CN110288089A - Method and apparatus for sending information - Google Patents
Method and apparatus for sending information Download PDFInfo
- Publication number
- CN110288089A CN110288089A CN201910575820.6A CN201910575820A CN110288089A CN 110288089 A CN110288089 A CN 110288089A CN 201910575820 A CN201910575820 A CN 201910575820A CN 110288089 A CN110288089 A CN 110288089A
- Authority
- CN
- China
- Prior art keywords
- target
- classification information
- information
- external
- training
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
- G06N3/084—Backpropagation, e.g. using gradient descent
Abstract
Embodiment of the disclosure discloses the method and apparatus for sending information.This method is related to field of cloud calculation, and a specific embodiment of this method includes: to obtain the classification information set of target terminal transmission as external classification information set;In response to target external classification information is not present in external classification information set, model trained in advance is determined as the disaggregated model that training obtains, wherein, target external classification information is the external classification information not matched with the internal sort information in predetermined internal sort information aggregate, trained model is obtained based on the training of predetermined training sample set in advance, and the training sample in training sample set includes internal sort information corresponding with internal data in internal data and internal sort information aggregate;The calling interface for the disaggregated model that training obtains is sent to target terminal.This embodiment improves the speed of model training, help to improve the accuracy rate and recall rate of the model that training obtains.
Description
Technical field
Embodiment of the disclosure is related to field of computer technology, and in particular to the method and apparatus for sending information.
Background technique
Artificial intelligence (Artificial Intelligence, AI) technology is with Covering domain is wide, technical threshold is high, place
Manage the features such as process is complicated.In practice, although significant progress has been obtained in the technologies such as machine learning, deep learning, its
Before being employed for solving practical problems, it is still necessary to carry out a large amount of preparation.For example, to select which kind of technology come analyze with
It handles data and needs a longer period.
In general, technical staff needs as follows, just to may be implemented the use of model: model selection, data preparation,
Model training, model measurement, model deployment etc..
Summary of the invention
The present disclosure proposes the method and apparatus for sending information.
In a first aspect, embodiment of the disclosure provides a kind of method for sending information, this method comprises: obtaining mesh
The classification information set of terminal transmission is marked as external classification information set;In response to mesh is not present in external classification information set
External classification information is marked, model trained in advance is determined as the disaggregated model that training obtains, wherein target external classification information
For the external classification information not matched with the internal sort information in predetermined internal sort information aggregate, in advance training
Model be based on predetermined training sample set training obtain, the training sample in training sample set includes internal data
With internal sort information corresponding with internal data in internal sort information aggregate;Point that training obtains is sent to target terminal
The calling interface of class model.
In some embodiments, this method further include: obtain the test sample collection that target terminal is sent via calling interface
It closes, wherein the test sample in test sample set includes the classification information of data and data;By the number in test sample set
According to the corresponding disaggregated model of calling interface is input to, the classification information of disaggregated model output is obtained;Based on disaggregated model output
Classification information in classification information and test sample set, generates at least one of following assessment information of disaggregated model: accuracy rate,
Recall rate, F1 score (F1-score);Assessment information generated is sent to target terminal.
In some embodiments, before the calling interface for sending the disaggregated model that training obtains to target terminal, the party
Method further include: in response to there are target external classification informations in external classification information set, obtain that target terminal is sent, corresponding
The target external data acquisition system of target external classification information;Using machine learning algorithm, it is based on external classification information set, target
External data set and target internal data acquisition system, are trained initial model, and will meet predetermined training terminates item
The initial model of part is determined as the disaggregated model that training obtains;Wherein, target internal data acquisition system is and external classification information collection
The corresponding internal data set of the internal sort information that external classification information in conjunction matches.
In some embodiments, before the calling interface for sending the disaggregated model that training obtains to target terminal, the party
Method further include: the internal data in quantity and target internal data acquisition system based on the external data in target external data acquisition system
Quantity, calculate remaining time, wherein remaining time be used to indicate training obtain between the time of disaggregated model and current time
Time difference;Remaining time is sent to target terminal.
In some embodiments, this method further include: it is to be sorted to obtain the target that target terminal is sent via calling interface
Data;Target data to be sorted are input to the corresponding disaggregated model of calling interface, generate the classification letter of target data to be sorted
Breath;Classification information generated is sent to target terminal.
In some embodiments, internal sort information aggregate and with the internal sort information phase in internal sort information aggregate
Corresponding internal data set is obtained after Feature Engineering is handled.
Second aspect, embodiment of the disclosure provide a kind of for sending the device of information, which includes: first to obtain
Unit is taken, is configured to obtain the classification information set of target terminal transmission as external classification information set;Determination unit, quilt
It is configured in external classification information set be determined as instructing by model trained in advance there is no target external classification information
The disaggregated model got, wherein target external classification information be not in predetermined internal sort information aggregate in
The external classification information that portion's classification information matches, it is trained that model trained in advance is based on predetermined training sample set
It arrives, the training sample in training sample set includes corresponding with internal data in internal data and internal sort information aggregate
Internal sort information;First transmission unit is configured to send the calling interface for the disaggregated model that training obtains to target terminal.
In some embodiments, device further include: second acquisition unit is configured to obtain target terminal via calling
The test sample set that interface is sent, wherein the test sample in test sample set includes the classification information of data and data;
Input unit is configured to for the data in test sample set to be input to the corresponding disaggregated model of calling interface, be classified
The classification information of model output;Generation unit is configured to the classification information and test sample set exported based on disaggregated model
In classification information, generate at least one of following assessment information of disaggregated model: accuracy rate, recall rate, F1 score;Second sends
Unit is configured to assessment information generated being sent to target terminal.
In some embodiments, device further include: third acquiring unit is configured in response to external classification information collection
There are target external classification informations in conjunction, obtain the target external number of target external classification information that target terminal is sent, corresponding
According to set;Using machine learning algorithm, it is based on external classification information set, target external data acquisition system and target internal data set
It closes, initial model is trained, the initial model for meeting predetermined trained termination condition is determined as what training obtained
Disaggregated model;Wherein, target internal data acquisition system is in matching with the external classification information in external classification information set
The corresponding internal data set of portion's classification information.
In some embodiments, device further include: computing unit is configured to based in target external data acquisition system
The quantity of external data and the quantity of the internal data in target internal data acquisition system calculate remaining time, wherein remaining time
It is used to indicate training and obtains the time difference between the time of disaggregated model and current time;Third transmission unit, be configured to
Target terminal sends remaining time.
In some embodiments, device further include: the 4th acquiring unit is configured to obtain target terminal via calling
The target data to be sorted that interface is sent;It is corresponding to be configured to for target data to be sorted being input to calling interface for input unit
Disaggregated model, generate target data to be sorted classification information;4th transmission unit is configured to send institute to target terminal
The classification information of generation.
In some embodiments, internal sort information aggregate and with the internal sort information phase in internal sort information aggregate
Corresponding internal data set is obtained after Feature Engineering is handled.
The third aspect, embodiment of the disclosure provide a kind of for sending the server of information, comprising: one or more
Processor;Storage device is stored thereon with one or more programs, when said one or multiple programs are by said one or multiple
Processor executes, so that the one or more processors realize the side of any embodiment in the method as above-mentioned for sending information
Method.
Fourth aspect, embodiment of the disclosure provide a kind of for sending the computer-readable medium of information, deposit thereon
Computer program is contained, is realized when which is executed by processor in the method as above-mentioned for sending information any embodiment
Method.
The method and apparatus for sending information that embodiment of the disclosure provides, the class sent by obtaining target terminal
Other information aggregate is as external classification information set, then, in response to target external class is not present in external classification information set
Model trained in advance is determined as the disaggregated model that training obtains by other information, wherein target external classification information is not and in advance
The first external classification information that the internal sort information in determining internal sort information aggregate matches, the model base trained in advance
It is obtained in the training of predetermined training sample set, the training sample in training sample set includes internal data and inner classes
Internal sort information corresponding with internal data in other information aggregate, finally, sending the classification that training obtains to target terminal
The calling interface of model, to improve the speed of model training, help to improve the accuracy rate for the model that training obtains and call together
The rate of returning.
Detailed description of the invention
By reading a detailed description of non-restrictive embodiments in the light of the attached drawings below, the disclosure is other
Feature, objects and advantages will become more apparent upon:
Fig. 1 is that one embodiment of the disclosure can be applied to exemplary system architecture figure therein;
Fig. 2 is the flow chart according to one embodiment of the method for sending information of the disclosure;
Fig. 3 is the schematic diagram according to an application scenarios of the method for sending information of the disclosure;
Fig. 4 is the flow chart according to another embodiment of the method for sending information of the disclosure;
Fig. 5 A- Fig. 5 F is the interactive process schematic diagram according to the target terminal of the method for sending information of the disclosure;
Fig. 6 is the structural schematic diagram according to one embodiment of the device for sending information of the disclosure;
Fig. 7 is adapted for the structural schematic diagram for the computer system for realizing the server of embodiment of the disclosure.
Specific embodiment
The disclosure is described in further detail with reference to the accompanying drawings and examples.It is understood that this place is retouched
The specific embodiment stated is used only for explaining related invention, rather than the restriction to the invention.It also should be noted that in order to
Convenient for description, part relevant to related invention is illustrated only in attached drawing.
It should be noted that in the absence of conflict, the feature in embodiment and embodiment in the disclosure can phase
Mutually combination.The disclosure is described in detail below with reference to the accompanying drawings and in conjunction with the embodiments.
Fig. 1 is shown can the method for sending information using embodiment of the disclosure or the dress for sending information
The exemplary system architecture 100 for the embodiment set.
As shown in Figure 1, system architecture 100 may include terminal device 101,102,103, network 104 and server 105.
Network 104 between terminal device 101,102,103 and server 105 to provide the medium of communication link.Network 104 can be with
Including various connection types, such as wired, wireless communication link or fiber optic cables etc..
User can be used terminal device 101,102,103 and be interacted by network 104 with server 105, to receive or send out
Send data (such as external classification information set) etc..Various client applications can be installed on terminal device 101,102,103,
Such as the application of model training class, video jukebox software, the application of Domestic News class, image processing class application, web browser applications,
Shopping class application, searching class application, instant messaging tools, mailbox client, social platform software etc..
Terminal device 101,102,103 can be hardware, be also possible to software.When terminal device 101,102,103 is hard
When part, it can be the various electronic equipments with display screen, including but not limited to smart phone, tablet computer, e-book reading
Device, MP3 player (MovingPicture Experts Group Audio Layer III, dynamic image expert's compression standard
Audio level 3), MP4 (Moving Picture Experts Group Audio Layer IV, dynamic image expert compress mark
Quasi- audio level 4) player, pocket computer on knee and desktop computer etc..When terminal device 101,102,103 is soft
When part, it may be mounted in above-mentioned cited electronic equipment.Its may be implemented into multiple softwares or software module (such as
The software or software module of Distributed Services are provided), single software or software module also may be implemented into.Specific limit is not done herein
It is fixed.
Server 105 can be to provide the server of various services, such as send to terminal device 101,102,103 outer
The background server that the data such as the other information aggregate of category are handled.Background server can determine in external classification information set
With the presence or absence of target external classification information, and in the other information aggregate of outer category be not present target external classification information the case where
Under, model trained in advance is determined as the disaggregated model that training obtains, and send and train to terminal device 101,102,103
The calling interface of obtained disaggregated model.As an example, server 105 can be cloud server, it is also possible to physical services
Device
It should be noted that server can be hardware, it is also possible to software.When server is hardware, may be implemented
At the distributed server cluster that multiple servers form, individual server also may be implemented into.It, can when server is software
To be implemented as multiple softwares or software module (such as providing the software of Distributed Services or software module), also may be implemented
At single software or software module.It is not specifically limited herein.
It should also be noted that, the method provided by embodiment of the disclosure for sending information can be held by server
Row.Correspondingly, the various pieces (such as each unit, subelement, module, submodule) that the device for sending information includes can
To be set in server.
It should be understood that the number of terminal device, network and server in Fig. 1 is only schematical.According to realization need
It wants, can have any number of terminal device, network and server.When for sending the electronics of information approach operation thereon
When equipment does not need to carry out data transmission with other electronic equipments, which can only include for sending information approach fortune
The electronic equipment (such as server) of row thereon.
With continued reference to Fig. 2, the process of one embodiment of the method for sending information according to the disclosure is shown
200.The method for being used to send information, comprising the following steps:
Step 201, the classification information set of target terminal transmission is obtained as external classification information set.
In the present embodiment, can lead to for sending the executing subject (such as server shown in FIG. 1) of the method for information
It crosses wired connection mode or radio connection obtains the classification information set of target terminal transmission as external classification information
Set.
Wherein, above-mentioned target terminal can be the terminal with the communication connection of above-mentioned executing subject.Above-mentioned target terminal is sent
Classification information set in classification information can serve to indicate that classification.As an example, classification information can serve to indicate that it is following
Any one: figure kind, plant, includes face, not comprising face etc. at animal class.
The classification information in classification information set that above-mentioned target terminal is sent can serve to indicate that the classification of video, text
Classification, the classification of image, can be used for the classification for indicating other data.
Step 202, in response to target external classification information is not present in external classification information set, by mould trained in advance
Type is determined as the disaggregated model that training obtains.
In the present embodiment, it is above-mentioned in the case where target external classification information is not present in the other information aggregate of outer category
Model trained in advance can be determined as the disaggregated model that training obtains by executing subject.
Wherein, target external classification information be not with the internal sort information in predetermined internal sort information aggregate
The external classification information to match.Trained model is obtained based on the training of predetermined training sample set in advance.Training sample
Training sample in this set includes internal sort corresponding with internal data in internal data and internal sort information aggregate
Information.Disaggregated model from external classification information set for determining the corresponding external classification information of inputted data.
Internal sort information in internal sort information aggregate is used to indicate predetermined classification.As an example, internal
Internal sort information in classification information set is used to indicate following any classification: vehicle class, figure kind, plant etc..Outside
The classification information that portion's classification information can send for target terminal.As an example, external classification information can serve to indicate that it is following
Any classification: automotive-type, landscape class, plant etc..Herein, technical staff can predefine matching rule, thus really
Determine whether internal sort information matches with external classification information.For example, above-mentioned matching rule can be " if internal sort is believed
Breath is identical as external classification information, then, the internal sort information and the outside classification information match ".Above-mentioned execution as a result,
Main body can determine the internal sort information for being used to indicate " plant " and the external classification information phase that is used to indicate " plant "
Matching.
Target internal data acquisition system are as follows: the internal sort to match with the external classification information in external classification information set
The corresponding internal data set of information.The classification of target external data in target external data acquisition system are as follows: with target external number
According to the classification for gathering corresponding external classification information instruction.
Above-mentioned model trained in advance can be obtained using following steps training:
Firstly, obtaining training sample set.Wherein, the training sample in above-mentioned training sample set includes internal sort letter
Internal sort information and the corresponding internal data of internal sort information in breath set.
Then, using machine learning algorithm, the internal data for including by the training sample in above-mentioned training sample set is made
It is for the input data of initial model, internal sort information corresponding with the internal data of input is defeated as the expectation of initial model
Data out are trained initial model, so that the initial model for meeting predetermined trained termination condition is determined as instructing
The model got.
Wherein, above-mentioned initial model may include various model structures, for example, AlexNet, ZFNet, OverFeat, VGG
(Visual Geometry Group) Network etc..As an example, initial model can be convolutional neural networks.Above-mentioned instruction
White silk termination condition can include but is not limited at least one of following: training duration is more than to preset more than preset duration, frequency of training
Number is less than preset threshold based on the functional value that predetermined loss function is calculated.
It is appreciated that when initial model is unsatisfactory for above-mentioned trained termination condition, it can be by adjusting the mould of initial model
Shape parameter, so that initial model meets above-mentioned training and terminates, so that training obtains above-mentioned model.
In practice, above-mentioned model trained in advance can using batch gradient decline (BatchGradient Descent,
BGD), stochastic gradient descent (Stochastic Gradient Descent, SGD), most small quantities of gradient decline (Mini-batch
Gradient Descent) scheduling algorithm, Lai Xunlian initial model.
In some optional implementations of the present embodiment, above-mentioned executing subject can also be by executing following any step
Suddenly (i.e. step 1 or step 2), to determine in external classification information set with the presence or absence of target external classification information:
Step 1, for the external classification information in external classification information set, in response to determining internal sort information collection
In conjunction, there is no the internal sort information that incidence relation is pre-established with the outside classification information, determine external classification information collection
There are target external classification informations in conjunction.
Herein, in the local or electronic equipment with the communication connection of above-mentioned executing subject of above-mentioned executing subject, Ke Yicun
Internal sort information aggregate is contained, and for each internal sort information in above-mentioned internal sort information aggregate, this is interior
Portion's classification information and external classification information are associated storage.As a result, in determining internal sort information aggregate there is no with it is outer
In the case that portion's classification information pre-establishes the internal sort information of incidence relation, above-mentioned executing subject can determine external classification
There are target external classification informations in information aggregate;Exist in each external classification information and pre-establishes incidence relation with it
In the case where internal sort information, above-mentioned executing subject can determine that there is no target external classifications in external classification information set
Information.
Step 2, for the external classification information in external classification information set, in response to determining internal sort information collection
In conjunction, there is no the internal sort information for being more than or equal to preset threshold with the similarity of the outside classification information, determine outer
There are target external classification informations in the other information aggregate of category.
Herein, above-mentioned executing subject can adopt in various manners come determine external classification information and internal sort information it
Between similarity, for example, deep structure semantic model (Deep StructuredSemantic Models, DSSM), cosine phase
Like property etc..
It is appreciated that in determining internal sort information aggregate there is no being greater than with the similarity of the outside classification information or
In the case that person is equal to the internal sort information of preset threshold, above-mentioned executing subject can determine deposits in external classification information set
In target external classification information;Similarity between each external classification information and internal sort information is respectively less than above-mentioned default
In the case where threshold value, above-mentioned executing subject can determine that there is no target external classification informations in external classification information set.
Step 203, the calling interface for the disaggregated model that training obtains is sent to target terminal.
In the present embodiment, above-mentioned executing subject can dispose disaggregated model after obtaining disaggregated model, with
And calling interface (API, the Application Programming for the disaggregated model that training obtains is sent to target terminal
Interface).Wherein, calling interface may include the address for disposing disaggregated model and port numbers.
Optionally, disaggregated model can be deployed to public cloud, private clound, among embedded device by above-mentioned executing subject.
It is appreciated that user can use instruction by the calling interface after target terminal receives calling interface
The disaggregated model got.
In practice, deployment disaggregated model be may include steps of: disaggregated model is packaged, by the disaggregated model after packing
It is uploaded to memory (such as BOS), disposes Clustering OS (such as matrix).
In some optional implementations of the present embodiment, following steps are can also be performed in above-mentioned executing subject:
Step 1 obtains the test sample set that target terminal is sent via calling interface.Wherein, in test sample set
Test sample include data and data classification information.As an example, the data that test sample includes can be video, test
The classification information that sample includes can be " figure kind ", it will be understood that the classification information " figure kind " that the test sample includes can
To be used to indicate the classification for the video that the test sample includes as " figure kind ".
Data in test sample set are input to the corresponding disaggregated model of calling interface by step 2, obtain classification mould
The classification information of type output.
It is appreciated that herein, the classification information that above-mentioned steps two obtain is disaggregated model in test sample set
Data obtained output data after calculating.
Step 3, the classification information in classification information and test sample set exported based on disaggregated model, generates classification
At least one following assessment information of model: accuracy rate, recall rate, F1 score.
Assessment information generated is sent to target terminal by step 4.
It is appreciated that this optional implementation can send assessment information generated to target terminal, for using
The user of target terminal assesses obtained disaggregated model by assessing information, so that it is determined that whether disaggregated model meets
Actual demand, and then determine and continue train classification models, re -training disaggregated model or begin to use disaggregated model.As a result,
Help to obtain disaggregated model that is corresponding, meeting user demand for the different demands training of user, enriches the instruction of model
The mode of white silk helps to improve in the accuracy rate, recall rate, F1 score of model under the premise of reducing the training duration of model
At least one of.
In some optional implementations of the present embodiment, before executing above-mentioned steps 203, above-mentioned executing subject is also
Following steps can be executed:
There are in the case where target external classification information in the other information aggregate of outer category, obtain it is that target terminal is sent,
The target external data acquisition system of corresponding target external classification information.Then, using machine learning algorithm, it is based on external classification information
Set, target external data acquisition system and target internal data acquisition system, are trained initial model, will meet predetermined instruction
The initial model for practicing termination condition is determined as the disaggregated model that training obtains.Wherein, target internal data acquisition system are as follows: with outer category
The set of the corresponding internal data composition of the internal sort information that external classification information in other information aggregate matches.
It is appreciated that herein, each target external data acquisition system can correspond to an external classification information, different mesh
Mark external data set can correspond to different external classification informations.In practice, above-mentioned executing subject is local or holds with above-mentioned
In the electronic equipment of row main body communication connection, multiple internal data set and internal sort information aggregate can be previously stored with.
Each internal data set can correspond to an internal sort information in internal sort information aggregate, which uses
In the classification for indicating each internal data in the internal data set, in other words, internal data in internal data set
Classification is the classification of internal sort information corresponding with internal data set instruction.
The class of target external data in the target external data acquisition system of target external classification information that terminal is sent, corresponding
Not, it can be the classification of target external classification information instruction.
In some optional implementations of the present embodiment, above-mentioned executing subject can be obtained using following steps training
Disaggregated model:
Step 1 obtains training sample set.Wherein, the training sample that above-mentioned training sample is concentrated includes target data and mesh
Mark the classification information of data.Target data are as follows: the target external data in above-mentioned target external data acquisition system, alternatively, in target
Target internal data in portion's data acquisition system.The classification information of target data is the external classification letter in external classification information set
Breath.
Step 2, using machine learning algorithm, the target data for including using the training sample that training sample is concentrated is as defeated
Enter data, will classification information corresponding with input data as desired output data, initial model is trained, thus will
The initial model for meeting predetermined trained termination condition is determined as the disaggregated model that training obtains.
Wherein, above-mentioned initial model may include various model structures, for example, AlexNet, ZFNet, OverFeat, VGG
(Visual Geometry Group) Network etc..As an example, initial model can be convolutional neural networks.Above-mentioned instruction
White silk termination condition can include but is not limited at least one of following: training duration is more than to preset more than preset duration, frequency of training
Number is less than preset threshold based on the functional value that predetermined loss function is calculated.
It is appreciated that when initial model is unsatisfactory for above-mentioned trained termination condition, it can be using gradient descent method, reversed biography
Method scheduling algorithm is broadcast to adjust the model parameter of initial model.
In some optional implementations of the present embodiment, above-mentioned executing subject can also be trained using following steps
To disaggregated model:
The first step obtains the model of pre-training.Wherein, the hyper parameter of the model of pre-training is the mesh in set of object models
The optimal hyper parameter of model is marked, for example, the hyper parameter of the most fast object module of pace of learning.Target mould in set of object models
Type is to carry out hyper parameter tune to initial model using the hyper parameter adjustment mode in predetermined multiple hyper parameter adjustment modes
The model obtained after whole.Wherein, hyper parameter can include but is not limited at least one of following: learning rate, regularization parameter, nerve
The number of neuron, the rounds of study, the size of small lot data, output neuron in the number of plies of network, each hidden layer
Coding mode, the selection of cost function, the method for weights initialisation, the type of neuron activation functions, participate in training pattern
Scale of data etc..Here, for different initial models, different hyper parameters can be correspondingly set.Above-mentioned hyper parameter
Adjustment mode may include at least one of following: grid search, Bayes's optimization, random search, the optimization based on gradient etc..
Second step, using machine learning algorithm, based in external classification information set, target external data acquisition system and target
Portion's data acquisition system is trained the model of pre-training, obtains disaggregated model.
Specifically, above-mentioned executing subject can be using target data as the input data of the model of pre-training, will be with input
The desired output data of model of the external classification information as pre-training in the corresponding target internal data acquisition system of data are right
The model of pre-training is trained, so that the model for the pre-training for meeting predetermined trained termination condition is determined as training
Obtained disaggregated model.
Wherein, above-mentioned initial model may include various model structures, for example, AlexNet, ZFNet, OverFeat, VGG
(Visual Geometry Group) Network etc..As an example, initial model can be convolutional neural networks.Above-mentioned instruction
White silk termination condition can include but is not limited at least one of following: training duration is more than to preset more than preset duration, frequency of training
Number is less than preset threshold based on the functional value that predetermined loss function is calculated.
It, can be using gradient descent method, anti-it is appreciated that when the model of pre-training is unsatisfactory for above-mentioned trained termination condition
The model parameter of the model of pre-training is adjusted to Law of Communication scheduling algorithm.
It should be appreciated that in this optional implementation, can using a variety of hyper parameter adjustment modes to initial model into
Row training, to obtain multiple object modules, then chosen from obtained multiple object modules it is optimal (such as accuracy rate and
In recall rate at least one of it is maximum, alternatively, pace of learning is most fast) model of the object module as pre-training, to improve
Subsequent training obtains the speed of disaggregated model, alternatively, improve subsequent training obtain the accuracy rate of disaggregated model, recall rate and
At least one of in F1 score.
In some optional implementations of the present embodiment, before executing above-mentioned steps 203, above-mentioned executing subject is also
Following steps can be executed:
Step 1, in quantity and target internal data acquisition system based on the external data in target external data acquisition system in
The quantity of portion's data calculates the remaining time for generating disaggregated model.Wherein, remaining time is used to indicate training and obtains disaggregated model
Time and current time between time difference.
As an example, above-mentioned remaining time can be calculated in the following way:
Firstly, technical staff can determine quantity for characterizing training sample and trained by a large amount of statistics
Corresponding relationship between the duration of disaggregated model.Here, above-mentioned corresponding relationship can pass through bivariate table, curve graph or line chart
Etc. forms characterization.Wherein, the quantity of training sample is the quantity and target internal of the external data in target external data acquisition system
The sum of quantity of internal data in data acquisition system.Wherein, current time can serve to indicate that the time for executing the step 1.
Then, when above-mentioned executing subject can determine that the quantity of current training sample is corresponding according to above-mentioned corresponding relationship
It is long.Wherein, the quantity of current training sample is the quantity and target of the external data in step 1 in target external data acquisition system
The sum of the quantity of internal data in internal data set.
Finally, the duration determined and the difference for having been subjected to duration can be determined as remaining time by above-mentioned executing subject.
Wherein, duration characterization is had been subjected to from the duration for having got training sample set and being passed through to current time.Wherein, above-mentioned current
Time can be the time for executing above-mentioned steps one.
Optionally, above-mentioned executing subject can also in the following way, to calculate above-mentioned remaining time:
Firstly, calculating the inside in the quantity and target internal data acquisition system of the external data in target external data acquisition system
The sum of quantity of data is used as training samples number.
Then, the dimension of the external data in target external data acquisition system is calculated, and calculates target internal data acquisition system
In internal data dimension.
Later, by the sum of above-mentioned two dimension, the product of training samples number and preset duration three, it is total to be determined as training
Duration.Wherein, preset duration is a pre-set training sample train classification models institute for characterizing using single dimension
The duration used.
Finally, the training total duration determined and the difference for having been subjected to duration can be determined as remaining by above-mentioned executing subject
The remaining time.Wherein, duration characterization is had been subjected to from the duration for having got training sample set and being passed through to current time.Wherein,
Above-mentioned current time can be the time for executing above-mentioned steps one.
Step 2 sends the remaining time being calculated in above-mentioned steps one to target terminal.
It is appreciated that this optional implementation can send residual time length to user terminal, user can be known as a result,
Obtain the specific time of disaggregated model.
In some optional implementations of the present embodiment, following steps are can also be performed in above-mentioned executing subject:
Step 1 obtains the target data to be sorted that target terminal is sent via calling interface.Wherein, above-mentioned target waits for point
Class data can be the data to classify to it.In practice, target data to be sorted can be with above-mentioned internal data, outer
Portion's data are the data of same type.For example, if internal data, external data are text, target data to be sorted
It can be text;If internal data, external data are image, target data to be sorted may be image.
Target data to be sorted are input to the corresponding disaggregated model of calling interface, generate target number to be sorted by step 2
According to classification information.
Step 3 sends classification information generated to target terminal.
It is appreciated that the disaggregated model that this optional implementation is obtained by training, target terminal is connect via calling
The target data to be sorted that mouth is sent are classified, to generate the classification information of target data to be sorted, and whole to target
End sends classification information generated.As a result, it is not necessary that disaggregated model is stored in target terminal used by a user, can be realized
User reduces the resource occupation of target terminal to the calling of disaggregated model, saves the computing resource of target terminal, reduces
The hardware deterioration of target terminal.
Optionally, before the classification information for generating target data to be sorted, above-mentioned executing subject can also be verified,
To improve the accuracy of the classification information of final output.
In some optional implementations of the present embodiment, internal sort information aggregate and with internal sort information aggregate
In the corresponding internal data set of internal sort information be to be obtained after Feature Engineering is handled.
It is appreciated that Feature Engineering processing can include but is not limited to it is at least one of following: feature construction, feature extraction,
Feature selecting etc..In general, model training personnel require a great deal of time with energy to training sample before training pattern
This progress Feature Engineering processing, so that it is relatively excellent to train the accuracy rate of obtained model, recall rate etc. to have
Performance.This optional implementation can be local in above-mentioned executing subject or be set with the electronics of above-mentioned executing subject communication connection
In standby, it is stored in advance by Feature Engineering treated internal sort information aggregate and internal data set, thus, it is possible to make
The user of target terminal needs not move through Feature Engineering processing, can be obtained using Feature Engineering treated internal sort information collection
It closes and internal data set trains obtained disaggregated model.The data that user face when model training are avoided as a result, to obtain
The obstacle for taking aspect, under the premise of the formation efficiency of raising model, has ensured mould at the step of simplifying trained and calling model
The accuracy rate of type.
With continued reference to the signal that Fig. 3, Fig. 3 are according to the application scenarios of the method for sending information of the present embodiment
Figure.In the application scenarios of Fig. 3, server 301 obtains the classification information set 303 of the transmission of target terminal 302 as outer first
The other information aggregate 303 of category.Then, in the case where target external classification information is not present in the other information aggregate 303 of outer category,
Model 305 trained in advance is determined as the disaggregated model 305 that training obtains by server 301.Wherein, target external classification information
For the external classification information not matched with the internal sort information in predetermined internal sort information aggregate 304.In advance
Trained model 305 is based on the training of predetermined training sample set and obtains.Training sample in training sample set includes
Internal sort information and the corresponding internal data of internal sort information in internal sort information aggregate 304.Finally, server
301 send the calling interface 306 for the disaggregated model that training obtains to target terminal 302.
In the prior art, before using model, it usually needs user carries out following steps first, and model just may be implemented
Use: model selection, data preparation, model training, model measurement, model deployment etc..
The method provided by the above embodiment of the disclosure, by obtaining the classification information set of target terminal transmission as outer
The other information aggregate of category then, will be preparatory in the case where target external classification information is not present in the other information aggregate of outer category
Trained model is determined as the obtained disaggregated model of training, wherein target external classification information be not with predetermined inside
The external classification information that internal sort information in classification information set matches, trained model is based on predetermined in advance
The training of training sample set obtains, and the training sample in training sample set includes the internal sort in internal sort information aggregate
Information and the corresponding internal data of internal sort information, finally, sending the calling for the disaggregated model that training obtains to target terminal
Interface, user need to only upload the data (such as classification information set) for training pattern as a result, that is, produce to data according to
The disaggregated model that the classification for the classification information instruction that user uploads is classified, improves the speed of model training, helps to mention
The accuracy rate and recall rate for the model that height training obtains, also, the method provided by the above embodiment of the disclosure will be without that will classify
Model is stored in target terminal used by a user, and user can be realized to the calling of disaggregated model, reduce target terminal
Resource occupation saves the computing resource of target terminal, reduces the hardware deterioration of target terminal.
With further reference to Fig. 4, it illustrates the processes 400 of another embodiment of the method for sending information.The use
In the process 400 for the method for sending information, comprising the following steps:
Step 401, the classification information set of target terminal transmission is obtained as external classification information set.Later, it executes
Step 402.
In the present embodiment, step 401 and the step 201 in Fig. 2 corresponding embodiment are almost the same, and which is not described herein again.
Step 402, determining in external classification information set whether there is target external classification information.Later, if so, holding
Row step 404;If it is not, thening follow the steps 403.
In the present embodiment, above-mentioned executing subject can use and side described in optional implementation shown in Fig. 2
Method whether there is target external classification information to determine in external classification information set, details are not described herein.
Step 403, model trained in advance is determined as the disaggregated model that training obtains.Later, step 406 is executed.
In the present embodiment, model trained in advance can be determined as the classification mould that training obtains by above-mentioned executing subject
Type.Above-mentioned executing subject can be using method described in Fig. 2, and the model that Lai Xunlian is trained in advance, details are not described herein.
It is appreciated that in the case where target external classification information is not present in the other information aggregate of outer category, above-mentioned execution
Model trained in advance directly can be determined as the disaggregated model that training obtains by main body.Also, due to model trained in advance
Internal sort information aggregate can be in advance based on and the training of internal data set obtains, it is above-mentioned to hold as a result, under this application scenarios
Row main body no longer needs to carry out model training, and the disaggregated model met the needs of users can be obtained, which thereby enhance the life of model
At efficiency.
Step 404, the target external data acquisition system of target external classification information that target terminal is sent, corresponding is obtained.It
Afterwards, step 405 is executed.
In the present embodiment, the available target terminal of above-mentioned executing subject is sending, corresponding target external classification information
Target external data acquisition system.It is appreciated that herein, each target external data acquisition system can correspond to an external classification letter
Breath, different target external data acquisition systems can correspond to different external classification informations.In practice, above-mentioned executing subject it is local or
In the electronic equipment of person and the communication connection of above-mentioned executing subject, multiple internal data set and internal sort can be previously stored with
Information aggregate.Each internal data set can correspond to an internal sort information in internal sort information aggregate, inside this
Classification information is used to indicate the classification of each internal data in the internal data set, in other words, in internal data set
The classification of internal data is the classification of internal sort information corresponding with internal data set instruction.
It is appreciated that internal sort information aggregate and corresponding with the internal sort information in internal sort information aggregate
Internal data set, which can be, to be obtained after Feature Engineering is handled.For example, internal sort information aggregate and internal data collection
Conjunction can be structural data (for example, data that realization is expressed with two-dimentional table structure).Thus, it is possible to which user is avoided to carry out mould
The data acquisition obstacle faced when type training, avoids user and starts from scratch and carry out algorithm investigation, Scheme Choice, arameter optimization etc.
Compared with the work of high-tech threshold, the formation efficiency of model can be improved.
Step 405, using machine learning algorithm, external classification information set, target external data acquisition system and target are based on
Internal data set, is trained initial model, and the initial model for meeting predetermined trained termination condition is determined as
The disaggregated model that training obtains.Later, step 406 is executed.
In the present embodiment, above-mentioned executing subject can use machine learning algorithm, be based on external classification information set, mesh
External data set and target internal data acquisition system are marked, initial model is trained, will meet predetermined training terminates
The initial model of condition is determined as the disaggregated model that training obtains.Wherein, terminal transmission, correspondence target external classification information
The classification of target external data in target external data acquisition system can be the classification of target external classification information instruction.
It is appreciated that being target external classification information in the external classification information of each of other information aggregate of outer category
In the case of, target internal data acquisition system be empty set, as a result, above-mentioned executing subject can be directly based upon external classification information set and
The training of external data set obtains disaggregated model;The external classification information in part (and not all) in the other information aggregate of outer category
In the case where for target external classification information, above-mentioned executing subject can be based on external classification information set and external data set
Training obtains disaggregated model.In this way, in external classification information set and in predetermined internal sort information aggregate
The external classification information that internal sort information matches, above-mentioned executing subject can be using matching with the outside classification information
The corresponding internal data set of internal sort information, instead of external data set corresponding with the outside classification information, to instruct
Practice disaggregated model, as a result, in the case where the set of the data obtained after target internal data acquisition system is characterized project treatment, from
Above-mentioned executing subject gets external data set and external classification information, to generate disaggregated model usually only need 10 hours with
It is interior, to improve the speed of model generation under the premise of ensuring to train the accuracy rate of obtained disaggregated model.
In some optional implementations of the present embodiment, following steps are can also be performed in above-mentioned executing subject:
In the case where receiving the target modification information of target terminal transmission, using target external data acquisition system, modification
External classification information set and internal number corresponding with the internal sort information that modified external classification information matches afterwards
According to set, re -training obtains disaggregated model.Wherein, target modification information is corresponding for modifying external data or internal data
External classification information.
Herein, above-mentioned executing subject can be using target data as the input data of initial model, will be with target data
Desired output data of the corresponding modified external classification information as initial model, re -training obtain disaggregated model.Its
In, target data is the target external data in target external data acquisition system, alternatively, the inside number in above-mentioned internal data set
According to.
It is appreciated that the mode that re -training obtains disaggregated model can be with instruction above in this optional implementation
The mode for getting disaggregated model is almost the same, and details are not described herein.
It should be appreciated that this optional implementation can modify external data or internal number by target terminal in user
After corresponding external classification information, re -training obtains the disaggregated model for meeting the new demand of user.Mould is enriched as a result,
The mode of type training can train the disaggregated model to be met the needs of different users.Later, above-mentioned executing subject can also be to
Target user sends the calling interface for the disaggregated model that re -training obtains, with the classification that the re -training obtains for users to use
Model.
Step 406, the calling interface for the disaggregated model that training obtains is sent to target terminal.
In the present embodiment, above-mentioned executing subject can dispose disaggregated model after obtaining disaggregated model, with
And the calling interface for the disaggregated model that training obtains is sent to target terminal.Wherein, calling interface may include dividing for disposing
The address of class model and port numbers.
It is appreciated that user can use instruction by the calling interface after target terminal receives calling interface
The disaggregated model got.In the case where target external classification information is not present in the other information aggregate of outer category, above-mentioned steps
Calling interface sent in 406 is the calling interface of the disaggregated model in step 403, i.e. the calling of model trained in advance connects
Mouthful;There are in the case where target external classification information in the other information aggregate of outer category, calling sent in above-mentioned steps 406
Interface is the calling interface for the disaggregated model that training obtains in step 405.
In practice, deployment disaggregated model be may include steps of: disaggregated model is packaged, by the disaggregated model after packing
It is uploaded to memory (such as BOS), disposes Clustering OS (such as matrix).
It is appreciated that the disaggregated model that this optional implementation is trained can be used for classifying, detecting, identify etc.
Scene.
In some optional implementations of the present embodiment, following steps are can also be performed in above-mentioned executing subject:
Step 1 generates the identification information for the disaggregated model that training obtains.Wherein, above-mentioned identification information can be used for identifying
Disaggregated model is also possible to the version of disaggregated model as an example, identification information can be the timestamp for generating disaggregated model
Number.
Step 2, each storage data of associated storage, wherein each storage data include following at least two: point
The data acquisition system and external classification letter of the identification information of class model, the disaggregated model for training obtained disaggregated model, training to obtain
Breath set, trains the data in the data acquisition system of obtained disaggregated model to be one of the following: internal data or external data.
Step 3, based on the storage data received from target terminal, to target terminal send it is being found, with connect
The storage data that the storage received is stored with data correlation.
It is appreciated that due to the identification information of associated storage disaggregated model, obtained disaggregated model, training is trained to obtain
Disaggregated model data acquisition system and external classification information set at least two item datas (i.e. storage data), thus, this
Optional implementation can be searched and be stored with the storage with data correlation according to the storage data received from target terminal
Other storage data.For example, when what the above-mentioned executing subject associated storage identification information of disaggregated model, training obtained divides
When the data acquisition system for the disaggregated model that class model, training obtain and external classification information set, above-mentioned executing subject can basis
The identification information of the disaggregated model received from target terminal finds the instruction with the identification information associated storage of the disaggregated model
The data acquisition system of the disaggregated model, trained obtained disaggregated model that get and external classification information set, thus, it is possible to use
Know training sample employed in the iteration situation or training process of disaggregated model in family.
It should be noted that the embodiment of the present application can also include reality corresponding with Fig. 2 in addition to documented content above
The same or similar feature of example, effect are applied, details are not described herein.
Turn next to the mesh for the method for sending information that Fig. 5 A- Fig. 5 F, Fig. 5 A- Fig. 5 F is shown according to the disclosure
Mark the interactive process schematic diagram of terminal.
As shown in Figure 5A, during creating model, user first is to target terminal input model title " short-sighted frequency division
Class ", category of model " scene classification ".It is appreciated that the demand of user is to carry out scene point for short-sighted frequency under this application scenarios
Class.
Later, Fig. 5 B is please referred to, it (is " life, science and technology, joy in Fig. 5 B that user, which uploads classification information set to target terminal,
It is happy ") it is used as external classification information set.Later, if target external classification information is not present in external classification information set, on
It states executing subject and model trained in advance is determined as the disaggregated model that training obtains.Wherein, target external classification information is not
The external classification information to match with the internal sort information in predetermined internal sort information aggregate.Trained mould in advance
Type is based on the training of predetermined training sample set and obtains.Training sample in training sample set includes internal sort information
Internal sort information and the corresponding internal data of internal sort information in set.If in external classification information set, there are targets
External classification information, then above-mentioned executing subject obtains the target external of target external classification information that target terminal is sent, corresponding
Data acquisition system.Then, using machine learning algorithm, based in external classification information set, target external data acquisition system and target
Portion's data acquisition system, is trained initial model, and the initial model for meeting predetermined trained termination condition is determined as instructing
The disaggregated model got.As shown in Figure 5 B, in the other information aggregate of outer category there are in the case where target external classification information,
Target terminal sends external data set " video 1, video 2, video 3, video 4, video 5, video 6, view to above-mentioned executing subject
Frequently 7 ... ".
As shown in Figure 5 C, for each external data in external data set, user is from external classification information set
Choose the external classification information of the external data.Target terminal obtains corresponding target external classification information (in diagram 5B as a result,
Target external classification information be " life ") target external data acquisition system (such as video 1, video 3 ...).
Hereafter, above-mentioned executing subject starts training pattern, and as shown in Figure 5 D, the current residual time is " 10 minutes ", characterization
Disaggregated model is completed in training to above-mentioned executing subject after 10 min.
As shown in fig. 5e, after disaggregated model is completed in training, essential information, test report, each mark is presented in target terminal
Sign the information such as (i.e. classification) Accuracy Analysis.
Finally, as illustrated in figure 5f, target terminal has received the disaggregated model that above-mentioned executing subject is sent, training obtains
Calling interface " http://xxx.xx.com/xx:8088 ".
Fig. 4 is returned below, figure 4, it is seen that being used to send out in the present embodiment compared with the corresponding embodiment of Fig. 2
The process 400 of method of breath of delivering letters highlights in the other information aggregate of outer category there are in the case where target external classification information,
The step of training obtains disaggregated model.The classification information collection that the scheme of the present embodiment description can be sent in target terminal as a result,
There are in the case where target external classification information in conjunction, it is based on external classification information set, target external data acquisition system and target
Internal data set, training obtains disaggregated model, to enrich the training method of model, also, in target internal data set
In the case where the set for being combined into the data obtained after Feature Engineering processing, this optional implementation can be improved what training obtained
The accuracy rate and speed of disaggregated model.
With further reference to Fig. 6, as the realization to method shown in above-mentioned each figure, present disclose provides one kind for sending letter
One embodiment of the device of breath, the Installation practice is corresponding with embodiment of the method shown in Fig. 2, except following documented special
Sign is outer, which can also include feature identical or corresponding with embodiment of the method shown in Fig. 2, and generates and scheme
Embodiment of the method shown in 2 is identical or corresponding effect.The device specifically can be applied in various electronic equipments.
As shown in fig. 6, the device 600 for sending information of the present embodiment includes: first acquisition unit 601, determines list
Member 602 and the first transmission unit 603.Wherein, first acquisition unit 601 are configured to obtain the classification letter of target terminal transmission
Breath set is used as external classification information set;Determination unit 602 is configured in response to be not present in external classification information set
Model trained in advance is determined as the disaggregated model that training obtains by target external classification information, wherein target external classification letter
Breath is the external classification information not matched with the internal sort information in predetermined internal sort information aggregate, is instructed in advance
Experienced model is based on the training of predetermined training sample set and obtains, and the training sample in training sample set includes inner classes
Internal sort information and the corresponding internal data of internal sort information in other information aggregate;First transmission unit 603, is configured
At the calling interface for sending the disaggregated model that training obtains to target terminal.
It in the present embodiment, can be by wired connection side for sending the first acquisition unit 601 of the device 600 of information
Formula or radio connection obtain the classification information set of target terminal transmission as external classification information set.
Wherein, above-mentioned target terminal can be the terminal communicated to connect with above-mentioned apparatus 600.What above-mentioned target terminal was sent
Classification information in classification information set can serve to indicate that classification.As an example, classification information can serve to indicate that following
One: figure kind, animal class, plant etc..The classification information in classification information set that above-mentioned target terminal is sent can be with
It is used to indicate the classification of video, the classification of text, the classification of image, can be used for the classification for indicating other data.
In the present embodiment, it in the case where target external classification information is not present in the other information aggregate of outer category, determines
Model trained in advance can be determined as the disaggregated model that training obtains by unit 602.Wherein, target external classification information is not
The external classification information to match with the internal sort information in predetermined internal sort information aggregate, the mould trained in advance
Type is based on the training of predetermined training sample set and obtains, and the training sample in training sample set includes internal sort information
Internal sort information and the corresponding internal data of internal sort information in set.
Wherein, target external classification information be not with the internal sort information in predetermined internal sort information aggregate
The external classification information to match.Trained model is obtained based on the training of predetermined training sample set in advance.Training sample
Training sample in this set include internal sort information in internal sort information aggregate and internal sort information it is corresponding in
Portion's data.Disaggregated model from external classification information set for determining the corresponding external classification information of inputted data.
Internal sort information in internal sort information aggregate is used to indicate predetermined classification.As an example, internal
Internal sort information in classification information set is used to indicate following any classification: vehicle class, figure kind, plant etc..Outside
The classification information that portion's classification information can send for target terminal.As an example, external classification information can serve to indicate that it is following
Any classification: automotive-type, landscape class, plant etc..
In the present embodiment, above-mentioned first transmission unit 603 can send the training of determination unit 602 to target terminal and obtain
Disaggregated model calling interface.Wherein, calling interface may include the address for disposing disaggregated model and port numbers.
In some optional implementations of the present embodiment, the device 600 further include: second acquisition unit is (in figure not
Show) it is configured to obtain the test sample set that target terminal is sent via calling interface, wherein in test sample set
Test sample includes the classification information of data and data.Input unit (not shown) is configured to will be in test sample set
Data be input to the corresponding disaggregated model of calling interface, obtain disaggregated model output classification information.Generation unit is (in figure not
Show) it is configured to the classification information in the classification information and test sample set based on disaggregated model output, generate classification mould
At least one following assessment information of type: accuracy rate, recall rate, F1 score.Second transmission unit (not shown) is configured
Target terminal is sent at by assessment information generated.
In some optional implementations of the present embodiment, the device 600 further include: third acquiring unit is (in figure not
Show) be configured in response in external classification information set there are target external classification information, obtain it is that target terminal is sent,
The target external data acquisition system of corresponding target external classification information;Using machine learning algorithm, based on external classification information set,
Target external data acquisition system and target internal data acquisition system, are trained initial model, will meet predetermined training knot
The initial model of beam condition is determined as the disaggregated model that training obtains;Wherein, target internal data acquisition system is to believe with external classification
The corresponding internal data set of internal sort information that external classification information in breath set matches.
In some optional implementations of the present embodiment, the device 600 further include: computing unit (not shown)
The internal data being configured in the quantity based on the external data in target external data acquisition system and target internal data acquisition system
Quantity, calculate remaining time, wherein remaining time be used to indicate training obtain between the time of disaggregated model and current time
Time difference.Third transmission unit (not shown) is configured to send remaining time to target terminal.
In some optional implementations of the present embodiment, the device 600 further include: the 4th acquiring unit is (in figure not
Show) it is configured to obtain the target data to be sorted that target terminal is sent via calling interface.Input unit (does not show in figure
It is configured to for target data to be sorted being input to the corresponding disaggregated model of calling interface out), generates target data to be sorted
Classification information.4th transmission unit (not shown) is configured to send classification information generated to target terminal.
In some optional implementations of the present embodiment, internal sort information aggregate and with internal sort information aggregate
In the corresponding internal data set of internal sort information be to be obtained after Feature Engineering is handled.
The device provided by the above embodiment of the disclosure obtains the class that target terminal is sent by first acquisition unit 601
Other information aggregate is as external classification information set, then, in response to target external class is not present in external classification information set
Model trained in advance is determined as the disaggregated model that training obtains by other information, determination unit 602, wherein target external classification
Information is the external classification information not matched with the internal sort information in predetermined internal sort information aggregate, in advance
Trained model is based on the training of predetermined training sample set and obtains, and the training sample in training sample set includes inside
Internal sort information and the corresponding internal data of internal sort information in classification information set, finally, the first transmission unit 603
The calling interface for the disaggregated model that training obtains is sent to target terminal, user need to only upload the number for training pattern as a result,
According to (such as classification information set), i.e., what the classification of the producible classification information instruction uploaded to data according to user was classified
Disaggregated model improves the speed of model training, helps to improve the accuracy rate and recall rate of the model that training obtains, also,
The method provided by the above embodiment of the disclosure is not necessarily to disaggregated model being stored in target terminal used by a user, can be realized
User reduces the resource occupation of target terminal to the calling of disaggregated model, saves the computing resource of target terminal, reduces
The hardware deterioration of target terminal.
Below with reference to Fig. 7, it illustrates the computer systems 700 for the server for being suitable for being used to realize embodiment of the disclosure
Structural schematic diagram.Server shown in Fig. 7 is only an example, function to embodiment of the disclosure and should not use model
Shroud carrys out any restrictions.
As shown in fig. 7, computer system 700 includes central processing unit (CPU) 701, it can be read-only according to being stored in
Program in memory (ROM) 702 or be loaded into the program in random access storage device (RAM) 703 from storage section 708 and
Execute various movements appropriate and processing.In RAM 703, also it is stored with system 700 and operates required various programs and data.
CPU 701, ROM702 and RAM 703 are connected with each other by bus 704.Input/output (I/O) interface 705 is also connected to always
Line 704.
I/O interface 705 is connected to lower component: the importation 706 including keyboard, mouse etc.;It is penetrated including such as cathode
The output par, c 707 of spool (CRT), liquid crystal display (LCD) etc. and loudspeaker etc.;Storage section 708 including hard disk etc.;
And the communications portion 709 of the network interface card including LAN card, modem etc..Communications portion 709 via such as because
The network of spy's net executes communication process.Driver 710 is also connected to I/O interface 705 as needed.Detachable media 711, such as
Disk, CD, magneto-optic disk, semiconductor memory etc. are mounted on as needed on driver 710, in order to read from thereon
Computer program be mounted into storage section 708 as needed.
Particularly, in accordance with an embodiment of the present disclosure, it may be implemented as computer above with reference to the process of flow chart description
Software program.For example, embodiment of the disclosure includes a kind of computer program product comprising be carried on computer-readable medium
On computer program, which includes the program code for method shown in execution flow chart.In such reality
It applies in example, which can be downloaded and installed from network by communications portion 709, and/or from detachable media
711 are mounted.When the computer program is executed by central processing unit (CPU) 701, limited in execution disclosed method
Above-mentioned function.
It should be noted that computer-readable medium described in the disclosure can be computer-readable signal media or meter
Calculation machine readable storage medium storing program for executing either the two any combination.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 any above combination.Meter
The more specific example of calculation machine readable storage medium storing program for executing can include but is not limited to: have the electrical connection, just of one or more conducting wires
Taking formula computer disk, hard disk, random access storage device (RAM), read-only memory (ROM), erasable type 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 disclosure, computer readable storage medium can be it is any include or storage journey
The tangible medium of sequence, the program can be commanded execution system, device or device use or in connection.And at this
In open, computer-readable signal media may include in a base band or as the data-signal that carrier wave a part is propagated,
Wherein carry computer-readable program code.The data-signal of this propagation can take various forms, 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 the use of instruction execution system, device or device or program in connection.Include on computer-readable medium
Program code can transmit with any suitable medium, including but not limited to: wireless, electric wire, optical cable, RF etc. are above-mentioned
Any appropriate combination.
The calculating of the operation for executing the disclosure can be write with one or more programming languages or combinations thereof
Machine program code, described program design language include object-oriented programming language-such as Python, Java,
Smalltalk, C++ further include conventional procedural programming language-such as " C " language or similar program design language
Speech.Program code can be executed fully on the user computer, partly be executed on the user computer, as an independence
Software package execute, part on the user computer part execute on the remote computer or completely in remote computer or
It is executed on server.In situations involving remote computers, remote computer can pass through the network of any kind --- packet
It includes local area network (LAN) or wide area network (WAN)-is connected to subscriber computer, or, it may be connected to outer computer (such as benefit
It is connected with ISP by internet).
Flow chart and block diagram in attached drawing are illustrated according to the system of the various embodiments of the disclosure, 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 of one module, program segment or code of table, a part of the module, program segment or code include one or more use
The executable instruction of the logic function as defined in realizing.It should also be noted that in some implementations as replacements, being marked in box
The function of note can also occur in a different order than that indicated in the drawings.For example, two boxes succeedingly indicated are actually
It can be basically executed in parallel, they can also be executed in the opposite order sometimes, and this depends on the function involved.Also it to infuse
Meaning, the combination of each box in block diagram and or flow chart and the box in block diagram and or flow chart can be with holding
The dedicated hardware based system of functions or operations as defined in row is realized, or can use specialized hardware and computer instruction
Combination realize.
Being described in unit involved in embodiment of the disclosure can be realized by way of software, can also be passed through
The mode of hardware is realized.Described unit also can be set in the processor, for example, can be described as: a kind of processor
Including first acquisition unit, determination unit and the first transmission unit.Wherein, the title of these units not structure under certain conditions
The restriction of the pairs of unit itself, for example, first acquisition unit is also described as " obtaining the classification letter that target terminal is sent
Cease the unit of set ".
As on the other hand, the disclosure additionally provides a kind of computer-readable medium, which can be
Included in server described in above-described embodiment;It is also possible to individualism, and without in the supplying server.It is above-mentioned
Computer-readable medium carries one or more program, when said one or multiple programs are executed by the server,
So that the server: obtaining the classification information set of target terminal transmission as external classification information set;In response to outer category
Target external classification information is not present in other information aggregate, model trained in advance is determined as the disaggregated model that training obtains,
Wherein, target external classification information is not match with the internal sort information in predetermined internal sort information aggregate
External classification information, trained model is obtained based on the training of predetermined training sample set in advance, in training sample set
Training sample include internal sort information and the corresponding internal data of internal sort information in internal sort information aggregate;To
Target terminal sends the calling interface for the disaggregated model that training obtains.
Above description is only the preferred embodiment of the disclosure and the explanation to institute's application technology principle.Those skilled in the art
Member is it should be appreciated that invention scope involved in the disclosure, however it is not limited to technology made of the specific combination of above-mentioned technical characteristic
Scheme, while should also cover in the case where not departing from foregoing invention design, it is carried out by above-mentioned technical characteristic or its equivalent feature
Any combination and the other technical solutions formed.Such as features described above has similar function with (but being not limited to) disclosed in the disclosure
Can technical characteristic replaced mutually and the technical solution that is formed.
Claims (14)
1. a kind of method for sending information, comprising:
The classification information set of target terminal transmission is obtained as external classification information set;
In response to target external classification information is not present in the external classification information set, model trained in advance is determined as
The obtained disaggregated model of training, wherein the target external classification information be not with predetermined internal sort information aggregate
In the external classification information that matches of internal sort information, the model trained in advance is based on predetermined training sample
Set training obtains, and the training sample in the training sample set includes in internal data and the internal sort information aggregate
Internal sort information corresponding with internal data;
The calling interface for the disaggregated model that training obtains is sent to the target terminal.
2. according to the method described in claim 1, wherein, the method also includes:
Obtain the test sample set that the target terminal is sent via the calling interface, wherein the test sample set
In test sample include data and data classification information;
Data in the test sample set are input to the disaggregated model of the corresponding calling interface, it is defeated to obtain disaggregated model
Classification information out;
Based on disaggregated model output classification information and the test sample set in classification information, generate disaggregated model with
At least one of lower assessment information: accuracy rate, recall rate, F1 score;
Assessment information generated is sent to the target terminal.
3. method according to claim 1 or 2, wherein send the classification that training obtains to the target terminal described
Before the calling interface of model, the method also includes:
In response to there are target external classification information, obtained in the external classification information set it is that the target terminal is sent,
The target external data acquisition system of the corresponding target external classification information;Using machine learning algorithm, based on the external classification
Information aggregate, the target external data acquisition system and target internal data acquisition system, are trained initial model, preparatory by meeting
The initial model of determining training termination condition is determined as the disaggregated model that training obtains;Wherein, the target internal data set
It is combined into internal data corresponding with the internal sort information that the external classification information in the external classification information set matches
Set.
4. according to the method described in claim 3, wherein, sending the disaggregated model that training obtains to the target terminal described
Calling interface before, the method also includes:
Inside in quantity and the target internal data acquisition system based on the external data in the target external data acquisition system
The quantity of data calculates remaining time, wherein the remaining time is used to indicate training and obtains time of disaggregated model and current
Time difference between time;
The remaining time is sent to the target terminal.
5. method described in one of -4 according to claim 1, wherein the method also includes:
Obtain the target data to be sorted that the target terminal is sent via the calling interface;
Target data to be sorted are input to the disaggregated model of the corresponding calling interface, generate target number to be sorted
According to classification information;
Classification information generated is sent to the target terminal.
6. method described in one of -4 according to claim 1, wherein the internal sort information aggregate and with the internal sort
The corresponding internal data set of internal sort information in information aggregate is obtained after Feature Engineering is handled.
7. a kind of for sending the device of information, comprising:
First acquisition unit is configured to obtain the classification information set of target terminal transmission as external classification information set;
Determination unit is configured in response to that target external classification information is not present in the external classification information set, will be pre-
First trained model is determined as the disaggregated model that training obtains, wherein the target external classification information be not with predefine
Internal sort information aggregate in the external classification information that matches of internal sort information, the model trained in advance is based on
The training of predetermined training sample set obtains, and the training sample in the training sample set includes internal data and described
Internal sort information corresponding with internal data in internal sort information aggregate;
First transmission unit is configured to send the calling interface for the disaggregated model that training obtains to the target terminal.
8. device according to claim 7, wherein described device further include:
Second acquisition unit is configured to obtain the test sample set that the target terminal is sent, wherein the test sample
Test sample in set includes the classification information of data and data;
Input unit is configured to for the data in the test sample set being input to the disaggregated model that training obtains, obtains
The classification information of disaggregated model output;
Generation unit, the classification in classification information and the test sample set for being configured to be exported based on disaggregated model are believed
Breath generates at least one following assessment information of disaggregated model: accuracy rate, recall rate, F1 score;
Second transmission unit is configured to assessment information generated being sent to the target terminal.
9. device according to claim 7 or 8, wherein described device further include:
Third acquiring unit is configured in response to obtain in the external classification information set there are target external classification information
Take the target terminal transmission, the correspondence target external classification information target external data acquisition system;Using machine learning
Algorithm, based on the external classification information set, the target external data acquisition system and target internal data acquisition system, to introductory die
Type is trained, and the initial model for meeting predetermined trained termination condition is determined as the disaggregated model that training obtains;Its
In, the target internal data acquisition system is the inner classes to match with the external classification information in the external classification information set
The corresponding internal data set of other information.
10. device according to claim 9, wherein described device further include:
Computing unit is configured to quantity and the target internal based on the external data in the target external data acquisition system
The quantity of internal data in data acquisition system calculates remaining time, wherein the remaining time is used to indicate training and is classified
Time difference between the time and current time of model;
Third transmission unit is configured to send the remaining time to the target terminal.
11. the device according to one of claim 7-10, wherein described device further include:
4th acquiring unit is configured to obtain the target number to be sorted that the target terminal is sent via the calling interface
According to;
Input unit is configured to target data to be sorted being input to the corresponding disaggregated model of the calling interface, raw
At the classification information of target data to be sorted;
4th transmission unit is configured to send classification information generated to the target terminal.
12. the device according to one of claim 7-10, wherein the internal sort information aggregate and with the inner classes
The corresponding internal data set of internal sort information in other information aggregate is obtained after Feature Engineering is handled.
13. a kind of server, comprising:
One or more processors;
Storage device is stored thereon with 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
Now such as method as claimed in any one of claims 1 to 6.
14. a kind of computer-readable medium, is stored thereon with computer program, wherein real when described program is executed by processor
Now such as method as claimed in any one of claims 1 to 6.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910575820.6A CN110288089B (en) | 2019-06-28 | 2019-06-28 | Method and apparatus for transmitting information |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910575820.6A CN110288089B (en) | 2019-06-28 | 2019-06-28 | Method and apparatus for transmitting information |
Publications (2)
Publication Number | Publication Date |
---|---|
CN110288089A true CN110288089A (en) | 2019-09-27 |
CN110288089B CN110288089B (en) | 2021-07-09 |
Family
ID=68019591
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201910575820.6A Active CN110288089B (en) | 2019-06-28 | 2019-06-28 | Method and apparatus for transmitting information |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN110288089B (en) |
Citations (8)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20160180214A1 (en) * | 2014-12-19 | 2016-06-23 | Google Inc. | Sharp discrepancy learning |
US20160210552A1 (en) * | 2013-08-26 | 2016-07-21 | Auckland University Of Technology | Improved Method And System For Predicting Outcomes Based On Spatio/Spectro-Temporal Data |
CN106446754A (en) * | 2015-08-11 | 2017-02-22 | 阿里巴巴集团控股有限公司 | Image identification method, metric learning method, image source identification method and devices |
CN107766940A (en) * | 2017-11-20 | 2018-03-06 | 北京百度网讯科技有限公司 | Method and apparatus for generation model |
CN108197664A (en) * | 2018-01-24 | 2018-06-22 | 北京墨丘科技有限公司 | Model acquisition methods, device, electronic equipment and computer readable storage medium |
CN108304936A (en) * | 2017-07-12 | 2018-07-20 | 腾讯科技(深圳)有限公司 | Machine learning model training method and device, facial expression image sorting technique and device |
US10115032B2 (en) * | 2015-11-04 | 2018-10-30 | Nec Corporation | Universal correspondence network |
CN108830235A (en) * | 2018-06-21 | 2018-11-16 | 北京字节跳动网络技术有限公司 | Method and apparatus for generating information |
-
2019
- 2019-06-28 CN CN201910575820.6A patent/CN110288089B/en active Active
Patent Citations (8)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20160210552A1 (en) * | 2013-08-26 | 2016-07-21 | Auckland University Of Technology | Improved Method And System For Predicting Outcomes Based On Spatio/Spectro-Temporal Data |
US20160180214A1 (en) * | 2014-12-19 | 2016-06-23 | Google Inc. | Sharp discrepancy learning |
CN106446754A (en) * | 2015-08-11 | 2017-02-22 | 阿里巴巴集团控股有限公司 | Image identification method, metric learning method, image source identification method and devices |
US10115032B2 (en) * | 2015-11-04 | 2018-10-30 | Nec Corporation | Universal correspondence network |
CN108304936A (en) * | 2017-07-12 | 2018-07-20 | 腾讯科技(深圳)有限公司 | Machine learning model training method and device, facial expression image sorting technique and device |
CN107766940A (en) * | 2017-11-20 | 2018-03-06 | 北京百度网讯科技有限公司 | Method and apparatus for generation model |
CN108197664A (en) * | 2018-01-24 | 2018-06-22 | 北京墨丘科技有限公司 | Model acquisition methods, device, electronic equipment and computer readable storage medium |
CN108830235A (en) * | 2018-06-21 | 2018-11-16 | 北京字节跳动网络技术有限公司 | Method and apparatus for generating information |
Non-Patent Citations (1)
Title |
---|
LI BINQUAN 等: "Effective distributed convolutional neural network architecture for remote sensing images target classification with a pre-training approach", 《JOURNAL OF SYSTEMS ENGINEERING AND ELECTRONICS》 * |
Also Published As
Publication number | Publication date |
---|---|
CN110288089B (en) | 2021-07-09 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN110458107A (en) | Method and apparatus for image recognition | |
CN109976997A (en) | Test method and device | |
CN109446990A (en) | Method and apparatus for generating information | |
CN108989882A (en) | Method and apparatus for exporting the snatch of music in video | |
CN109189544B (en) | Method and device for generating dial plate | |
CN109086719A (en) | Method and apparatus for output data | |
CN109101919A (en) | Method and apparatus for generating information | |
CN108960316A (en) | Method and apparatus for generating model | |
CN109410253B (en) | For generating method, apparatus, electronic equipment and the computer-readable medium of information | |
CN109993150A (en) | The method and apparatus at age for identification | |
CN108345387A (en) | Method and apparatus for output information | |
CN110084317A (en) | The method and apparatus of image for identification | |
CN109815365A (en) | Method and apparatus for handling video | |
CN109545193A (en) | Method and apparatus for generating model | |
CN109977839A (en) | Information processing method and device | |
CN109495552A (en) | Method and apparatus for updating clicking rate prediction model | |
CN112149699B (en) | Method and device for generating model and method and device for identifying image | |
CN109582825A (en) | Method and apparatus for generating information | |
CN108960110A (en) | Method and apparatus for generating information | |
CN109920431A (en) | Method and apparatus for output information | |
CN110209658A (en) | Data cleaning method and device | |
CN108062416B (en) | Method and apparatus for generating label on map | |
CN110211121A (en) | Method and apparatus for pushing model | |
CN108573054A (en) | Method and apparatus for pushed information | |
CN109325178A (en) | Method and apparatus for handling information |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PB01 | Publication | ||
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
GR01 | Patent grant | ||
GR01 | Patent grant |