Skip to main content

CLI

important

This page covers clearml-data, ClearML's file-based data management solution. See Hyper-Datasets for ClearML's advanced queryable dataset management solution.

clearml-data is a data management CLI tool that comes as part of the clearml python package. Use clearml-data to create, modify, and manage your datasets. You can upload your dataset to any storage service of your choice (S3 / GS / Azure / Network Storage) by setting the dataset’s upload destination (see --storage). Once you have uploaded your dataset, you can access it from any machine.

The following page provides a reference to clearml-data's CLI commands.

create#

Creates a new dataset.

clearml-data create [-h] [--parents [PARENTS [PARENTS ...]]] [--project PROJECT]
--name NAME [--version VERSION] [--output-uri OUTPUT_URI]
[--tags [TAGS [TAGS ...]]]

Parameters

NameDescriptionOptional
--nameDataset's nameNo
--projectDataset's projectNo
--versionDataset version. Use the semantic versioning scheme. If not specified a version will automatically be assignedYes
--parentsIDs of the dataset's parents. The dataset inherits all of its parents' content. Multiple parents can be entered, but they are merged in the order they were enteredYes
--output-uriSets where dataset and its previews are uploaded toYes
--tagsDataset user tags. The dataset can be labeled, which can be useful for organizing datasetsYes
Dataset ID
  • For datasets created with clearml v1.6 or newer on ClearML Server v1.6 or newer, find the ID in the dataset version’s info panel in the Dataset UI.
    For datasets created with earlier versions of clearml, or if using an earlier version of ClearML Server, find the ID in the task header of the dataset task's info panel.
  • clearml-data works in a stateful mode so once a new dataset is created, the following commands do not require the --id flag.

add#

Add individual files or complete folders to the dataset.

clearml-data add [-h] [--id ID] [--dataset-folder DATASET_FOLDER]
[--files [FILES [FILES ...]]] [--wildcard [WILDCARD [WILDCARD ...]]]
[--links [LINKS [LINKS ...]]] [--non-recursive] [--verbose]

Parameters

NameDescriptionOptional
--idDataset's ID. Default: previously created / accessed datasetYes
--filesFiles / folders to add. Items will be uploaded to the dataset’s designated storage.Yes
--wildcardAdd specific set of files, denoted by these wildcards. For example: ~/data/*.jpg ~/data/json. Multiple wildcards can be passed.Yes
--linksFiles / folders link to add. Supports s3, gs, azure links. Example: s3://bucket/data azure://bucket/folder. Items remain in their original location.Yes
--dataset-folderDataset base folder to add the files to in the dataset. Default: dataset rootYes
--non-recursiveDisable recursive scan of filesYes
--verboseVerbose reportingYes

remove#

Remove files/links from the dataset.

clearml-data remove [-h] [--id ID] [--files [FILES [FILES ...]]]
[--non-recursive] [--verbose]

Parameters

NameDescriptionOptional
--idDataset's ID. Default: previously created / accessed datasetYes
--filesFiles / folders to remove (wildcard selection is supported, for example: ~/data/*.jpg ~/data/json). Notice: file path is the path within the dataset, not the local path. For links, you can specify their URL (e.g. s3://bucket/data)No
--non-recursiveDisable recursive scan of filesYes
--verboseVerbose reportingYes

upload#

Upload the local dataset changes to the server. By default, it's uploaded to the ClearML Server. It's possible to specify a different storage medium by entering an upload destination, such as s3://bucket, gs://, azure://, /mnt/shared/.

clearml-data upload [-h] [--id ID] [--storage STORAGE] [--chunk-size CHUNK_SIZE]
[--verbose]

Parameters

NameDescriptionOptional
--idDataset's ID. Default: previously created / accessed datasetYes
--storageRemote storage to use for the dataset files. Default: files_serverYes
--chunk-sizeSet dataset artifact upload chunk size in MB. Default 512, (pass -1 for a single chunk). Example: 512, dataset will be split and uploaded in 512 MB chunks.Yes
--verboseVerbose reportingYes

close#

Finalize the dataset and make it ready to be consumed. This automatically uploads all files that were not previously uploaded. Once a dataset is finalized, it can no longer be modified.

clearml-data close [-h] [--id ID] [--storage STORAGE] [--disable-upload]
[--chunk-size CHUNK_SIZE] [--verbose]

Parameters

NameDescriptionOptional
--idDataset's ID. Default: previously created / accessed datasetYes
--storageRemote storage to use for the dataset files. Default: files_serverYes
--disable-uploadDisable automatic upload when closing the datasetYes
--chunk-sizeSet dataset artifact upload chunk size in MB. Default 512, (pass -1 for a single chunk). Example: 512, dataset will be split and uploaded in 512 MB chunks.Yes
--verboseVerbose reportingYes

sync#

Sync a folder's content with ClearML. This option is useful in case a user has a single point of truth (i.e. a folder) which updates from time to time.

Once an update should be reflected in ClearML's system, call clearml-data sync and pass the folder path, and the changes (either file addition, modification and removal) will be reflected in ClearML.

This command also uploads the data and finalizes the dataset automatically.

clearml-data sync [-h] [--id ID] [--dataset-folder DATASET_FOLDER] --folder FOLDER
[--parents [PARENTS [PARENTS ...]]] [--project PROJECT] [--name NAME]
[--version VERSION] [--output-uri OUTPUT_URI] [--tags [TAGS [TAGS ...]]]
[--storage STORAGE] [--skip-close] [--chunk-size CHUNK_SIZE] [--verbose]

Parameters

NameDescriptionOptional
--idDataset's ID. Default: previously created / accessed datasetYes
--dataset-folderDataset base folder to add the files to (default: Dataset root)Yes
--folderLocal folder to sync. Wildcard selection is supported, for example: ~/data/*.jpg ~/data/jsonNo
--storageRemote storage to use for the dataset files. Default: files serverYes
--parentsIDs of the dataset's parents (i.e. merge all parents). All modifications made to the folder since the parents were synced will be reflected in the datasetYes
--projectIf creating a new dataset, specify the dataset's project nameYes
--nameIf creating a new dataset, specify the dataset's nameYes
--versionSpecify the dataset’s version using the semantic versioning scheme. Default: 1.0.0Yes
--tagsDataset user tagsYes
--skip-closeDo not auto close dataset after syncing foldersYes
--chunk-sizeSet dataset artifact upload chunk size in MB. Default 512, (pass -1 for a single chunk). Example: 512, dataset will be split and uploaded in 512 MB chunks.Yes
--verboseVerbose reportingYes

list#

List a dataset's contents.

clearml-data list [-h] [--id ID] [--project PROJECT] [--name NAME] [--version VERSION]
[--filter [FILTER [FILTER ...]]] [--modified]

Parameters

NameDescriptionOptional
--idDataset ID whose contents will be shown (alternatively, use project / name combination). Default: previously accessed datasetYes
--projectSpecify dataset project name (if used instead of ID, dataset name is also required)Yes
--nameSpecify dataset name (if used instead of ID, dataset project is also required)Yes
--versionSpecify dataset version. Default: most recent versionYes
--filterFilter files based on folder / wildcard. Multiple filters are supported. Example: folder/date_*.json folder/sub-folderYes
--modifiedOnly list file changes (add / remove / modify) introduced in this versionYes

set-description#

Sets the description of an existing dataset.

clearml-data set-description [-h] [--id ID] [--description DESCRIPTION]

Parameters

NameDescriptionOptional
--idDataset’s IDNo
--descriptionDescription to be setNo

delete#

Deletes dataset(s). Pass any of the attributes of the dataset(s) you want to delete. Multiple datasets matching the request will raise an exception, unless you pass --entire-dataset and --force. In this case, all matching datasets will be deleted.

If a dataset is a parent to a dataset(s), you must pass --force in order to delete it.

warning

Deleting a parent dataset may cause child datasets to lose data!

clearml-data delete [-h] [--id ID] [--project PROJECT] [--name NAME]
[--version VERSION] [--force] [--entire-dataset]

Parameters

NameDescriptionOptional
--idID of the dataset to delete (alternatively, use project / name combination).Yes
--projectSpecify dataset project name (if used instead of ID, dataset name is also required)Yes
--nameSpecify dataset name (if used instead of ID, dataset project is also required)Yes
--versionSpecify dataset versionYes
-–forceForce dataset deletion even if other dataset versions depend on it. Must also be used if --entire-dataset flag is usedYes
--entire-datasetDelete all found datasetsYes

rename#

Rename a dataset (and all of its versions).

clearml-data rename [-h] --new-name NEW_NAME --project PROJECT --name NAME

Parameters

NameDescriptionOptional
--new-nameThe new name of the datasetNo
--projectThe project the dataset to be renamed belongs toNo
--nameThe current name of the dataset(s) to be renamedNo

move#

Moves a dataset to another project

clearml-data move [-h] --new-project NEW_PROJECT --project PROJECT --name NAME

Parameters

NameDescriptionOptional
--new-projectThe new project of the datasetNo
--projectThe current project the dataset to be move belongs toNo
--nameThe name of the dataset to be movedNo

search#

Search datasets in the system by project, name, ID, and/or tags.

Returns list of all datasets in the system that match the search request, sorted by creation time.

clearml-data search [-h] [--ids [IDS [IDS ...]]] [--project PROJECT]
[--name NAME] [--tags [TAGS [TAGS ...]]]

Parameters

NameDescriptionOptional
--idsA list of dataset IDsYes
--projectThe project name of the datasetsYes
--nameA dataset name or a partial name to filter datasets byYes
--tagsA list of dataset user tagsYes

compare#

Compare two datasets (target vs. source). The command returns a comparison summary that looks like this: Comparison summary: 4 files removed, 3 files modified, 0 files added

clearml-data compare [-h] --source SOURCE --target TARGET [--verbose]

Parameters

NameDescriptionOptional
--sourceSource dataset ID (used as baseline)No
--targetTarget dataset ID (compare against the source baseline dataset)No
--verboseVerbose report all file changes (instead of summary)Yes

squash#

Squash multiple datasets into a single dataset version (merge down).

clearml-data squash [-h] --name NAME --ids [IDS [IDS ...]] [--storage STORAGE] [--verbose]

Parameters

NameDescriptionOptional
--nameCreate squashed dataset nameNo
--idsSource dataset IDs to squash (merge down)No
--storageRemote storage to use for the dataset files. Default: files_serverYes
--verboseVerbose report all file changes (instead of summary)Yes

verify#

Verify that the dataset content matches the data from the local source.

clearml-data verify [-h] [--id ID] [--folder FOLDER] [--filesize] [--verbose]

Parameters

NameDescriptionOptional
--idSpecify dataset ID. Default: previously created/accessed datasetYes
--folderSpecify dataset local copy (if not provided the local cache folder will be verified)Yes
--filesizeIf True, only verify file size and skip hash checks (default: False)Yes
--verboseVerbose report all file changes (instead of summary)Yes

get#

Get a local copy of a dataset. By default, you get a read only cached folder, but you can get a mutable copy by using the --copy flag.

clearml-data get [-h] [--id ID] [--copy COPY] [--link LINK] [--part PART]
[--num-parts NUM_PARTS] [--overwrite] [--verbose]

Parameters

NameDescriptionOptional
--idSpecify dataset ID. Default: previously created / accessed datasetYes
--copyGet a writable copy of the dataset to a specific output folderYes
--linkCreate a soft link (not supported on Windows) to a read-only cached folder containing the datasetYes
--partRetrieve a partial copy of the dataset. Part number (0 to --num-parts-1) of total parts --num-parts.Yes
--num-partsTotal number of parts to divide the dataset into. Notice, minimum retrieved part is a single chunk in a dataset (or its parents). Example: Dataset gen4, with 3 parents, each with a single chunk, can be divided into 4 partsYes
--overwriteIf True, overwrite the target folderYes
--verboseVerbose report all file changes (instead of summary)Yes

publish#

Publish the dataset for public use. The dataset must be finalized before it is published.

clearml-data publish [-h] --id ID

Parameters

NameDescriptionOptional
--idThe dataset task ID to be published.No