CN109886408A - A kind of deep learning method and device - Google Patents
A kind of deep learning method and device Download PDFInfo
- Publication number
- CN109886408A CN109886408A CN201910151308.9A CN201910151308A CN109886408A CN 109886408 A CN109886408 A CN 109886408A CN 201910151308 A CN201910151308 A CN 201910151308A CN 109886408 A CN109886408 A CN 109886408A
- Authority
- CN
- China
- Prior art keywords
- access log
- deep learning
- log data
- data set
- learning 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
Landscapes
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The present invention provides a kind of deep learning method and device, this method comprises: obtaining access log data set, wherein the access log data set includes the access log data of at least one access log in the service of the first deep learning model reasoning;The training access log data set, generates the second deep learning model.Deep learning method provided in an embodiment of the present invention, during deep learning, it can be by having the access log data set in deep learning model service, training obtains new deep learning model, so as to realize the update to deep learning model, and new deep learning model is obtained by the training of access log data set, is suitable for new application scenarios, and then promote prediction result accuracy.
Description
Technical field
The present invention relates to field of artificial intelligence more particularly to a kind of deep learning method and devices.
Background technique
Depth learning technology be machine learning research in a new field, motivation be establish and simulate human brain into
The neural network of row analytic learning, it can imitate the mechanism of human brain to explain data, such as image, sound and text, allow mould
Type itself learns the character representation of object, and can often match in excellence or beauty or even surmount the accuracy of identification of the mankind, thus is widely used
In artificial intelligence field.
Wherein, it is to be input to a large amount of training data that depth learning technology application, which is divided into trained and reasoning, training process,
In neural network, weight is updated by backpropagation and finally obtains ideal deep learning mould to constantly restrain model
Type;Reasoning process is that the deep learning model that will be trained is applied to external service platform to obtain prediction result.But by
In current deep learning model be one-step building in training process, when practical application scene changes in reasoning process,
Deep learning model possibly can not adapt to the application scenarios after changing, low so as to cause prediction result accuracy.
Summary of the invention
The embodiment of the present invention provides a kind of deep learning method and device, there is prediction to solve current deep learning model
As a result the low problem of accuracy.
In a first aspect, the embodiment of the invention provides a kind of deep learning methods, comprising:
Obtain access log data set, wherein the access log data set takes including the first deep learning model reasoning
The access log data of at least one access log in business;
The training access log data set, generates the second deep learning model.
Second aspect, the embodiment of the invention also provides a kind of deep learning devices, comprising:
Data set acquisition module, for obtaining access log data set, wherein the access log data set includes first
The access log data of at least one access log in the service of deep learning model reasoning;
Model training module generates the second deep learning model for training the access log data set.
The third aspect, the embodiment of the invention also provides a kind of electronic equipment, which is characterized in that including processor, storage
Device and it is stored in the computer program that can be run on the memory and on the processor, the computer program is described
The step of processor realizes above-mentioned deep learning method when executing.
Fourth aspect, the embodiment of the invention also provides a kind of computer readable storage mediums, are stored thereon with computer
Program, which is characterized in that the computer program realizes the step of above-mentioned deep learning method when being executed by processor.
The embodiment of the present invention obtains access log data set, wherein access log data set includes the first deep learning mould
The access log data of at least one access log in type inference service;The training access log data set, generates second
Deep learning model.In this way, during deep learning, it can be by having the access log number in deep learning model service
According to collection, training obtains new deep learning model, so as to realize the update to deep learning model, and new deep learning
Model is obtained by the training of access log data set, is suitable for new application scenarios, and then promote prediction result accuracy.
Detailed description of the invention
In order to illustrate the technical solution of the embodiments of the present invention more clearly, needed in being described below to the embodiment of the present invention
Attached drawing to be used is briefly described, it should be apparent that, drawings in the following description are only some embodiments of the invention,
For those of ordinary skill in the art, without any creative labor, it can also obtain according to these attached drawings
Obtain other attached drawings.
Fig. 1 is the flow diagram of deep learning method provided in an embodiment of the present invention;
Fig. 2 is the total system schematic diagram of deep learning method application provided in an embodiment of the present invention;
Fig. 3 is one of the structural schematic diagram of deep learning device provided in an embodiment of the present invention;
Fig. 4 is the second structural representation of deep learning device provided in an embodiment of the present invention;
Fig. 5 is the structural schematic diagram of data set acquisition module provided in an embodiment of the present invention;
Fig. 6 is the second structural representation of deep learning device provided in an embodiment of the present invention;
Fig. 7 is the third structural representation of deep learning device provided in an embodiment of the present invention;
Fig. 8 is the four of the structural schematic diagram of deep learning device provided in an embodiment of the present invention;
Fig. 9 is the structural schematic diagram of electronic equipment provided in an embodiment of the present invention.
Specific embodiment
Following will be combined with the drawings in the embodiments of the present invention, and technical solution in the embodiment of the present invention carries out clear, complete
Site preparation description, it is clear that described embodiments are some of the embodiments of the present invention, instead of all the embodiments.Based on this hair
Embodiment in bright, every other implementation obtained by those of ordinary skill in the art without making creative efforts
Example, shall fall within the protection scope of the present invention.
It is a kind of flow chart of deep learning method provided in an embodiment of the present invention referring to Fig. 1, Fig. 1, as shown in Figure 1, packet
Include following steps:
Step 101 obtains access log data set, wherein access log data set includes that the first deep learning model pushes away
The access log data of at least one access log in reason service;
Step 102, training access log data set, generate the second deep learning model.
Here, during deep learning, electronic equipment can be by having the access day in deep learning model service
Will data set, training obtains new deep learning model, so as to realize the update to deep learning model, and new depth
Learning model is obtained by the training of access log data set, is suitable for new application scenarios, and then it is accurate to promote prediction result
Property.
It should be noted that above-mentioned electronic equipment can be any equipment that can apply above-mentioned deep learning method, example
Such as: can be automatic Pilot car-mounted terminal, image processing server or other equipment, etc..
In addition, above-mentioned electronic equipment is also possible to a device systems, which can combine shape by multiple equipment
At, such as: as shown in Fig. 2, above-mentioned electronic equipment may include business platform, model repository, inference service, data mark and
Modules such as data set warehouse, and each module can be and be made of an individual server, alternatively, be also possible to multiple modules by
One server is constituted, if business platform and model repository can be in same server, etc., herein and without limit
It is fixed.
In above-mentioned steps 101, access log data set in the service of above-mentioned acquisition the first deep learning model reasoning can
To be: electronic equipment can provide inference service after loading the first deep learning model for external equipment, and in external equipment
A series of access logs can be generated during calling the first deep learning model to make inferences service, electronic equipment can be collected
Each access log and the access log data for generating access log finally construct the visit being made of a large amount of access log data
Ask log data set.
Wherein, above-mentioned first deep learning model can be existing deep learning model in electronic equipment, can be
It is obtained by training data or the training of the access log data set of history, and at least one depth can be loaded in electronic equipment
Learning model is spent, above-mentioned first deep learning model, which can be, provides appointing for inference service at least one deep learning model
One deep learning model.
Such as: as shown in Figure 2, the business platform in electronic equipment can be obtained by the training of the training dataset of acquisition
Deep learning model, and the deep learning model that training obtains is pushed into model repository, it can store in the model repository
At least one deep learning model, when external equipment needs a certain deep learning model in calling model warehouse, i.e., first is deep
When spending learning model, inference service module loads the first deep learning model from model repository and provides inference service.
In addition, above-mentioned access log is that electronic equipment obtains during the first deep learning model provides inference service
Log may include time, access request mark, the input fields such as parameter (source) and prediction result that log generates.
Such as: by taking image classification as an example, it is cat or dog, the model that the first deep learning model, which is used to distinguish incoming picture,
It requires that end is called to be passed to picture to be predicted when On-line accoun service is provided, access log is specific as follows:
[21/Oct/2016:15:48:54+0800]reuqeust_id:10001;request_img:http://
tao.goule w.com/users/upfile/20190127/fd11998f-1f42-4dbb-9d38-
78ec0a533ca4.jpg;inferen ce_result:1;
Here, tri- words of requst_id, request_img and inference_result are included at least in access log
Section, wherein requst_id is used to indicate certain specific request, can be associated with the feedback of subsequent calls person user, i.e., above-mentioned
Access request mark;Request_img is exactly to the picture in requisition for prediction, i.e. the input parameter of the first deep learning model;
Inferece_result is the prediction result of the first deep learning model, 1 represents cat here, 0 represents dog.
It should be noted that the first deep learning model and the second deep learning model are to provide the depth of identical inference service
Spend learning model, thus the first deep learning of the data format of the access log data in above-mentioned access log data set and training
The data format of data used by model is identical, such as: the training data of the first deep learning model generally includes input ginseng
Several and output parameter, if input parameter includes input parameter 1 and input parameter 2, above-mentioned access log data also include input
Parameter 1, input parameter 2 and output parameter (i.e. prediction result).
In the embodiment of the present invention, since above-mentioned access log generally includes input parameter, output parameter and timestamp etc.
Field, therefore above-mentioned acquisition access log data set can be electronic equipment according to preset rules, extract in each access log with
The identical field of data format of the training data of first deep learning model, obtains corresponding access log data, thus raw
At above-mentioned access log data set.Above-mentioned preset rules can be it is any can extract in each access log with the first depth
The rule for practising the identical field of data format of the training data of model, is not defined herein.
Optionally, above-mentioned acquisition access log data set, comprising: in acquisition the first deep learning model reasoning service extremely
A few access log;According to the mapping relations of access log predetermined and access log data, mapping obtains at least one
The access log data of each access log in access log;Generate the access log number including at least one access log
According to access log data set.
Here, electronic equipment passes through the mapping relations of access log and access log data, can map to obtain each visit
It asks the access log data of log, and generates the access log data set being made of access log data, make to obtain access log
The mode of data set is simple, promotes the treatment effeciency of electronic equipment.
Wherein, the mapping relations for having access log Yu access log data are pre-defined in above-mentioned electronic equipment, and are passed through
The mapping relations can map to obtain the data format phase in each access log with the training data of the first deep learning model
Same field such as inputs the field of parameter and prediction result, to extract the access log data of access log.
Such as: as shown in Figure 2, the business platform in electronic equipment is preset with [data-interface definition], should [data-interface
Definition] in definition have the mapping relations of access log Yu access log data, during access log collection, electronic equipment can
With according to business platform push [data-interface definition] defined in mapping relations, the field of access log is mapped to corresponding
Access log data, specifically by taking above-mentioned image classification as an example, the training data of deep learning model include request_img and
Two fields of inferece_result can define access log and request_img in then above-mentioned [data-interface definition]
With the mapping relations of inferece_result, electronic equipment can extract request_img in each access log and
Two fields of inferece_result form access log data.
Certainly, above-mentioned electronic equipment obtains access log data set, can also be according to preset rules, and reasoning is combined to take
Access log concentration each access log is converted a valuable access day by prediction result and confidence level of business etc.
Will data, to generate access log data set.
In the embodiment of the present invention, after electronic equipment gets above-mentioned access log data set, in above-mentioned steps 102
In, the access log data set that electronic equipment can directly will acquire trains above-mentioned second deep learning model.
It should be noted that the precision due to the first deep learning model may be limited, so that the access log got
There are deviations for prediction result and the result of objective in part access log data in data set, thus can pass through mark
Operation is modified the prediction result in above-mentioned part access log data, obtains through revised access log data
Collect, which is incremental data set, and trains the second deep learning model by incremental data set,
And then the precision of prediction of the second deep learning model can be promoted.
Wherein, above-mentioned labeling operation can be any input data based in access log data and carry out to prediction result
Amendment, with realize obtain inference service request objective result purpose operation, can be by by hand mark or
The modes such as person's user feedback realize, such as: by taking above-mentioned image classification as an example, picture, that is, request_img is inputted in inference service
In image it is practical be dog, objective result should be 0, and prediction result, that is, inference_result of access log be 1,
It is labeled so as to the prediction result to the access log, can be modification prediction result is 0, to accurately be visited
Ask log.
In addition, the process of above-mentioned artificial labeling operation can be and carry out on an electronic device, it is also possible in external equipment
Upper progress, to keep the mode for carrying out the labeling operation to access log data set more flexible.
Optionally, above-mentioned trained access log data set can also include: to connect before generating the second deep learning model
Receive the labeling operation carried out to access log data set, wherein labeling operation is that mark personnel are based in access log data set
The input parameter of at least one access log data, the behaviour that the prediction result of at least one access log data is modified
Make;In response to labeling operation, corresponding first incremental data set of access log data set is generated;Above-mentioned trained access log data
Collection generates the second deep learning model, comprising: and the first incremental data set of training generates the second deep learning model, so as to
So that mark personnel is completed the labeling operation to access day to data set on an electronic device, promotes treatment effeciency.
Alternatively, optional, above-mentioned trained access log data set, before generating the second deep learning model, further includes: will
Access log data set is sent to external equipment;Receive the second incremental data set of external equipment feedback, wherein the second incremental number
According to the data set integrated as external equipment based on the second labeling operation generation received, and the second labeling operation is mark personnel base
The input parameter of at least one access log data in access log data set, the prediction at least one access log data
As a result the operation being modified;Above-mentioned trained access log data set generates the second deep learning model, comprising: training second
Incremental data set generates the second deep learning model, so as to complete the mark to access day to data set on external equipment
Note operation, saves the process resource expense of electronic equipment.
Such as: as shown in Figure 2, the inference service module in electronic equipment can access log collection, will collect
To access log generate and access log data set and push to data labeling module, data labeling module can be by manually marking
The modes such as note or crowdsourcing access, which are realized, is labeled operation to the access log data in access log data set, and in crowdsourcing
In access way, the access log data set of Inference Forecast service can be delivered to the by introducing third party's crowdsourcing service
Three method, apparatus (i.e. external equipment), it is one by one that access log data are tagged by third party device, the label to indicate by
Prediction result amendment, to generate the incremental data set (i.e. the second incremental data set) marked, and by above-mentioned incremental data set
It is back to above-mentioned electronic equipment;Alternatively, data labeling module provides an artificial mark for developer in artificial notation methods
Tool, by system access log data one by one show developer, be manually labeled by developer, to generate mark
Complete incremental data set (i.e. the first incremental data set);In addition, after above-mentioned data labeling module gets incremental data set,
Incremental data set is pushed into data set warehouse, data set warehouse can store at least one set of incremental data set, and business platform
It can be concentrated from data warehouse and call incremental data set and carry out model training, and then obtain the second deep learning model.
In the embodiment of the present invention, in above-mentioned steps 102, electronic equipment can be visited by the training of preset model training algorithm
Log data set is asked, to generate the second deep learning model.
Wherein, the process of above-mentioned model training may is that after the triggering of new model training mission, get access log
In the case where data set, electronic equipment can load preset model version, and access is carried on the basis of the model version of load
Log data set (or the incremental data set generated after labeling operation) is trained, and when reaching the condition of training termination, is obtained
To above-mentioned second deep learning model.
Such as: can in model repository as shown in Figure 2 the highest deep learning model of choice accuracy as above-mentioned pre-
If model version, the data in access log data set are pre-processed;Pretreated data are inputted into preset mould
The neural network forward-propagating of stencilling sheet, obtains score;By score error originated from input function, error is obtained compared with expected value, it is more
It is a to be then and degree is identified by error judgment;Gradient vector is determined by backpropagation;It is adjusted finally by gradient vector
Each whole weight makes score tend to 0 or the adjusting of convergent trend to error;It repeats the above process until reaching scheduled training
Wheel number or the average value for damaging error mistake no longer decline, and training is completed simultaneously to obtain above-mentioned second deep learning model.
In the embodiment of the present invention, electronic equipment is according to the access log data in the service of the deep learning model reasoning of calling
Collection repeats the training of new model.
Optionally, before above-mentioned acquisition access log data set, further includes: determine the access day of the first deep learning model
Whether the collection process of will data meets preset condition;Obtain access log data set, comprising: determining that it is pre- that collection process meets
If in the case where condition, executing and obtaining access log data set;Wherein, access log data set is collected during being included in collection
The access log data arrived.
Here, electronic equipment can meet default item in the collection process of the access log data of the first deep learning model
In the case where part, automatic trigger new model training mission, to allow electronic equipment based on the access log during collection
Data set training obtains new model, and it is convenient to operate.
Optionally, whether the collection process of the access log data of above-mentioned the first deep learning of determination model meets default item
Part, comprising: determine the access log data being collected into during the collection of the access log data of the first deep learning model
Whether preset quantity is reached, if so, determining that collection process meets preset condition;Access is collected in the process alternatively, determining and collecting
Whether the duration of daily record data reaches preset duration, if so, determining that collection process meets preset condition, keeps electronic equipment automatic
The mode for triggering new model training mission is more flexible.
Certainly, above-mentioned preset condition can also be other conditions, not be defined herein.
In the specific embodiment of the invention, after electronic equipment gets the second deep learning model in above-mentioned steps 102,
The deep learning model modification called in inference service can be the second deep learning model by electronic equipment, i.e., by the first depth
Learning model is updated to the second deep learning model.
Wherein, it is above-mentioned by the first deep learning model modification be the second deep learning model, can be electronic equipment and obtaining
It is directly the second deep learning model by the first deep learning model modification after getting the second deep learning model.
Alternatively, optional, the training access log data set, after generating the second deep learning model, comprising:
The first deep learning model and the second deep learning model are assessed, first depth is obtained
Second assessment result of the first assessment result of learning model and the second deep learning model;
It is deep by described first in the case where second assessment result and first assessment result meet preset condition
Degree learning model is changed to the second deep learning model.
Here, in the case where above-mentioned second assessment result and the first assessment result meet preset condition, the second depth
The performance (such as precision or response efficiency) of habit model reasoning service should be above the first deep learning model reasoning service
Performance, so as to further promote the performance of inference service.
Such as: as shown in Figure 2, [appraisal procedure definition] is preset in the business platform module of electronic equipment, by [commenting
Estimate method definition] new deep learning model can be assessed, and obtain the assessment score i.e. assessment result of new model, and
Compare the assessment score (i.e. the first assessment result) of existing model and the assessment score (i.e. the second assessment result) of new model, and
In the case that the assessment score of new model is higher than the assessment score of existing model, new model is substituted into inference service.
Optionally, first assessment result and second assessment result include precision of prediction;It is described described
In the case that two assessment results and first assessment result meet preset condition, the first deep learning model is changed to
The second deep learning model, comprising: be more than or equal to the feelings of first assessment result in second assessment result
Under condition, the first deep learning model is changed to the second deep learning model, so as to promote inference service
Accuracy.
Certainly, above-mentioned first assessment result and the second assessment result can also include in response speed and response time extremely
One item missing, such as: above-mentioned first assessment result and the second assessment result include precision of prediction and response time, then,
The precision of prediction of two assessment results is more than or equal to the precision of prediction of the first assessment result, and when the response of the second assessment result
In the case where the long response time for being less than or equal to the first assessment result, the first deep learning model is changed to the second depth
Learning model.
In addition, it is above-mentioned by the first deep learning model modification be the second deep learning model, can be will be in electronic equipment
First deep learning model of storage is changed to the second deep learning model, i.e. the second deep learning model covers the first depth
Model is practised, so as to the second deep learning model is called in subsequent inference service, such as: it is as shown in Figure 2, deep above-mentioned first
In the case where spending learning model for model 1, business platform can be pushed to the second deep learning model in model repository, and will
Model 1 in model repository is changed to the second deep learning model.
It should be noted that above-mentioned assess the first deep learning model and the second deep learning model, can be
Using preset test data as the input parameter of the first deep learning model and the second deep learning model reasoning service, and root
According to the prediction result of the first deep learning model and the second deep learning model, test set data are calculated in the first deep learning mould
Error in type and the second deep learning model, to obtain the first assessment result and the second assessment result.Wherein, above-mentioned default
Test data can be it is different from the training data of above-mentioned first deep learning model and the second deep learning model, to protect
Demonstrate,prove the accuracy of assessment.
In the embodiment of the present invention, by obtaining access log data set, wherein access log data set includes the first depth
The access log data of at least one access log in learning model inference service;The training access log data set, it is raw
At the second deep learning model.In this way, during deep learning, it can be by having the access in deep learning model service
Log data set, training obtains new deep learning model, so as to realize the update to deep learning model, and new depth
Degree learning model is obtained by the training of access log data set, is suitable for new application scenarios, and then it is quasi- to promote prediction result
True property.
It is the structure chart of deep learning device provided in an embodiment of the present invention referring to Fig. 3, Fig. 3, as shown in figure 3, depth
Practising device 300 includes:
Data set acquisition module 301, for obtaining access log data set, wherein the access log data set includes
The access log data of at least one access log in the service of first deep learning model reasoning;
Model training module 302 generates the second deep learning model for training the access log data set.
Optionally, as shown in figure 4, the deep learning device 300, further includes:
Evaluation module 303, for assessing the first deep learning model and the second deep learning model,
Obtain the first assessment result of the first deep learning model and the second assessment result of the second deep learning model;
Model replaces module 304, for meeting preset condition in second assessment result and first assessment result
In the case where, the first deep learning model is changed to the second deep learning model.
Optionally, first assessment result and second assessment result include in precision of prediction and response speed
At least one of;
The model replaces module 304, is specifically used for:
In the case where second assessment result is more than or equal to first assessment result, by first depth
Learning model is changed to the second deep learning model.
Optionally, as shown in figure 5, the data set acquisition module 301, comprising:
Access log acquisition unit 3011, for acquiring at least one in the first deep learning model reasoning service
Access log;
Map unit 3012, for the mapping relations according to access log predetermined and access log data, mapping
Obtain the access log data of each access log at least one access log;
Data set generating unit 3013, for generating the institute of the access log data including at least one access log
State access log data set.
Optionally, as shown in fig. 6, the deep learning device 300 further include:
Labeling operation receiving module 305, for receiving the labeling operation carried out to the access log data set, wherein
The labeling operation is input parameter of the mark personnel based at least one access log data in the access log data set,
The operation that the prediction result of at least one access log data is modified;
Incremental data set generation module 306, for generating the access log data set pair in response to the labeling operation
The first incremental data set answered;
The model training module 302, is specifically used for:
Training first incremental data set, generates the second deep learning model.
Optionally, as shown in fig. 7, the deep learning device 300 further include:
Sending module 307, for the access log data set to be sent to external equipment;
Receiving module 308, for receiving the second incremental data set of the external equipment feedback, wherein described second increases
The data set that amount data set is generated for the external equipment based on the second labeling operation received, and second labeling operation
To mark input parameter of the personnel based at least one access log data in the access log data set, to described at least one
The operation that the prediction result of access log data is modified;
The model training module 302, is specifically used for:
Training second incremental data set, generates the second deep learning model.
Optionally, as shown in figure 8, the deep learning device 300 further include:
Determining module 309, for determine the first deep learning model access log data collection process whether
Meet preset condition;
The data set acquisition module 301, is also used to:
In the case where determining that the collection process meets preset condition, the acquisition access log data set is executed;
Wherein, the access log data set includes the access log data being collected into during the collection.
Optionally, the determining module 309, is specifically used for:
Determine the access log being collected into during the collection of the access log data of the first deep learning model
Whether data reach preset quantity, if so, determining that the collection process meets preset condition;Alternatively,
Determine whether the duration that access log data are collected during the collection reaches preset duration, if so, determining
The collection process meets preset condition.
It is real that deep learning device 300 provided in an embodiment of the present invention can be realized in Fig. 1 electronic equipment in embodiment of the method
Existing each process, to avoid repeating, which is not described herein again.
The hardware structural diagram of Fig. 9 a kind of electronic equipment of each embodiment to realize the present invention.
The electronic equipment 900 includes but is not limited to: radio frequency unit 901, network module 902, audio output unit 903, defeated
Enter unit 904, sensor 905, display unit 906, user input unit 907, interface unit 908, memory 909, processor
The components such as 910 and power supply 911.It will be understood by those skilled in the art that electronic devices structure shown in Fig. 9 is not constituted
Restriction to electronic equipment, electronic equipment may include than illustrating more or fewer components, perhaps combine certain components or
Different component layouts.
Wherein, processor 910 are used for:
Obtain access log data set, wherein the access log data set takes including the first deep learning model reasoning
The access log data of at least one access log in business;
The training access log data set, generates the second deep learning model.
Optionally, processor 910 are also used to:
The first deep learning model and the second deep learning model are assessed, first depth is obtained
Second assessment result of the first assessment result of learning model and the second deep learning model;
It is deep by described first in the case where second assessment result and first assessment result meet preset condition
Degree learning model is changed to the second deep learning model.
Optionally, first assessment result and second assessment result include in precision of prediction and response speed
At least one of;
Processor 910, is specifically used for:
In the case where second assessment result is more than or equal to first assessment result, by first depth
Learning model is changed to the second deep learning model.
Optionally, the access log data set includes at least one access log data;
Processor 910, is specifically used for:
Acquire at least one access log in the first deep learning model reasoning service;
According to the mapping relations of access log predetermined and access log data, mapping obtains described at least one and visits
Ask the access log data of each access log in log;
Generate the access log data set of the access log data including at least one access log.
Optionally, processor 910 are also used to:
Receive the labeling operation carried out to the access log data set, wherein the labeling operation is mark personnel base
The input parameter of at least one access log data in the access log data set, at least one access log number
According to the operation that is modified of prediction result;
In response to the labeling operation, corresponding first incremental data set of the access log data set is generated;
Training first incremental data set, generates the second deep learning model.
Optionally, processor 910 are also used to:
The access log data set is sent to external equipment;
Receive the second incremental data set of the external equipment feedback, wherein second incremental data set is described outer
The data set that portion's equipment is generated based on the second labeling operation received, and second labeling operation is based on institute for mark personnel
The input parameter for stating at least one access log data in access log data set, at least one access log data
The operation that prediction result is modified;
Training second incremental data set, generates the second deep learning model.
Optionally, processor 910 are also used to:
Determine whether the collection process of the access log data of the first deep learning model meets preset condition;
In the case where determining that the collection process meets preset condition, the acquisition access log data set is executed;
Wherein, the access log data set includes the access log data being collected into during the collection.
Optionally, processor 910 are specifically used for:
Determine the access log being collected into during the collection of the access log data of the first deep learning model
Whether data reach preset quantity, if so, determining that the collection process meets preset condition;Alternatively,
Determine whether the duration that access log data are collected during the collection reaches preset duration, if so, determining
The collection process meets preset condition.
It should be understood that the embodiment of the present invention in, radio frequency unit 901 can be used for receiving and sending messages or communication process in, signal
Send and receive, specifically, by from base station downlink data receive after, to processor 910 handle;In addition, by uplink
Data are sent to base station.In general, radio frequency unit 901 includes but is not limited to antenna, at least one amplifier, transceiver, coupling
Device, low-noise amplifier, duplexer etc..In addition, radio frequency unit 901 can also by wireless communication system and network and other set
Standby communication.
Electronic equipment provides wireless broadband internet by network module 902 for user and accesses, and such as user is helped to receive
It sends e-mails, browse webpage and access streaming video etc..
Audio output unit 903 can be received by radio frequency unit 901 or network module 902 or in memory 909
The audio data of storage is converted into audio signal and exports to be sound.Moreover, audio output unit 903 can also provide and electricity
The relevant audio output of specific function that sub- equipment 900 executes is (for example, call signal receives sound, message sink sound etc.
Deng).Audio output unit 903 includes loudspeaker, buzzer and receiver etc..
Input unit 904 is for receiving audio or video signal.Input unit 904 may include graphics processor
(Graphics Processing Unit, GPU) 9041 and microphone 9042, graphics processor 9041 is in video acquisition mode
Or the image data of the static images or video obtained in image capture mode by image capture apparatus (such as camera) carries out
Reason.Treated, and picture frame may be displayed on display unit 906.Through graphics processor 9041, treated that picture frame can be deposited
Storage is sent in memory 909 (or other storage mediums) or via radio frequency unit 901 or network module 902.Mike
Wind 9042 can receive sound, and can be audio data by such acoustic processing.Treated audio data can be
The format output that mobile communication base station can be sent to via radio frequency unit 901 is converted in the case where telephone calling model.
Electronic equipment 900 further includes at least one sensor 905, such as optical sensor, motion sensor and other biographies
Sensor.Specifically, optical sensor includes ambient light sensor and proximity sensor, wherein ambient light sensor can be according to environment
The light and shade of light adjusts the brightness of display panel 9091, and proximity sensor can close when electronic equipment 900 is moved in one's ear
Display panel 9091 and/or backlight.As a kind of motion sensor, accelerometer sensor can detect in all directions (general
For three axis) size of acceleration, it can detect that size and the direction of gravity when static, can be used to identify electronic equipment posture (ratio
Such as horizontal/vertical screen switching, dependent game, magnetometer pose calibrating), Vibration identification correlation function (such as pedometer, tap);It passes
Sensor 905 can also include fingerprint sensor, pressure sensor, iris sensor, molecule sensor, gyroscope, barometer, wet
Meter, thermometer, infrared sensor etc. are spent, details are not described herein.
Display unit 906 is for showing information input by user or being supplied to the information of user.Display unit 906 can wrap
Display panel 9091 is included, liquid crystal display (Liquid Crystal Display, LCD), Organic Light Emitting Diode can be used
Forms such as (Organic Light-Emitting Diode, OLED) configure display panel 9091.
User input unit 907 can be used for receiving the number or character information of input, and generate the use with electronic equipment
Family setting and the related key signals input of function control.Specifically, user input unit 907 include touch panel 9091 and
Other input equipments 9072.Touch panel 9091, also referred to as touch screen collect the touch operation of user on it or nearby
(for example user uses any suitable objects or attachment such as finger, stylus on touch panel 9091 or in touch panel 9091
Neighbouring operation).Touch panel 9091 may include both touch detecting apparatus and touch controller.Wherein, touch detection
Device detects the touch orientation of user, and detects touch operation bring signal, transmits a signal to touch controller;Touch control
Device processed receives touch information from touch detecting apparatus, and is converted into contact coordinate, then gives processor 910, receiving area
It manages the order that device 910 is sent and is executed.Furthermore, it is possible to more using resistance-type, condenser type, infrared ray and surface acoustic wave etc.
Seed type realizes touch panel 9091.In addition to touch panel 9091, user input unit 907 can also include other input equipments
9072.Specifically, other input equipments 9072 can include but is not limited to physical keyboard, function key (such as volume control button,
Switch key etc.), trace ball, mouse, operating stick, details are not described herein.
Further, touch panel 9091 can be covered on display panel 9091, when touch panel 9091 is detected at it
On or near touch operation after, send processor 910 to determine the type of touch event, be followed by subsequent processing device 910 according to touching
The type for touching event provides corresponding visual output on display panel 9091.Although in Fig. 9, touch panel 9091 and display
Panel 9091 is the function that outputs and inputs of realizing electronic equipment as two independent components, but in some embodiments
In, can be integrated by touch panel 9091 and display panel 9091 and realize the function that outputs and inputs of electronic equipment, it is specific this
Place is without limitation.
Interface unit 908 is the interface that external device (ED) is connect with electronic equipment 900.For example, external device (ED) may include having
Line or wireless head-band earphone port, external power supply (or battery charger) port, wired or wireless data port, storage card end
Mouth, port, the port audio input/output (I/O), video i/o port, earphone end for connecting the device with identification module
Mouthful etc..Interface unit 908 can be used for receiving the input (for example, data information, electric power etc.) from external device (ED) and
By one or more elements that the input received is transferred in electronic equipment 900 or can be used in 900 He of electronic equipment
Data are transmitted between external device (ED).
Memory 909 can be used for storing software program and various data.Memory 909 can mainly include storing program area
The storage data area and, wherein storing program area can (such as the sound of application program needed for storage program area, at least one function
Sound playing function, image player function etc.) etc.;Storage data area can store according to mobile phone use created data (such as
Audio data, phone directory etc.) etc..In addition, memory 909 may include high-speed random access memory, it can also include non-easy
The property lost memory, a for example, at least disk memory, flush memory device or other volatile solid-state parts.
Processor 910 is the control centre of electronic equipment, utilizes each of various interfaces and the entire electronic equipment of connection
A part by running or execute the software program and/or module that are stored in memory 909, and calls and is stored in storage
Data in device 909 execute the various functions and processing data of electronic equipment, to carry out integral monitoring to electronic equipment.Place
Managing device 910 may include one or more processing units;Preferably, processor 910 can integrate application processor and modulatedemodulate is mediated
Manage device, wherein the main processing operation system of application processor, user interface and application program etc., modem processor is main
Processing wireless communication.It is understood that above-mentioned modem processor can not also be integrated into processor 910.
Electronic equipment 900 can also include the power supply 911 (such as battery) powered to all parts, it is preferred that power supply 911
Can be logically contiguous by power-supply management system and processor 910, to realize management charging by power-supply management system, put
The functions such as electricity and power managed.
In addition, electronic equipment 900 includes some unshowned functional modules, details are not described herein.
Preferably, the embodiment of the present invention also provides a kind of electronic equipment, including processor 910, and memory 909 is stored in
On memory 909 and the computer program that can run on the processor 910, the computer program are executed by processor 910
Each process of the above-mentioned deep learning embodiment of the method for Shi Shixian, and identical technical effect can be reached, to avoid repeating, here
It repeats no more.
The embodiment of the present invention also provides a kind of computer readable storage medium, and meter is stored on computer readable storage medium
Calculation machine program, the computer program realize each process of above-mentioned deep learning embodiment of the method, and energy when being executed by processor
Reach identical technical effect, to avoid repeating, which is not described herein again.Wherein, the computer readable storage medium, such as only
Read memory (Read-Only Memory, abbreviation ROM), random access memory (Random Access Memory, abbreviation
RAM), magnetic or disk etc..
Those of ordinary skill in the art may be aware that list described in conjunction with the examples disclosed in the embodiments of the present disclosure
Member and algorithm steps can be realized with the combination of electronic hardware or computer software and electronic hardware.These functions are actually
It is implemented in hardware or software, the specific application and design constraint depending on technical solution.Professional technician
Each specific application can be used different methods to achieve the described function, but this realization is it is not considered that exceed
The scope of the present invention.
It is apparent to those skilled in the art that for convenience and simplicity of description, the system of foregoing description,
The specific work process of device and unit, can refer to corresponding processes in the foregoing method embodiment, and details are not described herein.
In embodiment provided herein, it should be understood that disclosed device and method can pass through others
Mode is realized.For example, the apparatus embodiments described above are merely exemplary, for example, the division of the unit, only
A kind of logical function partition, there may be another division manner in actual implementation, for example, multiple units or components can combine or
Person is desirably integrated into another system, or some features can be ignored or not executed.Another point, shown or discussed is mutual
Between coupling, direct-coupling or communication connection can be through some interfaces, the INDIRECT COUPLING or communication link of device or unit
It connects, can be electrical property, mechanical or other forms.
The unit as illustrated by the separation member may or may not be physically separated, aobvious as unit
The component shown may or may not be physical unit, it can and it is in one place, or may be distributed over multiple
In network unit.Some or all of unit therein can be selected to realize the embodiment of the present invention according to the actual needs
Purpose.
It, can also be in addition, the functional units in various embodiments of the present invention may be integrated into one processing unit
It is that each unit physically exists alone, can also be integrated in one unit with two or more units.
It, can be with if the function is realized in the form of SFU software functional unit and when sold or used as an independent product
It is stored in a computer readable storage medium.Based on this understanding, technical solution of the present invention is substantially in other words
The part of the part that contributes to existing technology or the technical solution can be embodied in the form of software products, the meter
Calculation machine software product is stored in a storage medium, including some instructions are used so that a computer equipment (can be a
People's computer, server or network equipment etc.) it performs all or part of the steps of the method described in the various embodiments of the present invention.
And storage medium above-mentioned includes: that USB flash disk, mobile hard disk, ROM, RAM, magnetic or disk etc. are various can store program code
Medium.
The above description is merely a specific embodiment, but scope of protection of the present invention is not limited thereto, any
Those familiar with the art in the technical scope disclosed by the present invention, can easily think of the change or the replacement, and should all contain
Lid is within protection scope of the present invention.Therefore, protection scope of the present invention should be subject to the protection scope in claims.
Claims (18)
1. a kind of deep learning method characterized by comprising
Obtain access log data set, wherein the access log data set includes in the service of the first deep learning model reasoning
At least one access log access log data;
The training access log data set, generates the second deep learning model.
2. the method according to claim 1, wherein the training access log data set, generates second
After deep learning model, comprising:
The first deep learning model and the second deep learning model are assessed, first deep learning is obtained
Second assessment result of the first assessment result of model and the second deep learning model;
In the case where second assessment result and first assessment result meet preset condition, by first depth
It practises model and is changed to the second deep learning model.
3. according to the method described in claim 2, it is characterized in that, first assessment result and second assessment result are equal
Including precision of prediction;
It is described in the case where second assessment result and first assessment result meet preset condition, will first depth
Degree learning model is changed to the second deep learning model, comprising:
In the case where second assessment result is more than or equal to first assessment result, by first deep learning
Model is changed to the second deep learning model.
4. the method according to claim 1, wherein the acquisition access log data set, comprising:
Acquire at least one access log in the first deep learning model reasoning service;
According to the mapping relations of access log predetermined and access log data, mapping obtains at least one access day
The access log data of each access log in will;
Generate the access log data set of the access log data including at least one access log.
5. method according to claim 1 to 4, which is characterized in that the training access log data
Collect, before the second deep learning model of generation, further includes:
Receive the labeling operation carried out to the access log data set, wherein the labeling operation is based on institute for mark personnel
The input parameter for stating at least one access log data in access log data set, at least one access log data
The operation that prediction result is modified;
In response to the labeling operation, corresponding first incremental data set of the access log data set is generated;
The training access log data set generates the second deep learning model, comprising:
Training first incremental data set, generates the second deep learning model.
6. method according to claim 1 to 4, which is characterized in that the training access log data
Collect, before the second deep learning model of generation, further includes:
The access log data set is sent to external equipment;
Receive the second incremental data set of the external equipment feedback, wherein second incremental data set is that the outside is set
The standby data set generated based on the second labeling operation received, and second labeling operation is that mark personnel are based on the visit
Ask the input parameter that daily record data concentrates at least one access log data, the prediction at least one access log data
As a result the operation being modified;
The training access log data set generates the second deep learning model, comprising:
Training second incremental data set, generates the second deep learning model.
7. method according to claim 1 to 4, which is characterized in that the acquisition access log data set it
Before, further includes:
Determine whether the collection process of the access log data of the first deep learning model meets preset condition;
The acquisition access log data set, comprising:
In the case where determining that the collection process meets preset condition, the acquisition access log data set is executed;
Wherein, the access log data set includes the access log data being collected into during the collection.
8. the method according to the description of claim 7 is characterized in that the access day of the determination the first deep learning model
Whether the collection process of will data meets preset condition, comprising:
Determine the access log data being collected into during the collection of the access log data of the first deep learning model
Whether preset quantity is reached, if so, determining that the collection process meets preset condition;Alternatively,
Determine whether the duration that access log data are collected during the collection reaches preset duration, if so, described in determining
Collection process meets preset condition.
9. a kind of deep learning device characterized by comprising
Data set acquisition module, for obtaining access log data set, wherein the access log data set includes the first depth
The access log data of at least one access log in learning model inference service;
Model training module generates the second deep learning model for training the access log data set.
10. deep learning device according to claim 9, which is characterized in that the deep learning device, further includes:
Evaluation module obtains institute for assessing the first deep learning model and the second deep learning model
State the first assessment result of the first deep learning model and the second assessment result of the second deep learning model;
Model replaces module, for the case where second assessment result and first assessment result meet preset condition
Under, the first deep learning model is changed to the second deep learning model.
11. deep learning device according to claim 10, which is characterized in that first assessment result and described second
Assessment result includes at least one in precision of prediction and response speed;
The model replaces module, is specifically used for:
In the case where second assessment result is more than or equal to first assessment result, by first deep learning
Model is changed to the second deep learning model.
12. deep learning device according to claim 9, which is characterized in that the data set acquisition module, comprising:
Access log acquisition unit, for acquiring at least one access day in the first deep learning model reasoning service
Will;
Map unit, for the mapping relations according to access log predetermined and access log data, mapping obtains described
The access log data of each access log at least one access log;
Data set generating unit, for generating the access day of the access log data including at least one access log
Will data set.
13. the deep learning device according to any one of claim 9 to 12, which is characterized in that the deep learning dress
It sets further include:
Labeling operation receiving module, for receiving the labeling operation carried out to the access log data set, wherein the mark
Operation be input parameter of the mark personnel based at least one access log data in the access log data set, to it is described extremely
The operation that the prediction result of few access log data is modified;
Incremental data set generation module, in response to the labeling operation, generating the access log data set corresponding the
One incremental data set;
The model training module, is specifically used for:
Training first incremental data set, generates the second deep learning model.
14. the deep learning device according to any one of claim 9 to 12, which is characterized in that the deep learning dress
It sets further include:
Sending module, for the access log data set to be sent to external equipment;
Receiving module, for receiving the second incremental data set of the external equipment feedback, wherein second incremental data set
Data set for the external equipment based on the second labeling operation generation received, and second labeling operation is mark people
Input parameter of the member based at least one access log data in the access log data set, at least one access day
The operation that the prediction result of will data is modified;
The model training module, is specifically used for:
Training second incremental data set, generates the second deep learning model.
15. the deep learning device according to any one of claim 9 to 12, which is characterized in that the deep learning dress
It sets further include:
Determining module, for determining it is default whether the collection process of access log data of the first deep learning model meets
Condition;
The data set acquisition module, is also used to:
In the case where determining that the collection process meets preset condition, the acquisition access log data set is executed;
Wherein, the access log data set includes the access log data being collected into during the collection.
16. deep learning device according to claim 15, which is characterized in that the determining module is specifically used for:
Determine the access log data being collected into during the collection of the access log data of the first deep learning model
Whether preset quantity is reached, if so, determining that the collection process meets preset condition;Alternatively,
Determine whether the duration that access log data are collected during the collection reaches preset duration, if so, described in determining
Collection process meets preset condition.
17. a kind of electronic equipment, which is characterized in that including processor, memory and be stored on the memory and can be in institute
The computer program run on processor is stated, such as claim 1 to 8 is realized when the computer program is executed by the processor
Any one of described in deep learning method the step of.
18. a kind of computer readable storage medium, is stored thereon with computer program, which is characterized in that the computer program
The step of deep learning method described in any item of the claim 1 to 8 is realized when being executed by processor.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910151308.9A CN109886408A (en) | 2019-02-28 | 2019-02-28 | A kind of deep learning method and device |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910151308.9A CN109886408A (en) | 2019-02-28 | 2019-02-28 | A kind of deep learning method and device |
Publications (1)
Publication Number | Publication Date |
---|---|
CN109886408A true CN109886408A (en) | 2019-06-14 |
Family
ID=66929992
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201910151308.9A Pending CN109886408A (en) | 2019-02-28 | 2019-02-28 | A kind of deep learning method and device |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN109886408A (en) |
Cited By (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110414187A (en) * | 2019-07-03 | 2019-11-05 | 北京百度网讯科技有限公司 | Model safety delivers the system and method for automation |
CN111190804A (en) * | 2019-12-28 | 2020-05-22 | 同济大学 | Multi-level deep learning log fault detection method for cloud native system |
CN112104659A (en) * | 2020-09-18 | 2020-12-18 | 宋清云 | Real-time monitoring platform based on government affair application safety |
CN112722356A (en) * | 2020-12-22 | 2021-04-30 | 宋清云 | Ampoule seals all-in-one by melting |
Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN106951926A (en) * | 2017-03-29 | 2017-07-14 | 山东英特力数据技术有限公司 | The deep learning systems approach and device of a kind of mixed architecture |
CN107423535A (en) * | 2016-05-23 | 2017-12-01 | 百度(美国)有限责任公司 | For the methods, devices and systems for the medical conditions for determining user |
CN108280462A (en) * | 2017-12-11 | 2018-07-13 | 北京三快在线科技有限公司 | A kind of model training method and device, electronic equipment |
CN108364028A (en) * | 2018-03-06 | 2018-08-03 | 中国科学院信息工程研究所 | A kind of internet site automatic classification method based on deep learning |
-
2019
- 2019-02-28 CN CN201910151308.9A patent/CN109886408A/en active Pending
Patent Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN107423535A (en) * | 2016-05-23 | 2017-12-01 | 百度(美国)有限责任公司 | For the methods, devices and systems for the medical conditions for determining user |
CN106951926A (en) * | 2017-03-29 | 2017-07-14 | 山东英特力数据技术有限公司 | The deep learning systems approach and device of a kind of mixed architecture |
CN108280462A (en) * | 2017-12-11 | 2018-07-13 | 北京三快在线科技有限公司 | A kind of model training method and device, electronic equipment |
CN108364028A (en) * | 2018-03-06 | 2018-08-03 | 中国科学院信息工程研究所 | A kind of internet site automatic classification method based on deep learning |
Non-Patent Citations (1)
Title |
---|
吴向阳等: "面向订票服务器端爬虫的可视检测方法研究", 《计算机辅助设计与图形学学报》 * |
Cited By (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110414187A (en) * | 2019-07-03 | 2019-11-05 | 北京百度网讯科技有限公司 | Model safety delivers the system and method for automation |
CN110414187B (en) * | 2019-07-03 | 2021-09-17 | 北京百度网讯科技有限公司 | System and method for model safety delivery automation |
CN111190804A (en) * | 2019-12-28 | 2020-05-22 | 同济大学 | Multi-level deep learning log fault detection method for cloud native system |
CN112104659A (en) * | 2020-09-18 | 2020-12-18 | 宋清云 | Real-time monitoring platform based on government affair application safety |
CN112722356A (en) * | 2020-12-22 | 2021-04-30 | 宋清云 | Ampoule seals all-in-one by melting |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN109886408A (en) | A kind of deep learning method and device | |
CN107101665A (en) | Movable information provides method and supports the electronic installation of methods described | |
CN110096580A (en) | A kind of FAQ dialogue method, device and electronic equipment | |
CN108090855A (en) | Method and mobile terminal are recommended in a kind of study plan | |
CN108121803A (en) | A kind of method and server of definite page layout | |
CN108089891A (en) | A kind of application program launching method, mobile terminal | |
CN108984064A (en) | Multi-screen display method, device, storage medium and electronic equipment | |
CN109543099A (en) | A kind of content recommendation method and terminal device | |
CN106649010A (en) | Terminal device testing method and terminal device | |
CN109871246A (en) | A kind of message display method and terminal device | |
CN109005336A (en) | A kind of image capturing method and terminal device | |
CN108616448A (en) | A kind of the path recommendation method and mobile terminal of Information Sharing | |
CN108135033A (en) | A kind of Bluetooth connecting method and mobile terminal | |
CN109062468A (en) | Multi-screen display method, device, storage medium and electronic equipment | |
CN108037885A (en) | A kind of operation indicating method and mobile terminal | |
CN109271297A (en) | A kind of abnormal prompt method and mobile terminal | |
CN108681483A (en) | A kind of task processing method and device | |
CN108958593A (en) | A kind of method and mobile terminal of determining communication object | |
CN109085963A (en) | A kind of interface display method and terminal device | |
CN108958629A (en) | Split screen exits method, apparatus, storage medium and electronic equipment | |
CN108549660B (en) | Information pushing method and device | |
CN107317918A (en) | Parameter setting method and related product | |
CN107765954A (en) | A kind of application icon update method, mobile terminal and server | |
CN108762621B (en) | Message display method and mobile terminal | |
CN109783722A (en) | A kind of content outputting method and terminal device |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PB01 | Publication | ||
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
RJ01 | Rejection of invention patent application after publication |
Application publication date: 20190614 |
|
RJ01 | Rejection of invention patent application after publication |