sales prediction method, device and equipment |
CN111797078A (en) * | 2019-04-09 | 2020-10-20 | Oppo广东移动通信有限公司 | Data cleaning method, model training method, device, storage medium and equipment |
CN112785256A (en) * | 2021-01-14 | 2021-05-11 | 田进伟 | Real-time assessment method and system for clinical endpoint events in clinical trials |
CN113341883A (en) * | 2021-08-05 | 2021-09-03 | 山东豪泉软件技术有限公司 | Method and equipment for predicting machine tool machining working hours |
CN113606649A (en) * | 2021-07-23 | 2021-11-05 | 淄博热力有限公司 | Intelligent heat supply station control prediction system based on machine learning algorithm |
CN114239823A (en) * | 2021-12-17 | 2022-03-25 | 中国电信股份有限公司 | Modeling and using method of behavior prediction model of number card user and related equipment |
CN116061189A (en) * | 2023-03-08 | 2023-05-05 | 国网瑞嘉(天津)智能机器人有限公司 | Robot operation data processing system, method, device, equipment and medium |
CN116303382A (en) * | 2023-02-10 | 2023-06-23 | 重庆见芒信息技术咨询服务有限公司 | Multidimensional big data cleaning method and system |
Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN102968670A (en) * | 2012-10-23 | 2013-03-13 | 北京京东世纪贸易有限公司 | Method and device for predicting data |
CN104217355A (en) * | 2014-09-11 | 2014-12-17 | 北京京东尚科信息技术有限公司 | Method and device for predicting sales volume of promotion items |
CN104951843A (en) * | 2014-03-27 | 2015-09-30 | 日立(中国)研究开发有限公司 | Sales forecasting system and method |
CN106408341A (en) * | 2016-09-21 | 2017-02-15 | 北京小米移动软件有限公司 | Goods sales volume prediction method and device, and electronic equipment |
CN104200087B (en) * | 2014-06-05 | 2018-10-02 | 清华大学 | For the parameter optimization of machine learning and the method and system of feature tuning |
-
2017
- 2017-04-24 CN CN201710272081.4A patent/CN108734330A/en active Pending
Patent Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN102968670A (en) * | 2012-10-23 | 2013-03-13 | 北京京东世纪贸易有限公司 | Method and device for predicting data |
CN104951843A (en) * | 2014-03-27 | 2015-09-30 | 日立(中国)研究开发有限公司 | Sales forecasting system and method |
CN104200087B (en) * | 2014-06-05 | 2018-10-02 | 清华大学 | For the parameter optimization of machine learning and the method and system of feature tuning |
CN104217355A (en) * | 2014-09-11 | 2014-12-17 | 北京京东尚科信息技术有限公司 | Method and device for predicting sales volume of promotion items |
CN106408341A (en) * | 2016-09-21 | 2017-02-15 | 北京小米移动软件有限公司 | Goods sales volume prediction method and device, and electronic equipment |
Non-Patent Citations (1)
Title |
---|
张波等: "Web文档清洗技术", 《计算机科学》 * |
Cited By (11)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN109685160A (en) * | 2019-01-18 | 2019-04-26 | 创新奇智(合肥)科技有限公司 | A kind of on-time model trained and dispositions method and system automatically |
CN111797078A (en) * | 2019-04-09 | 2020-10-20 | Oppo广东移动通信有限公司 | Data cleaning method, model training method, device, storage medium and equipment |
CN111639798A (en) * | 2020-05-26 | 2020-09-08 | 华青融天(北京)软件股份有限公司 | Intelligent prediction model selection method and device |
CN111724211A (en) * | 2020-06-30 | 2020-09-29 | 名创优品(横琴)企业管理有限公司 | Offline store commodity sales prediction method, device and equipment |
CN112785256A (en) * | 2021-01-14 | 2021-05-11 | 田进伟 | Real-time assessment method and system for clinical endpoint events in clinical trials |
CN113606649A (en) * | 2021-07-23 | 2021-11-05 | 淄博热力有限公司 | Intelligent heat supply station control prediction system based on machine learning algorithm |
CN113341883A (en) * | 2021-08-05 | 2021-09-03 | 山东豪泉软件技术有限公司 | Method and equipment for predicting machine tool machining working hours |
CN113341883B (en) * | 2021-08-05 | 2021-11-09 | 山东豪泉软件技术有限公司 | Method and equipment for predicting machine tool machining working hours |
CN114239823A (en) * | 2021-12-17 | 2022-03-25 | 中国电信股份有限公司 | Modeling and using method of behavior prediction model of number card user and related equipment |
CN116303382A (en) * | 2023-02-10 | 2023-06-23 | 重庆见芒信息技术咨询服务有限公司 | Multidimensional big data cleaning method and system |
CN116061189A (en) * | 2023-03-08 | 2023-05-05 | 国网瑞嘉(天津)智能机器人有限公司 | Robot operation data processing system, method, device, equipment and medium |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN108734330A (en) | Data processing method and device | |
CN109242135B (en) | Model operation method, device and business server | |
CN109919684A (en) | For generating method, electronic equipment and the computer readable storage medium of information prediction model | |
CN108345958A (en) | A kind of order goes out to eat time prediction model construction, prediction technique, model and device | |
Van Volsem et al. | An evolutionary algorithm and discrete event simulation for optimizing inspection strategies for multi-stage processes | |
CN107220217A (en) | Characteristic coefficient training method and device that logic-based is returned | |
US20090043715A1 (en) | Method to Continuously Diagnose and Model Changes of Real-Valued Streaming Variables | |
CN109961248A (en) | Waybill complains prediction technique, device, equipment and its storage medium | |
US20100125487A1 (en) | System and method for estimating settings for managing a supply chain | |
Karimi-Mamaghan et al. | A learning-based metaheuristic for a multi-objective agile inspection planning model under uncertainty | |
Shadkam et al. | Multi-objective simulation optimization for selection and determination of order quantity in supplier selection problem under uncertainty and quality criteria | |
Zeiträg et al. | Surrogate-assisted automatic evolving of dispatching rules for multi-objective dynamic job shop scheduling using genetic programming | |
CN111402017A (en) | Credit scoring method and system based on big data | |
CN107729915A (en) | For the method and system for the key character for determining machine learning sample | |
Chen et al. | Extracting performance rules of suppliers in the manufacturing industry: an empirical study | |
Nguyen et al. | Genetic programming for evolving due-date assignment models in job shop environments | |
JP7304698B2 (en) | Water demand forecasting method and system | |
CN107357764A (en) | Data analysing method, electronic equipment and computer-readable storage medium | |
CN115422788B (en) | Power distribution network line loss analysis management method, device, storage medium and system | |
CN110335090A (en) | Replenishing method and system, electronic equipment based on Sales Volume of Commodity forecast of distribution | |
CN109242363A (en) | Full life cycle test management platform based on multiple quality control models | |
Silva et al. | A hybrid bi-objective optimization approach for joint determination of safety stock and safety time buffers in multi-item single-stage industrial supply chains | |
Guo et al. | Automatic design for shop scheduling strategies based on hyper-heuristics: A systematic review | |
US9697480B2 (en) | Process analysis, simulation, and optimization based on activity-based cost information | |
JP5831363B2 (en) | Manufacturing lead time prediction device, manufacturing lead time prediction method, and computer program |
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 | ||
RJ01 | Rejection of invention patent application after publication |
Application publication date: 20181102 |