Meet Metalad, an extension to datalad that supports metadata handling

This chapter will show you how to install metalad, create a datalad dataset and how to work with metadata. Working with metadata means, adding metadata, viewing metadata, and aggregating metadata.

Below you will find examples and instructions on how to install metalad, how to create an example dataset and how to work with metadata. You will learn how to add metadata, how to display metadata that is stored locally or remotely, and how to combine (aka aggregate) metadata from multiple datasets.


Install metalad and create a datalad dataset

Install metalad

Create a virtual python environment and activate it (Linux example shown).

> python3 -m venv venv/datalad-metalad
> source venv/datalad-metalad/bin/activate

Install the latest version of metalad

> pip install --upgrade datalad-metalad

Create a datalad dataset

The following command creates a datalad dataset that stores text-files in git and non-text files in git-annex.

> datalad create -c text2git example-ds

Add a text-file to the dataset

> cd example-ds
> echo some content > file1.txt
> datalad save

Add a binary file to the dataset

> dd if=/dev/zero of=file.bin count=1000
> datalad save

Working with metadata

This chapter provides an overview of commands in metalad. If you want to continue with the hands-on examples, skip to next chapter (and probably come back here later).

Metalad allows you to associate metadata with datalad datasets or files in datalad-datasets. More specifically, metadata can be associated with: datasets, sub-datasets, and files in a dataset or sub-dataset.

Metalad can associate an arbitrary amount of individual metadata instances with a single element (dataset or file). Each metadata instance is identified by a type-name that specifies the type of the data contained in the metadata instance. For example: metalad_core, bids, etc.


Implementation side note: The metadata associations can in principle be stored in any git-repository, but are by default stored in the git-repository of a root dataset.

The plumbing

Metalad has a few basic commands, aka plumbing commands, that perform essential elementary metadata operations:

  • meta-add add metadata to an element (dataset, sub-dataset, or file)
  • meta-dump show metadata stored in a local or remote dataset

The porcelain

To simplify working with metadata, metalad provides a number of higher level functions and tools that implement typical use cases.

  • meta-extract run an extractor (see below) on an existing dataset, sub-dataset, or file and emit metadata that can be fed to meta-add.
  • meta-conduct run pipelines of extractors and adders on locally available datatasets, sub-datasets and files, in order to automatate metadata extraction and adding tasks
  • meta-aggregate combine metadata from number of sub-datasets into the root-dataset.
  • meta-filter walk through metadata from multiple stores, apply a filter, and output new metadata

Metadata extractors

Datalad supports pluggable metadata extractors. Metadata extractors can perform arbitrary operations on the given element (dataset, sub-dataset, or file) and return arbitrary metadata in JSON-format. Meta-extract will associate the metadata with the metadata element.

Metalad comes with a number of extractors. Some extractors are provided by metalad, some are inherited from datalad. The provided extractors generate provenance records for datasets and data, or they extract metadata from specific files or data-structures, e.g. BIDS. In principle any processing is possible. There is also a generic extractor, which allows to invoke external commands to generate metadata.

Metadata extraction examples

Extract dataset-level metadata

Extract dataset-level metadata with the datalad command meta-extract. It takes a number of optional arguments and one required argument, the name of the metadata extractor that should be used. We use metalad_core for now.

> datalad meta-extract metalad_core

The extracted metadata will be written to stdout and will look similar to this (times, names, and UUIDs will be different for you):

{"type": "dataset", "dataset_id": "853d9356-fc2e-459e-96bc-02414a1fef93", "dataset_version": "8d6d0e50a27b7540717360e21332b1ad0c924415", "extractor_name": "metalad_core", "extractor_version": "1", "extraction_parameter": {}, "extraction_time": 1637921555.282522, "agent_name": "Your Name", "agent_email": "", "extracted_metadata": {"@context": {"@vocab": "", "datalad": ""}, "@graph": [{"@id": "59286713dacabfbce1cecf4c865fff5a", "@type": "agent", "name": "Your Name", "email": ""}, {"@id": "8d6d0e50a27b7540717360e21332b1ad0c924415", "identifier": "853d9331-fc2e-459e-96bc-02414a1fef93", "@type": "Dataset", "version": "0-3-g8d6d0e5", "dateCreated": "2021-11-26T11:03:25+01:00", "dateModified": "2021-11-26T11:09:27+01:00", "hasContributor": {"@id": "59286713dacabfbce1cecf4c865fff5a"}}]}}

The output is a JSON-serialized object. You can use `jq <>`__ to get a nicer formatting of the JSON-object. For example the command:

> datalad meta-extract metalad_core|jq .

would result in an output similar to:

  "type": "dataset",
  "dataset_id": "853d9356-fc2e-459e-96bc-02414a1fef93",
  "dataset_version": "ee512961b878a674c8068e54656e161d40566d9b",
  "extractor_name": "metalad_core",
  "extractor_version": "1",
  "extraction_parameter": {},
  "extraction_time": 1637923596.9511302,
  "agent_name": "Your Name",
  "agent_email": "",
  "extracted_metadata": {
    "@context": {
      "@vocab": "",
      "datalad": ""
    "@graph": [
        "@id": "59286713dacabfbce1cecf4c865fff5a",
        "@type": "agent",
        "name": "Your Name",
        "email": ""
        "@id": "ee512961b878a674c8068e54656e161d40566d9b",
        "identifier": "853d9356-fc2e-459e-96bc-02414a1fef93",
        "@type": "Dataset",
        "version": "0-4-gee51296",
        "dateCreated": "2021-11-26T11:03:25+01:00",
        "dateModified": "2021-11-26T11:13:58+01:00",
        "hasContributor": {
          "@id": "59286713dacabfbce1cecf4c865fff5a"

Extract file-level metadata

The datalad command meta-extract also support the extraction of file-level metadata. File-level metadata extraction requires a second argument, besides the extractor-name, to datalad meta-extract. The second argument identifies the file for which metadata should be extracted.

NB: you must specify an extractor that supports file-level extraction if a file-name is passed to datalad meta-extract, and an extractor that supports dataset-level extraction if no file-name is passed to datalad meta-extract. The extractor metalad_core supports both metadata levels.

To extract metadata for the file file1.txt, execute the following command:

> datalad meta-extract metalad_core file1.txt

which will lead to an output similar to:

{"type": "file", "dataset_id": "853d9331-fc2e-459e-96bc-02414a1fef93", "dataset_version": "ee512961b878a674c8068e54656e161d40566d9b", "path": "file1.txt", "extractor_name": "metalad_core", "extractor_version": "1", "extraction_parameter": {}, "extraction_time": 1637927097.2165475, "agent_name": "Your Name", "agent_email": "", "extracted_metadata": {"@id": "datalad:SHA1-s13--2ef267e25bd6c6a300bb473e604b092b6a48523b", "contentbytesize": 13}}

Add metadata

You can add extracted metadata to the dataset (metadata will be stored in a special area of the git-repository and not interfere with your data in the dataset).

To add metadata you use the datalad command meta-add. The meta-add command takes on required argument, the name of a file that contains metadata in JSON-format. It also supports reading JSON-metadata from stdin, if you provided - as the file name. That mean you can pipe the output of meta-extract directly into meta-add by specifying - as metadata file-name like this:

> datalad meta-extract metalad_core |datalad meta-add -

meta-add supports files that contain lists of JSON-records in “JSON Lines”-format (see

Let’s add the file-level metadata for file1.txt and file.bin to the metadata of the dataset by executing the two commands:

> datalad meta-extract metalad_core file1.txt |datalad meta-add -


> datalad meta-extract metalad_core file.bin |datalad meta-add -

Display (retrieve) metadata

To view the metadata that has been stored in a dataset, you can use the datalad command meta-dump. The following command will show all metadata that is stored in the dataset. Metadata is displayed in JSON Lines-format (aka newline-delimited JSON), which is a number of lines where each line contains a serialized JSON object.

datalad meta-dump -r

Its execution will generate a result similar to:

{"type": "dataset", "dataset_id": "853d9356-fc2e-459e-96bc-02414a1fef93", "dataset_version": "ee512961b878a674c8068e54656e161d40566d9b", "extraction_time": 1637924361.8114567, "agent_name": "Your Name", "agent_email": "", "extractor_name": "metalad_core", "extractor_version": "1", "extraction_parameter": {}, "extracted_metadata": {"@context": {"@vocab": "", "datalad": ""}, "@graph": [{"@id": "59286713dacabfbce1cecf4c865fff5a", "@type": "agent", "name": "Your Name", "email": ""}, {"@id": "ee512961b878a674c8068e54656e161d40566d9b", "identifier": "853d9356-fc2e-459e-96bc-02414a1fef93", "@type": "Dataset", "version": "0-4-gee51296", "dateCreated": "2021-11-26T11:03:25+01:00", "dateModified": "2021-11-26T11:13:58+01:00", "hasContributor": {"@id": "59286713dacabfbce1cecf4c865fff5a"}}]}}
{"type": "file", "path": "file1.txt", "dataset_id": "853d9356-fc2e-459e-96bc-02414a1fef93", "dataset_version": "ee512961b878a674c8068e54656e161d40566d9b", "extraction_time": 1637927239.2590044, "agent_name": "Your Name", "agent_email": "", "extractor_name": "metalad_core", "extractor_version": "1", "extraction_parameter": {}, "extracted_metadata": {"@id": "datalad:SHA1-s13--2ef267e25bd6c6a300bb473e604b092b6a48523b", "contentbytesize": 13}}
{"type": "file", "path": "file.bin", "dataset_id": "853d9356-fc2e-459e-96bc-02414a1fef93", "dataset_version": "ee512961b878a674c8068e54656e161d40566d9b", "extraction_time": 1637927246.2115273, "agent_name": "Your Name", "agent_email": "", "extractor_name": "metalad_core", "extractor_version": "1", "extraction_parameter": {}, "extracted_metadata": {"@id": "datalad:MD5E-s512000--816df6f64deba63b029ca19d880ee10a.bin", "contentbytesize": 512000}}

Adding a lot of metadata with meta-conduct

To extract and add metadata from a large number of files or from all files of a dataset you can use meta-conduct. Meta-conduct can be configured to execute a number of meta-extract and meta-add commands automatically in parallel. The operations that meta-conduct should perform are defined in pipeline definitions. A few pipeline definitions are provided with metalad, and we will use the extract_metadata pipeline.

Adding dataset-level metadata

Execute the following command:

datalad meta-conduct extract_metadata traverser:`pwd` traverser:dataset extractor:dataset extractor:metalad_core

You will get an output which is similar to:

meta_conduct(ok): <...>/gist/example-ds

What happened?

You just ran the extract_metadata pipeline and specified that you want to traverse the current directory (traverser:`pwd`), and that you want to operate on all datasets that are encountered (traverser:Dataset). You also specified that, for each element found during traversal, you would like to execute a dataset-level extractor (extractor:dataset) with the name metalad_core (extractor:metalad_core).

The pipeline found one dataset in the current directory and added the metadata to it. Since you have done that already before using meta-extract and meta-add, you have the same number of metadata entries in the metadata store. That means datalad meta-dump -r will give you three results. But you might notice that the extraction-time of the dataset-level entry has changed.

Metalad comes with different pre-built pipelines. Some allow to automatically fetch an annexed file and automatically drop said file, after is has been processed.

Adding file-level metadata

You can also add file-metadata using meta-conduct. Execute the following command:

datalad meta-conduct extract_metadata traverser:`pwd` traverser:file extractor:file extractor:metalad_core

You will get an output which is similar to:

meta_conduct(ok): <...>/example-ds/file1.txt
meta_conduct(ok): <...>/example-ds/file.bin
action summary:
  meta_conduct (ok: 2)

What happened here?

The traverser found two elements that fitted your description (traverser:Dataset), executed the specified extractor on them (extractor:metalad_core), and added the results to the metadata storage.

Again, you can verify this with the value of extraction_time in the output of datalad meta-dump -r.

Joining metadata from multiple datasets with meta-aggregate

Let’s have a look at meta-aggregate. The command meta-aggregate copies metadata from sub-datasets into the metadata store of the root dataset.

Subdataset creation

To see meta-aggregate in action we first create a sub-datasets:

> datalad create -d . -c text2git subds1

This command will yield an output similar to:

[INFO   ] Creating a new annex repo at <...>/example-ds/subds1
[INFO   ] Running procedure cfg_text2git
[INFO   ] == Command start (output follows) =====
[INFO   ] == Command exit (modification check follows) =====
add(ok): subds1 (file)
add(ok): .gitmodules (file)
save(ok): . (dataset)
create(ok): subds1 (dataset)
action summary:
  add (ok: 2)
  create (ok: 1)
  save (ok: 1)

Create some content and save it:

> cd subds1
> echo content of subds1/file_subds1.1.txt > file_subds1.1.txt
> datalad save

Now run the file level extractor in the subdataset:

> datalad meta-conduct extract_metadata traverser:`pwd` traverser:file extractor:file extractor:metalad_core

and the dataset-level extractor:

> datalad meta-conduct extract_metadata traverser:`pwd` traverser:dataset extractor:dataset extractor:metalad_core

If you want you can view the added metadata in the subdataset with the command datalad meta-dump -r.

Since we modified the subdataset, we should also save the root dataset:

> cd ..
> datalad save


After all the above commands are executed, we have metadata stored in two datasets (more precisely, in the metadata stores of the datasets which are the git repositories). In the metadata store of example-ds we have the following information:

And in the metadata store of subds1 we have:

Now let us aggregate the subdataset metadata into the root dataset with the command meta-aggregate:

> datalad meta-aggregate -d . subds1

And display the result:

> datalad meta-dump -r

The output will contain five JSON records (in 5 lines), three from the top-level datasets and two from the subdataset. It will look similar to this:

{"type": "dataset", "dataset_id": "ceeb844a-c6e8-4b2f-bb7c-62b7ae449a9f", "dataset_version": "bcf9cfde4a599d26094a58efbe4369e0878cb9c8", "extraction_time": 1638357863.4242253, "agent_name": "Your Name", "agent_email": "", "extractor_name": "metalad_core", "extractor_version": "1", "extraction_parameter": {}, "extracted_metadata": {"@context": {"@vocab": "", "datalad": ""}, "@graph": [{"@id": "59286713dacabfbce1cecf4c865fff5a", "@type": "agent", "name": "Your Name", "email": ""}, {"@id": "bcf9cfde4a599d26094a58efbe4369e0878cb9c8", "identifier": "ceeb844a-c6e8-4b2f-bb7c-62b7ae449a9f", "@type": "Dataset", "version": "0-4-gbcf9cfd", "dateCreated": "2021-12-01T12:24:17+01:00", "dateModified": "2021-12-01T12:24:19+01:00", "hasContributor": {"@id": "59286713dacabfbce1cecf4c865fff5a"}}]}}
{"type": "file", "path": "file1.txt", "dataset_id": "ceeb844a-c6e8-4b2f-bb7c-62b7ae449a9f", "dataset_version": "bcf9cfde4a599d26094a58efbe4369e0878cb9c8", "extraction_time": 1638357864.5259314, "agent_name": "Your Name", "agent_email": "", "extractor_name": "metalad_core", "extractor_version": "1", "extraction_parameter": {}, "extracted_metadata": {"@id": "datalad:SHA1-s13--2ef267e25bd6c6a300bb473e604b092b6a48523b", "contentbytesize": 13}}
{"type": "file", "path": "file.bin", "dataset_id": "ceeb844a-c6e8-4b2f-bb7c-62b7ae449a9f", "dataset_version": "bcf9cfde4a599d26094a58efbe4369e0878cb9c8", "extraction_time": 1638357864.5327883, "agent_name": "Your Name", "agent_email": "", "extractor_name": "metalad_core", "extractor_version": "1", "extraction_parameter": {}, "extracted_metadata": {"@id": "datalad:MD5E-s512000--816df6f64deba63b029ca19d880ee10a.bin", "contentbytesize": 512000}}
{"type": "dataset", "root_dataset_id": "<unknown>", "root_dataset_version": "7228f027171f7b8949a47812a651600412f2577e", "dataset_path": "subds1", "dataset_id": "4e3422f4-b606-4cf9-818a-a3bb840e3396", "dataset_version": "ddf2a2758fd6773a1171a6fbae4afe48cc982773", "extraction_time": 1638357869.7052076, "agent_name": "Your Name", "agent_email": "", "extractor_name": "metalad_core", "extractor_version": "1", "extraction_parameter": {}, "extracted_metadata": {"@context": {"@vocab": "", "datalad": ""}, "@graph": [{"@id": "59286713dacabfbce1cecf4c865fff5a", "@type": "agent", "name": "Your Name", "email": ""}, {"@id": "ddf2a2758fd6773a1171a6fbae4afe48cc982773", "identifier": "4e3422f4-b606-4cf9-818a-a3bb840e3396", "@type": "Dataset", "version": "0-3-gddf2a27", "dateCreated": "2021-12-01T12:24:25+01:00", "dateModified": "2021-12-01T12:24:27+01:00", "hasContributor": {"@id": "59286713dacabfbce1cecf4c865fff5a"}}]}}
{"type": "file", "path": "file_subds1.1.txt", "root_dataset_id": "<unknown>", "root_dataset_version": "7228f027171f7b8949a47812a651600412f2577e", "dataset_path": "subds1", "dataset_id": "4e3422f4-b606-4cf9-818a-a3bb840e3396", "dataset_version": "ddf2a2758fd6773a1171a6fbae4afe48cc982773", "extraction_time": 1638357868.706351, "agent_name": "Your Name", "agent_email": "", "extractor_name": "metalad_core", "extractor_version": "1", "extraction_parameter": {}, "extracted_metadata": {"@id": "datalad:SHA1-s36--9ce18068eb4126c23235d965c179b2a53546d104", "contentbytesize": 36}}


Upcoming: how to delete metadata, how to filter metadata, and how to export metadata.