A data training method, apparatus, and medium

Abstract

This invention relates to the field of data processing technology and discloses a data training method, apparatus, and medium. First, training data is acquired; then, a training task is run based on the training data, and training results are obtained. When the training results meet preset conditions, a modification file is acquired; the modification file includes a parameter file and a data file. Compared to current technologies, where parameter adjustments or data additions require retraining the model, wasting time and resources, this invention addresses this issue. When adjusting the training model, the modification file is uploaded to the corresponding directory based on the type of adjustment. An input identifier determines the starting position for training data through the modified file. This solution continues training from the position corresponding to the identifier, thus avoiding the waste of time and resources caused by starting training from scratch.

Description

Technical Field

[0001] This invention relates to the field of data processing technology, and in particular to a data training method, apparatus, and medium. Background Technology

[0002] An AI platform is a service platform for machine learning integrated development environments (IDEs) used for inference and training. It provides product design, optimization, and training support for other AI products. A good AI product requires the support of an excellent algorithm. The creation of an excellent algorithm often requires extensive computation and countless parameter adjustments and data additions to refine it.

[0003] In current technology, during the initial training of an algorithm model, a small amount of data is often used to quickly obtain the accuracy of the trained model. Each time, the model parameters are modified or the amount of training data is increased based on the training results. However, as the trained model stabilizes, the entire training cycle becomes very long. When adjusting parameters or adding more data later, it is necessary to retrain on the previously trained data, resulting in the consumption of more time and resources.

[0004] Therefore, how to avoid repeated training on historical training data and avoid wasting time and resources is a problem that urgently needs to be solved by those skilled in the art. Summary of the Invention

[0005] The purpose of this invention is to provide a data training method, apparatus, and medium to avoid repeated training on historical training data, thereby avoiding wasting time and resources.

[0006] To address the aforementioned technical problems, this invention provides a data training method, comprising:

[0007] Obtain training data;

[0008] Run the training task based on the training data and obtain the training results;

[0009] When the training result is detected to meet the preset conditions, a modified file is obtained; the modified file includes a parameter file and a data file.

[0010] Upload the modified file to the corresponding directory location according to the type of the modified file;

[0011] Based on the input identifier, confirm the starting position of the training data through the modified file for data training.

[0012] On the one hand, the modified file is a parameter file;

[0013] Correspondingly, the training results satisfy the following preset conditions:

[0014] The training results do not meet the expected output;

[0015] Correspondingly, uploading the modified file to the corresponding directory location according to the type of the modified file includes:

[0016] Upload the parameter file to the directory location that will be mounted after the training task is run.

[0017] On the other hand, the marker is: the marker used to mark the data positions in the training results that meet the expected results;

[0018] Furthermore, the input identifier is the identifier that is closest to the current time among all the identifiers.

[0019] On the other hand, the modified file is a data file;

[0020] Correspondingly, the training results satisfy the following preset conditions:

[0021] The training results of the training data meet the expected output;

[0022] Correspondingly, uploading the modified file to the corresponding directory location according to the type of the modified file includes:

[0023] Upload the data file to the same directory location as the training data.

[0024] On the other hand, the identifier is: the end identifier of the training data;

[0025] Furthermore, it also includes: using the end marker to mark the start position of the data file.

[0026] On the other hand, there is overlap between the training data and the data in the data file.

[0027] On the other hand, it also includes:

[0028] Command to obtain task running status;

[0029] If the task running status instruction is an end instruction, then training will stop after the training task is completed;

[0030] If the task running status instruction is a continue instruction, then the training of other training tasks will continue after the training task is completed.

[0031] To address the aforementioned technical problems, the present invention also provides a data training apparatus, comprising:

[0032] The acquisition module is used to acquire training data;

[0033] The execution module is used to run training tasks based on the training data and obtain training results;

[0034] The detection module is used to obtain a modified file when the training result is detected to meet preset conditions; the modified file includes a parameter file and a data file.

[0035] The upload module is used to upload the modified file to the corresponding directory location according to the type of the modified file;

[0036] The confirmation module is used to confirm the starting position of the training data through the modified file for data training based on the input identifier.

[0037] On the other hand, the detection module is used to obtain a parameter file when the training result does not meet the expected output;

[0038] Correspondingly, the upload module is used to upload the parameter file to the directory location that will be mounted after the training task is run.

[0039] On the other hand, the identifiers in the confirmation module are: identifiers used to mark the data positions in the training results that meet the expected results;

[0040] Furthermore, the input identifier is the identifier that is closest to the current time among all the identifiers.

[0041] On the other hand, the detection module is used to acquire a data file when the training result of the training data meets the expected output;

[0042] Correspondingly, the upload module is used to upload the data file to the same directory location as the training data.

[0043] On the other hand, the identifier in the confirmation module is: the end identifier of the training data;

[0044] Furthermore, it also includes a marking module for marking the start position of the data file using the end marker.

[0045] On the other hand, there is overlap between the data in the data file obtained by the detection module and the training data.

[0046] On the other hand, it also includes:

[0047] The processing module is used to obtain task running status instructions; if the task running status instructions are end instructions, training is stopped after the training task is completed; if the task running status instructions are continue instructions, training of other training tasks continues after the training task is completed.

[0048] To address the aforementioned technical problems, the present invention also provides a data training apparatus, including a memory for storing computer programs;

[0049] A processor is used to implement the steps of the data training method described above when executing the computer program.

[0050] To address the aforementioned technical problems, the present invention also provides a computer-readable storage medium storing a computer program, which, when executed by a processor, implements the steps of the data training method described above.

[0051] The data training method provided by this invention first acquires training data, runs a training task based on the training data, and obtains training results. When the training results meet preset conditions, a modification file is acquired; the modification file includes a parameter file and a data file. The modification file is uploaded to the corresponding directory location according to its type. Based on an input identifier, the starting position for training data through the modification file is determined to begin data training. Compared to current technologies, where parameter adjustments or data additions require retraining the model, wasting time and resources, this method, when adjusting the training model, uploads the modification file to the corresponding directory location according to the type of adjustment, and determines the starting position for training data through the modification file based on an input identifier. This method continues training from the position corresponding to the identifier, thus avoiding the waste of time and resources caused by training from scratch.

[0052] Furthermore, the data training apparatus and medium provided by this invention correspond to the data training method described above and have the same effect. Attached Figure Description

[0053] To more clearly illustrate the embodiments of the present invention, the accompanying drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0054] Figure 1 A flowchart of a data training method provided in an embodiment of the present invention;

[0055] Figure 2 An application flowchart for parameter file adjustment provided in an embodiment of the present invention;

[0056] Figure 3 An application flowchart for adjusting data files is provided in an embodiment of the present invention;

[0057] Figure 4A structural diagram of a data training device provided in an embodiment of the present invention;

[0058] Figure 5 This is a structural diagram of another data training device provided in an embodiment of the present invention. Detailed Implementation

[0059] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the protection scope of the present invention.

[0060] The core of this invention is to provide a data training method, apparatus, and medium to avoid repeated training on historical training data, thereby avoiding wasting time and resources.

[0061] To enable those skilled in the art to better understand the present invention, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0062] First, it should be noted that the data training method provided by this invention is mainly applied to artificial intelligence platforms. These platforms facilitate the establishment of algorithm models, and the data training method of this invention provides a more accurate and convenient data foundation for these models. The execution entity of the data training method provided by this invention can be a data training device, such as a processor. Technical personnel can perform data training through human-computer interaction with the data training device.

[0063] Figure 1 A flowchart of a data training method provided in an embodiment of the present invention is shown below. Figure 1 As shown, the method includes:

[0064] S10: Obtain training data;

[0065] S11: Run the training task based on the training data and obtain the training results;

[0066] S12: When the training results are detected to meet the preset conditions, obtain the modified files; the modified files include parameter files and data files;

[0067] S13: Upload the modified files to the corresponding directory location according to the type of modified files;

[0068] S14: Based on the input identifier, confirm the starting position of the training data in the modified file for data training.

[0069] In practical implementation, when training on training data, to more quickly confirm the accuracy of the training model, a small portion of the data can be used for training first, and the amount of data can be increased only when the training results meet expectations. Therefore, specifically, in this embodiment, after obtaining the training data, the process further includes: dividing the training data into multiple parts, and then submitting and running training tasks sequentially; this allows for timely adjustment of parameter information when the training result of a certain training task does not meet expectations, ensuring the accuracy of the algorithm model. When running training tasks, the training tasks need to be submitted sequentially according to the order of the training data to ensure the consistency and accuracy of the training model.

[0070] After a training task is run, the log file records the training results of each task. This log file can record whether a training task was successful or failed, or the name or flag of a successfully trained task. The determination of whether training was successful or failed can be based on whether the training output matches the expected output.

[0071] This embodiment provides a method for logging training results. The location of data training is marked by case_id. During the training task, the location of data training can be confirmed by reading case_id.

[0072] When the training results are detected to meet preset conditions, modification files are obtained. These files include parameter files and data files. The parameter files primarily contain the parameters used in training the model to ensure the algorithm's accuracy. The data files mainly contain newly added training data to expand the training data volume and improve the algorithm's accuracy. The detection of whether the training results meet the preset conditions can be performed by technicians inputting relevant instructions or by the data training device detecting certain conditions. Specifically, since the obtained modification files mainly address parameter adjustments and training data volume adjustments, the preset conditions for training results also address situations requiring parameter adjustments or increased data volume. For example, if the training results deviate, parameter adjustments are needed; or if the training data has been completed, new data needs to be added for further training.

[0073] Understandably, different modified files are needed for different situations, and their storage locations vary depending on the type of file. Therefore, the modified files need to be uploaded to the corresponding directory locations. For example, the location where parameters are stored and the location where data is stored.

[0074] After the modified file is uploaded to the corresponding location, the technician can input a marker, and the training task can then begin from the position corresponding to the marker, continuing data training by modifying the file. It should be noted that this marker can be the symbol used to mark the data training position as described above, or it can be other symbols (such as those used to mark the end or beginning of data). Through this marker, the technician and the data training device can confirm the starting position of the training task. In this invention, this marker avoids the training task from repeatedly performing data training from the beginning, thus avoiding waste of time and resources.

[0075] The data training method provided in this invention first acquires training data, runs a training task based on the training data, and obtains training results. When the training results meet preset conditions, a modification file is acquired. The modification file includes a parameter file and a data file. The modification file is uploaded to the corresponding directory location according to its type. Based on an input identifier, the starting position for training data through the modification file is determined to begin data training. Compared to current technologies, where parameter adjustments or data additions require retraining the model, wasting time and resources, this method, when adjusting the training model, uploads the modification file to the corresponding directory location according to the type of adjustment, and determines the starting position for training data through the modification file based on an input identifier. This method continues training from the position corresponding to the identifier, thus avoiding the waste of time and resources caused by training from scratch.

[0076] Based on the above embodiments, this embodiment provides a specific method for model adjustment. In this embodiment, the modified file is a parameter file;

[0077] Correspondingly, the training results must meet the following preset conditions:

[0078] The training results do not match the expected output;

[0079] Correspondingly, depending on the type of file being modified, the modified files will be uploaded to the corresponding directory location, including:

[0080] Upload the parameter file to the directory that will be mounted after the training task runs.

[0081] Understandably, in this invention, since technicians can obtain training results in real time through logs, parameter adjustments are necessary when the training results do not meet expectations. Therefore, a parameter file is required, which is used to modify the parameters in the current model based on the training results.

[0082] Figure 2 An application flowchart for parameter file adjustment provided in an embodiment of the present invention is shown below. Figure 2As shown, it includes:

[0083] S20: Submit the training task;

[0084] S21: Should the training task be run? If yes, proceed to step S220; otherwise, proceed to step S221.

[0085] S220: Output the running result; S221: Wait for execution;

[0086] S23: Should the parameters be updated? If yes, proceed to step S24; otherwise, proceed to step S25.

[0087] S24: Upload the parameter file to the corresponding directory location and enter the identifier;

[0088] S25: Task execution.

[0089] Understandably, in practice, after acquiring training data and submitting the training task, technicians can choose whether to start training and run the training task. This can be actively controlled by the technicians, such as by inputting relevant instructions, clicking the corresponding button to run the training task, or starting the training task when other judgment conditions are met (e.g., reaching a set time); otherwise, the process waits. After running the training task, the results are obtained. Technicians can then determine whether to update the parameters based on the results, specifically by checking whether the training results meet the expected output. When parameter updates are needed, the technicians upload the parameter file to the corresponding directory, enter a flag, and click "Run Task." The training model can then continue training from the location indicated by the flag.

[0090] The directory location refers to the directory where the parameter file is uploaded and mounted after the training task runs. The identifier is used to mark the location of data in the training results that matches the expected performance; further, the input identifier is the identifier closest to the current time step among all the identifiers.

[0091] As can be understood from the above embodiments, in this embodiment, after parameter changes, data training continues from the position corresponding to the input marker. The marker in this embodiment indicates the position where the training result matches the expected output. Each time a training result matching the expected output is output, a marker is added to that position. After updating the parameters, the user can input the most recent marker, and the training task can then start from that position. This allows for retraining on failed training data while avoiding repeated training on successfully trained data. Of course, in other embodiments, training can also start from the previous few markers at the current time, as needed by the technician.

[0092] Based on the above embodiments, this embodiment provides another specific method for model adjustment. In this embodiment, the modified file is a data file;

[0093] Correspondingly, the training results must meet the following preset conditions:

[0094] The training results on the training data meet the expected output.

[0095] Correspondingly, depending on the type of file being modified, the modified files will be uploaded to the corresponding directory location, including:

[0096] Upload the data files to the same directory location as the training data.

[0097] When adding data, the corresponding identifier is: the end identifier of the training data;

[0098] Furthermore, this also includes using an end marker to mark the beginning of the data file.

[0099] The above embodiments provide a data training method after parameter adjustment; this embodiment provides a training method after adding data. As can be understood from the description of the above embodiments, to more quickly confirm the training model, the training data can be divided into multiple parts and trained on multiple training tasks. Therefore, in this embodiment, after one training task is completed, other training tasks need to be trained as well. Or, after the entire training data has been trained, new data is acquired for training. In this case, the acquired modified file is the data file. The prerequisite for continuing data training is that the training of historical training data meets the expected output. Therefore, in this embodiment, the data file is acquired only after the training results of the training data meet the expected output, so that training can continue on the data in the data file after the training of the training data has been completed. The data file is uploaded to the same directory as the training data, so that training can continue on the data in the data file after the training of the training data has been completed.

[0100] Figure 3 This is an application flowchart for adjusting a data file provided in an embodiment of the present invention. Figure 2 Similarities include:

[0101] S30: Submit the training task;

[0102] S31: Should the training task be run? If yes, proceed to step S320; otherwise, proceed to step S321.

[0103] S320: Output the running result; S321: Wait for execution;

[0104] S33: Should the parameters be updated? If yes, proceed to step S34; otherwise, proceed to step S35.

[0105] S34: Upload the data file to the corresponding directory location, and enter the identifier, the name of the training data, and the name of the data file;

[0106] S35: Task execution.

[0107] In this embodiment, the application flowchart for data file adjustment differs from that for parameter file adjustment in that parameter adjustment starts from the marker position and continues training on the training data. However, when adjusting data, after acquiring newly added data, it is necessary to confirm the name of the training data and the name of the data file before starting training via the marker. This marker connects the training data and the data file. When the training data marker is detected, training switches to the position corresponding to the marker in the data file. This marker can still indicate a successful training point, but in practice, to ensure the accuracy of the training results, there may be overlapping data between the training data and the data file. Therefore, the marker can be an end-of-training marker, placed before the last data point. For example, if the training data is audio data with a length of 10 seconds, the marker would be placed at 9 seconds. The portion from 9 to 10 seconds is the overlapping data with the data file.

[0108] Based on the above embodiments, this embodiment further includes:

[0109] Command to obtain task running status;

[0110] If the task running status instruction is an end instruction, then training will stop after the training task is completed;

[0111] If the task running status instruction is a continue instruction, then training for other training tasks will continue after the training task is completed.

[0112] In this embodiment, before running the training task, users can choose to repeat the training after completion or end the task. Specifically, this can be determined based on the task running status command. When the task running status command is a continue command, users can choose to input a flag to run from a specified position or start from the beginning. When the task running status command is an end command, training stops after completion, and resources are released.

[0113] In summary, for ease of understanding, this embodiment provides a specific data training scenario. In this embodiment, the data to be trained is audio or video, which can be divided into several segments. First, the task is submitted, and the audio data for training is initially used, with a 1-minute audio segment. The task running status command selects the continue command.

[0114] After the task runs, based on the training results, when a deviation in audio recognition is found, the case_id that most recently matches the expectation is found according to the case_id printed in the log, and the training parameters are adjusted. The adjusted file is then uploaded to the specified directory (usually placed in the mounted directory after the task runs).

[0115] Select the task operation bar, adjust the parameters, enter case_id, click OK, and wait for the task to run from case_id.

[0116] When the training results of the 1-minute audio duration meet expectations, subsequent audio is needed for further training. The subsequent audio should be uploaded to the designated directory (usually in the same directory as the 1-minute audio data).

[0117] In the task operation bar, click Update Data, enter the data_tag automatically generated by the system, the name of the old training data, and the name of the new training data, and click OK.

[0118] To complete the task training, select to change the task completion status from the continue instruction to the end instruction.

[0119] If the training results do not meet the expected output, the above steps can be repeated based on the task log output.

[0120] In the above embodiments, the data training method has been described in detail. This invention also provides embodiments corresponding to a data training device. It should be noted that this invention describes the device embodiments from two perspectives: one based on functional modules, and the other based on hardware.

[0121] Figure 4 A structural diagram of a data training device provided in an embodiment of the present invention is shown below. Figure 4 As shown, the device includes:

[0122] Module 10 is used to acquire training data;

[0123] The execution module 11 is used to run training tasks based on training data and obtain training results;

[0124] The detection module 12 is used to obtain the modified file when the training result is detected to meet the preset conditions; the modified file includes a parameter file and a data file.

[0125] Upload module 13 is used to upload modified files to the corresponding directory location according to the type of modified file;

[0126] The confirmation module 14 is used to confirm, based on the input identifier, the starting position of the training data in the modified file for data training.

[0127] In other embodiments, the detection module is used to obtain a parameter file when the training result does not meet the expected output;

[0128] Correspondingly, the upload module is used to upload the parameter file to the directory location that will be mounted after the training task is run.

[0129] In other embodiments, the identifier in the confirmation module is: an identifier used to mark the data positions in the training results that meet the expected results;

[0130] In other embodiments, the input identifier is the identifier that is closest to the current time among all the identifiers.

[0131] In other embodiments, the detection module is used to acquire a data file when the training result of the training data meets the expected output;

[0132] Correspondingly, the upload module is used to upload the data file to the same directory location as the training data.

[0133] In other embodiments, the identifier in the confirmation module is: the end identifier of the training data;

[0134] Furthermore, it also includes a marking module for marking the start position of the data file using the end marker.

[0135] In other embodiments, the data in the data file obtained by the detection module overlaps with the training data.

[0136] In other embodiments, it also includes:

[0137] The processing module is used to obtain task running status instructions; if the task running status instruction is an end instruction, then training stops after the training task is completed; if the task running status instruction is a continue instruction, then training of other training tasks continues after the training task is completed.

[0138] Since the embodiments of the apparatus and the embodiments of the method correspond to each other, please refer to the description of the embodiments of the method for the embodiments of the apparatus, which will not be repeated here.

[0139] The data training apparatus provided in this embodiment of the invention first acquires training data, runs a training task based on the training data, and obtains training results. When the training results meet preset conditions, a modification file is acquired. The modification file includes a parameter file and a data file. The modification file is uploaded to the corresponding directory location according to its type. Based on an input identifier, the starting position for training data through the modification file is determined to begin data training. Compared to current technologies, where parameter adjustments or data additions require retraining the model, wasting time and resources, this technical solution, when adjusting the training model, uploads the modification file to the corresponding directory location according to the type of adjustment, and determines the starting position for training data through the modification file based on an input identifier. This solution continues training from the position corresponding to the identifier, thus avoiding the waste of time and resources caused by training from scratch.

[0140] Figure 5 A structural diagram of another data training device provided in an embodiment of the present invention is shown below. Figure 5 As shown, the device includes: a memory 20 for storing computer programs;

[0141] The processor 21 is used to execute a computer program to implement the steps of the data training method as described in the above embodiments.

[0142] The data training device provided in this embodiment may include, but is not limited to, smartphones, tablets, laptops, or desktop computers.

[0143] The processor 21 may include one or more processing cores, such as a quad-core processor or an octa-core processor. The processor 21 may be implemented using at least one of the following hardware forms: Digital Signal Processor (DSP), Field-Programmable Gate Array (FPGA), or Programmable Logic Array (PLA). The processor 21 may also include a main processor and a coprocessor. The main processor, also known as the Central Processing Unit (CPU), is used to process data in the wake-up state; the coprocessor is a low-power processor used to process data in the standby state. In some embodiments, the processor 21 may integrate a Graphics Processing Unit (GPU), which is responsible for rendering and drawing the content to be displayed on the screen. In some embodiments, the processor 21 may also include an Artificial Intelligence (AI) processor, which is used to handle computational operations related to machine learning.

[0144] The memory 20 may include one or more computer-readable storage media, which may be non-transitory. The memory 20 may also include high-speed random access memory and non-volatile memory, such as one or more disk storage devices or flash memory devices. In this embodiment, the memory 20 is used to store at least the following computer program 201, which, after being loaded and executed by the processor 21, is capable of implementing the relevant steps of the data training method disclosed in any of the foregoing embodiments. In addition, the resources stored in the memory 20 may also include an operating system 202 and data 203, and the storage method may be temporary storage or permanent storage. The operating system 202 may include Windows, Unix, Linux, etc. The data 203 may include, but is not limited to, identifiers.

[0145] In some embodiments, the data training device may further include a display screen 22, an input / output interface 23, a communication interface 24, a power supply 25, and a communication bus 26.

[0146] Those skilled in the art will understand that Figure 5 The structure shown does not constitute a limitation on the data training device and may include more or fewer components than illustrated.

[0147] The data training apparatus provided in this embodiment of the invention includes a memory and a processor. When the processor executes a program stored in the memory, it can perform the following methods: acquiring training data; running a training task based on the training data and acquiring training results; acquiring a modified file when the training results meet preset conditions; the modified file includes a parameter file and a data file; uploading the modified file to the corresponding directory location according to the type of modified file; and confirming the starting position of the training data through the modified file for data training based on the input identifier.

[0148] The data training apparatus provided in this embodiment of the invention first acquires training data, runs a training task based on the training data, and obtains training results. When the training results meet preset conditions, a modification file is acquired. The modification file includes a parameter file and a data file. The modification file is uploaded to the corresponding directory location according to its type. Based on an input identifier, the starting position for training data through the modification file is determined to begin data training. Compared to current technologies, where parameter adjustments or data additions require retraining the model, wasting time and resources, this technical solution, when adjusting the training model, uploads the modification file to the corresponding directory location according to the type of adjustment, and determines the starting position for training data through the modification file based on an input identifier. This solution continues training from the position corresponding to the identifier, thus avoiding the waste of time and resources caused by training from scratch.

[0149] Finally, the present invention also provides an embodiment corresponding to a computer-readable storage medium. The computer-readable storage medium stores a computer program, which, when executed by a processor, performs the steps described in the above method embodiments.

[0150] It is understood that if the methods in the above embodiments are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and executes all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0151] The computer-readable storage medium provided in this invention first acquires training data, runs a training task based on the training data, and obtains training results. When the training results meet preset conditions, a modification file is acquired. The modification file includes a parameter file and a data file. The modification file is uploaded to the corresponding directory location according to its type. Based on an input identifier, the starting position of the training data through the modification file is determined for data training. Compared to current technologies, where parameter adjustments or data additions require retraining the model, wasting time and resources, this technical solution, when adjusting the training model, uploads the modification file to the corresponding directory location according to the type of adjustment, and determines the starting position of the training data through the modification file based on an input identifier. This solution continues training from the position corresponding to the identifier, thus avoiding the waste of time and resources caused by training from scratch.

[0152] The data training method, apparatus, and medium provided by this invention have been described in detail above. The various embodiments in the specification are described in a progressive manner, with each embodiment focusing on its differences from other embodiments. Similar or identical parts between embodiments can be referred to interchangeably. For the apparatus disclosed in the embodiments, since it corresponds to the method disclosed in the embodiments, the description is relatively simple; relevant parts can be referred to in the method section. It should be noted that those skilled in the art can make several improvements and modifications to this invention without departing from the principles of this invention, and these improvements and modifications also fall within the protection scope of the claims of this invention.

[0153] It should also be noted that, in this specification, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

Claims

1. A data training method, characterized in that, The method comprises the following steps: acquiring training data; running a training task according to the training data and acquiring a training result; when it is detected that the training result meets a preset condition, acquiring a modification file; the modification file comprises a parameter file and a data file; the parameter file is a parameter in a training model, and the data file is newly added training data; uploading the modification file to a corresponding directory location according to a type of the modification file; confirming a starting position of the training data through the modification file according to an inputted marker to perform data training; the modification file is a parameter file; correspondingly, the training result meeting the preset condition comprises: the training result not meeting an expected output; correspondingly, the uploading of the modification file to the corresponding directory location according to the type of the modification file comprises: uploading the parameter file to a directory location mounted after the training task is run; the marker is a marker used to mark a data position with an output meeting an expected effect in the training result; further, the inputted marker is a marker closest to a current time among the markers; alternatively, the modification file is a data file; correspondingly, the training result meeting the preset condition comprises: the training result of the training data meeting an expected output; correspondingly, the uploading of the modification file to the corresponding directory location according to the type of the modification file comprises: uploading the data file to a same directory location as the training data; the marker is an ending marker of the training data; further, the method further comprises the following step:

2. The data training method of claim 1, wherein, marking a starting position of the data file using the ending marker.

3. The data training method of claim 1, wherein, The training data and data in the data file have overlapping data. The method further comprises the following steps: acquiring a task running state instruction; if the task running state instruction is an ending instruction, stopping training after the training task is completed; 4. A data training apparatus, characterized by, if the task running state instruction is a continuing instruction, continuing training of other training tasks after the training task is completed. The method comprises the following steps: an acquiring module, configured to acquire training data; a running module, configured to run a training task according to the training data and acquire a training result; a detecting module, configured to acquire a modification file when it is detected that the training result meets a preset condition; the modification file comprises a parameter file and a data file; the parameter file is a parameter in a training model, and the data file is newly added training data; an uploading module, configured to upload the modification file to a corresponding directory location according to a type of the modification file; a confirming module, configured to confirm a starting position of the training data through the modification file according to an inputted marker to perform data training; the detecting module is configured to acquire a parameter file when the training result does not meet an expected output; correspondingly, the uploading module is configured to upload the parameter file to a directory location mounted after the training task is run; the marker in the confirming module is a marker used to mark a data position with an output meeting an expected effect in the training result; the inputted marker is a marker closest to a current time among the markers; Alternatively, the detection module is configured to acquire a data file when a training result of the training data matches an expected output. Correspondingly, the uploading module is configured to upload the data file to a same directory location as the training data.

5. A data training apparatus, characterized by, The computer readable storage medium stores a computer program. The processor is configured to implement the steps of the data training method according to any one of claims 1 to 3 when executing the computer program.

6. A computer readable storage medium characterized by The computer readable storage medium stores a computer program, and the computer program is configured to implement the steps of the data training method according to any one of claims 1 to 3 when executed by the processor.

Images