Creating a Dataset Import Phase

Description

This phase integrates capabilities of the ModelArts dataset module, allowing you to import data to datasets. The dataset import phase is used to import data from a specified path to a dataset or a labeling job. The application scenarios are as follows:

This phase is used for continuous data update. You can import raw data or labeled data to a labeling job and label the data in the labeling phase.
Some labeled raw data can be directly imported to a dataset or labeling job, and the dataset with version information can be obtained in the dataset release phase.

Parameter Overview

You can use DatasetImportStep to create a dataset import phase. The following is an example of defining a DatasetImportStep.

**Table 1** **DatasetImportStep**
Parameter	Description	Mandatory	Data Type
name	Name of a dataset import phase. The name contains a maximum of 64 characters, including only letters, digits, underscores (_), and hyphens (-). It must start with a letter and must be unique in a workflow.	Yes	str
inputs	Inputs of the dataset import phase.	Yes	DatasetImportInput or DatasetImportInput list
outputs	Outputs of the dataset import phase.	Yes	DatasetImportOutput or DatasetImportOutput list
properties	Configurations for dataset import.	Yes	ImportDataInfo
title	Title for frontend display.	No	str
description	Description of the dataset import phase.	No	str
policy	Phase execution policy.	No	StepPolicy
depend_steps	Dependent phases.	No	Step or step list

**Table 2** **DatasetImportInput**
Parameter	Description	Mandatory	Data Type
name	Input name of the dataset import phase. The name can contain a maximum of 64 characters, including only letters, digits, underscores (_), and hyphens (-), and must start with a letter. The input name of a step must be unique.	Yes	str
data	Input data object of the dataset import phase.	Yes	Dataset, OBS, or labeling job object. Currently, only Dataset, DatasetConsumption, DatasetPlaceholder, OBSPath, OBSConsumption, OBSPlaceholder, LabelTask, LabelTaskPlaceholder, LabelTaskConsumption, and DataConsumptionSelector are supported.

**Table 3** **DatasetImportOutput**
Parameter	Description	Mandatory	Data Type
name	Output name of the dataset import phase. The name can contain a maximum of 64 characters, including only letters, digits, underscores (_), and hyphens (-), and must start with a letter. The output name of a step must be unique.	Yes	str

**Table 4** **ImportDataInfo**
Parameter	Description	Mandatory	Data Type
annotation_format_config	Configurations of the imported labeling format.	No	AnnotationFormatConfig
excluded_labels	Samples with specified labels are not imported.	No	Label list
import_annotated	Whether to import the labeled samples in the original dataset to the To Be Confirmed tab. The default value is false, indicating that the labeled samples in the original dataset are not imported to the To Be Confirmed tab. The options are as follows: true: The labeled samples in the original dataset are imported to the To Be Confirmed tab. false: The labeled samples in the original dataset are not imported to the To Be Confirmed tab.	No	bool
import_annotations	Whether to import labels. The options are as follows: true: The labels are imported. (Default) false: The labels are not imported.	No	bool
import_samples	Whether to import samples. The options are as follows: true: The samples are imported. (Default) false: The samples are not imported.	No	bool
import_type	Import mode. The options are as follows: dir: imported from an OBS path manifest: imported from a manifest file	No	ImportTypeEnum
included_labels	Samples with specified labels are imported.	No	Label list
label_format	Label format. This parameter is used only for text datasets.	No	LabelFormat

**Table 5** **AnnotationFormatConfig**
Parameter	Description	Mandatory	Data Type
format_name	Name of a labeling format	No	AnnotationFormatEnum
parameters	Advanced parameters of the labeling format	No	AnnotationFormatParameters
scene	Labeling scenario, which is optional	No	LabelTaskTypeEnum

**Table 6** **AnnotationFormatParameters**
Parameter	Description	Mandatory	Data Type
difficult_only	Whether to import only hard examples. The options are as follows: true: Only hard examples are imported. false: All the samples are imported. (Default)	No	bool
included_labels	Samples with specified labels are imported.	No	Label list
label_separator	Separator between labels. By default, the comma (,) is used as the separator. The separator needs to be escaped. The separator can contain only one character, which must be a letter, a digit, or any of the following special characters: !@#$%^&*_=\|?/':.;,	No	str
sample_label_separator	Separator between the text and label. By default, the Tab key is used as the separator. The separator needs to be escaped. The separator can contain only one character, which must be a letter, a digit, or any of the following special characters: !@#$%^&*_=\|?/':.;,	No	str

Examples

There are three scenarios:

Scenario 1: Updating a dataset by importing data from a specified path

You import labeled data (with label information) in a specified path to a dataset. Then, you can create a dataset release phase to release a version.

Data preparation: Create a dataset on the ModelArts console and upload labeled data to OBS.

from modelarts import workflow as wf
# Use DatasetImportStep to import data in a specified path to a dataset and output the dataset.

# Define a dataset.
dataset = wf.data.DatasetPlaceholder(name="input_dataset")

# Define OBS data.
obs = wf.data.OBSPlaceholder(name = "obs_placeholder_name", object_type = "directory" ) # object_type must be file or directory.

dataset_import = wf.steps.DatasetImportStep(
    name="data_import", # Name of the dataset import phase. The name contains a maximum of 64 characters, including only letters, digits, underscores (_), and hyphens (-). It must start with a letter and must be unique in a workflow.
    title="Dataset Import", # Title, which defaults to the value of name
    inputs=[
        wf.steps.DatasetImportInput(name="input_name_1", data=dataset), # The target dataset is configured when the workflow is running. You can also use wf.data.Dataset(dataset_name="dataset_name") for the data field.
        wf.steps.DatasetImportInput(name="input_name_2", data=obs) # Storage path to the imported dataset, configured when the workflow is running. You can also use wf.data.OBSPath(obs_path="obs_path") for the data field.
    ],# DatasetImportStep inputs
    outputs=wf.steps.DatasetImportOutput(name="output_name"), # DatasetImportStep outputs
    properties=wf.steps.ImportDataInfo(
        annotation_format_config=[
            wf.steps.AnnotationFormatConfig(
                format_name=wf.steps.AnnotationFormatEnum.MA_IMAGE_CLASSIFICATION_V1, # Labeling format of labeled data, for example, image classification
                scene=wf.data.LabelTaskTypeEnum.IMAGE_CLASSIFICATION # Labeling scene
            )
        ]
    )
)

workflow = wf.Workflow(
    name="dataset-import-demo",
    desc="this is a demo workflow",
    steps=[dataset_import]
)

You import unlabeled data in a specified path to a dataset. Then, you can add a labeling phase to label the imported data.

Data preparation: Create a dataset on the ModelArts console and upload unlabeled data to OBS.

from modelarts import workflow as wf
# Use DatasetImportStep to import data in a specified path to a dataset and output the dataset.

# Define a dataset.
dataset = wf.data.DatasetPlaceholder(name="input_dataset")

# Define OBS data.
obs = wf.data.OBSPlaceholder(name = "obs_placeholder_name", object_type = "directory" ) # object_type must be file or directory.

dataset_import = wf.steps.DatasetImportStep(
    name="data_import", # Name of the dataset import phase. The name contains a maximum of 64 characters, including only letters, digits, underscores (_), and hyphens (-). It must start with a letter and must be unique in a workflow.
    title="Dataset Import", # Title, which defaults to the value of name
    inputs=[
        wf.steps.DatasetImportInput(name="input_name_1", data=dataset), # The target dataset is configured when the workflow is running. You can also use wf.data.Dataset(dataset_name="dataset_name") for the data field.
        wf.steps.DatasetImportInput(name="input_name_2", data=obs) # Storage path to the imported dataset, configured when the workflow is running. You can also use wf.data.OBSPath(obs_path="obs_path") for the data field.
    ],# DatasetImportStep inputs
    outputs=wf.steps.DatasetImportOutput(name="output_name"), # DatasetImportStep outputs
)

workflow = wf.Workflow(
    name="dataset-import-demo",
    desc="this is a demo workflow",
    steps=[dataset_import]
)

Scenario 2: Updating a labeling job by importing data from a specified path

You import labeled data in a specified path to a labeling job. Then, you can create a dataset release phase to release a version.

Data preparation: Create a labeling job using a specified dataset and upload the labeled data to OBS.

from modelarts import workflow as wf
# Use DatasetImportStep to import data in a specified path to a labeling job and output the labeling job.

# Define a labeling job.
label_task = wf.data.LabelTaskPlaceholder(name="label_task_placeholder_name")

# Define the OBS data.
obs = wf.data.OBSPlaceholder(name = "obs_placeholder_name", object_type = "directory" ) # object_type must be file or directory.

dataset_import = wf.steps.DatasetImportStep(
    name="data_import", # Name of the dataset import phase. The name contains a maximum of 64 characters, including only letters, digits, underscores (_), and hyphens (-). It must start with a letter and must be unique in a workflow.
    title="Dataset Import", # Title, which defaults to the value of name
    inputs=[
        wf.steps.DatasetImportInput(name="input_name_1", data=label_task), # Labeling job object, configured when the workflow is running. You can also use wf.data.LabelTask(dataset_name="dataset_name", task_name="label_task_name") for the data field.
        wf.steps.DatasetImportInput(name="input_name_2", data=obs) # Storage path to the imported dataset, configured when the workflow is running. You can also use wf.data.OBSPath(obs_path="obs_path") for the data field.
    ],# DatasetImportStep inputs
    outputs=wf.steps.DatasetImportOutput(name="output_name"), # DatasetImportStep outputs
    properties=wf.steps.ImportDataInfo(
        annotation_format_config=[
            wf.steps.AnnotationFormatConfig(
                format_name=wf.steps.AnnotationFormatEnum.MA_IMAGE_CLASSIFICATION_V1, # Labeling format of labeled data, for example, image classification
                scene=wf.data.LabelTaskTypeEnum.IMAGE_CLASSIFICATION # Labeling scene
            )
        ]
    )
)

workflow = wf.Workflow(
    name="dataset-import-demo",
    desc="this is a demo workflow",
    steps=[dataset_import]
)

You import unlabeled data in a specified path to a labeling job. Then, you can add a labeling phase to label the imported data.

Data preparation: Create a labeling job using a specified dataset and upload the unlabeled data to OBS.

from modelarts import workflow as wf
# Use DatasetImportStep to import data in a specified path to a labeling job and output the labeling job.

# Define a labeling job.
label_task = wf.data.LabelTaskPlaceholder(name="label_task_placeholder_name")

# Define the OBS data.
obs = wf.data.OBSPlaceholder(name = "obs_placeholder_name", object_type = "directory" ) # object_type must be file or directory.

dataset_import = wf.steps.DatasetImportStep(
    name="data_import", # Name of the dataset import phase. The name contains a maximum of 64 characters, including only letters, digits, underscores (_), and hyphens (-). It must start with a letter and must be unique in a workflow.
    title="Dataset Import", # Title, which defaults to the value of name
    inputs=[
        wf.steps.DatasetImportInput(name="input_name_1", data=label_task), # Labeling job object, configured when the workflow is running. You can also use wf.data.LabelTask(dataset_name="dataset_name", task_name="label_task_name") for the data field.
        wf.steps.DatasetImportInput(name="input_name_2", data=obs) # Storage path to the imported dataset, configured when the workflow is running. You can also use wf.data.OBSPath(obs_path="obs_path") for the data field.
    ],# DatasetImportStep inputs
    outputs=wf.steps.DatasetImportOutput(name="output_name"), # DatasetImportStep outputs
)

workflow = wf.Workflow(
    name="dataset-import-demo",
    desc="this is a demo workflow",
    steps=[dataset_import]
)

Scenario 3: Creating a dataset import phase using the outputs of the dataset creation phase.

from modelarts import workflow as wf
# Use DatasetImportStep to import data in a specified path to a dataset and output the dataset.

# Define the OBS data.
obs = wf.data.OBSPlaceholder(name = "obs_placeholder_name", object_type = "directory" ) # object_type must be file or directory.

dataset_import = wf.steps.DatasetImportStep(
    name="data_import", # Name of the dataset import phase. The name contains a maximum of 64 characters, including only letters, digits, underscores (_), and hyphens (-). It must start with a letter and must be unique in a workflow.
    title="Dataset Import", # Title, which defaults to the value of name
    inputs=[
        wf.steps.DatasetImportInput(name="input_name_1", data=create_dataset.outputs["create_dataset_output"].as_input()), # The outputs of the dataset creation phase are used as the inputs of the dataset import phase.
        wf.steps.DatasetImportInput(name="input_name_2", data=obs) # Storage path to the imported dataset, configured when the workflow is running. You can also use wf.data.OBSPath(obs_path="obs_path") for the data field.
    ],# DatasetImportStep inputs
    outputs=wf.steps.DatasetImportOutput(name="output_name"), # DatasetImportStep outputs
    depend_steps=create_dataset # Preceding dataset creation phase
)
# create_dataset is an instance of wf.steps.CreateDatasetStep. create_dataset_output is the name field value of wf.steps.CreateDatasetOutput.

workflow = wf.Workflow(
    name="dataset-import-demo",
    desc="this is a demo workflow",
    steps=[dataset_import]
)