Query & search registries

This guide walks through all the ways of finding metadata records in LaminDB registries.

# !pip install lamindb
!lamin init --storage ./test-registries
Hide code cell output 
→ connected lamindb: testuser1/test-registries

We’ll need some toy data.

import lamindb as ln

# create toy data
ln.Artifact(ln.core.datasets.file_jpg_paradisi05(), description="My image").save()
ln.Artifact.from_df(ln.core.datasets.df_iris(), description="The iris collection").save()
ln.Artifact(ln.core.datasets.file_fastq(), description="My fastq").save()

# see the content of the artifact registry
ln.Artifact.df()
Hide code cell output 
→ connected lamindb: testuser1/test-registries

! no run & transform got linked, call `ln.track()` & re-run

! no run & transform got linked, call `ln.track()` & re-run

! no run & transform got linked, call `ln.track()` & re-run
uid version is_latest description key suffix type size hash n_objects n_observations _hash_type _accessor visibility _key_is_virtual storage_id transform_id run_id created_at created_by_id
id
3 de6fDVsyNVop7WNs0000 None True My fastq None .fastq.gz None 20 hi7ZmAzz8sfMd3vIQr-57Q None None md5 None 1 True 1 None None 2024-10-22 09:24:45.661507+00:00 1
2 v2MJDTEJ042Agknn0000 None True The iris collection None .parquet dataset 5629 5axHfO_2Pmlun2sJZHVuNg None None md5 DataFrame 1 True 1 None None 2024-10-22 09:24:45.652752+00:00 1
1 wJem73xED8zcmjFV0000 None True My image None .jpg None 29358 r4tnqmKI_SjrkdLzpuWp4g None None md5 None 1 True 1 None None 2024-10-22 09:24:40.487582+00:00 1

Look up metadata

For registries with less than 100k records, auto-completing a Lookup object is the most convenient way of finding a record.

For example, take the User registry:

# query the database for all users, optionally pass the field that creates the key
users = ln.User.lookup(field="handle")

# the lookup object is a NamedTuple
users
Hide code cell output 
Lookup(testuser1=User(uid='DzTjkKse', handle='testuser1', name='Test User1', created_at=2024-10-22 09:24:36 UTC), dict=<bound method Lookup.dict of <lamin_utils._lookup.Lookup object at 0x7fba903c6090>>)

With auto-complete, we find a specific user record:

user = users.testuser1
user
Hide code cell output 
User(uid='DzTjkKse', handle='testuser1', name='Test User1', created_at=2024-10-22 09:24:36 UTC)

You can also get a dictionary:

users_dict = ln.User.lookup().dict()
users_dict
Hide code cell output 
{'testuser1': User(uid='DzTjkKse', handle='testuser1', name='Test User1', created_at=2024-10-22 09:24:36 UTC)}

Query exactly one record

get errors if more than one matching records are found.

# by the universal base62 uid
ln.User.get("DzTjkKse")

# by any expression involving fields
ln.User.get(handle="testuser1")
Hide code cell output 
User(uid='DzTjkKse', handle='testuser1', name='Test User1', created_at=2024-10-22 09:24:36 UTC)

Query sets of records

Filter for all artifacts created by a user:

ln.Artifact.filter(created_by=user).df()
Hide code cell output
uid version is_latest description key suffix type size hash n_objects n_observations _hash_type _accessor visibility _key_is_virtual storage_id transform_id run_id created_at created_by_id
id
1 wJem73xED8zcmjFV0000 None True My image None .jpg None 29358 r4tnqmKI_SjrkdLzpuWp4g None None md5 None 1 True 1 None None 2024-10-22 09:24:40.487582+00:00 1
2 v2MJDTEJ042Agknn0000 None True The iris collection None .parquet dataset 5629 5axHfO_2Pmlun2sJZHVuNg None None md5 DataFrame 1 True 1 None None 2024-10-22 09:24:45.652752+00:00 1
3 de6fDVsyNVop7WNs0000 None True My fastq None .fastq.gz None 20 hi7ZmAzz8sfMd3vIQr-57Q None None md5 None 1 True 1 None None 2024-10-22 09:24:45.661507+00:00 1

To access the results encoded in a filter statement, execute its return value with one of:

  • .df(): A pandas DataFrame with each record in a row.

  • .all(): A QuerySet.

  • .one(): Exactly one record. Will raise an error if there is none. Is equivalent to the .get() method shown above.

  • .one_or_none(): Either one record or None if there is no query result.

Note

filter() returns a QuerySet.

The ORMs in LaminDB are Django Models and any Django query works. LaminDB extends Django’s API for data scientists.

Under the hood, any .filter() call translates into a SQL select statement.

.one() and .one_or_none() are two parts of LaminDB’s API that are borrowed from SQLAlchemy.

Search for records

Search the toy data:

ln.Artifact.search("iris").df()
Hide code cell output
uid version is_latest description key suffix type size hash n_objects n_observations _hash_type _accessor visibility _key_is_virtual storage_id transform_id run_id created_at created_by_id
id
2 v2MJDTEJ042Agknn0000 None True The iris collection None .parquet dataset 5629 5axHfO_2Pmlun2sJZHVuNg None None md5 DataFrame 1 True 1 None None 2024-10-22 09:24:45.652752+00:00 1

Let us create 500 notebook objects with fake titles, save, and search them:

transforms = [ln.Transform(name=title, type="notebook") for title in ln.core.datasets.fake_bio_notebook_titles(n=500)]
ln.save(transforms)

# search
ln.Transform.search("intestine").df().head(5)
Hide code cell output
uid version is_latest name key description type source_code hash reference reference_type _source_code_artifact_id created_at created_by_id
id
30 rFcG9Ubv3jFy0000 None True Cluster intestinal candidate IgG research inte... None None notebook None None None None None 2024-10-22 09:24:54.886469+00:00 1
61 PpS4YDwGxY9F0000 None True Mast Cell visualize IgA IgA intestine. None None notebook None None None None None 2024-10-22 09:24:54.889386+00:00 1
74 c3L9c9fvYYOg0000 None True Igg IgA Ciliary body Speed cells cluster intes... None None notebook None None None None None 2024-10-22 09:24:54.894755+00:00 1
83 jNDQ2NCnaaqU0000 None True Iga IgG3 IgA Cortical Ciliary body intestine I... None None notebook None None None None None 2024-10-22 09:24:54.895574+00:00 1
84 wjJ4z4zkJQw60000 None True Igg Cortical intestine IgG3 Efferent ducts IgD... None None notebook None None None None None 2024-10-22 09:24:54.895665+00:00 1

Note

Currently, the LaminHub UI search is more powerful than the search of the lamindb open-source package.

Leverage relations

Django has a double-under-score syntax to filter based on related tables.

This syntax enables you to traverse several layers of relations and leverage different comparators.

ln.Artifact.filter(created_by__handle__startswith="testuse").df()
Hide code cell output
uid version is_latest description key suffix type size hash n_objects n_observations _hash_type _accessor visibility _key_is_virtual storage_id transform_id run_id created_at created_by_id
id
1 wJem73xED8zcmjFV0000 None True My image None .jpg None 29358 r4tnqmKI_SjrkdLzpuWp4g None None md5 None 1 True 1 None None 2024-10-22 09:24:40.487582+00:00 1
2 v2MJDTEJ042Agknn0000 None True The iris collection None .parquet dataset 5629 5axHfO_2Pmlun2sJZHVuNg None None md5 DataFrame 1 True 1 None None 2024-10-22 09:24:45.652752+00:00 1
3 de6fDVsyNVop7WNs0000 None True My fastq None .fastq.gz None 20 hi7ZmAzz8sfMd3vIQr-57Q None None md5 None 1 True 1 None None 2024-10-22 09:24:45.661507+00:00 1

The filter selects all artifacts based on the users who ran the generating notebook.

Under the hood, in the SQL database, it’s joining the artifact table with the run and the user table.

Comparators

You can qualify the type of comparison in a query by using a comparator.

Below follows a list of the most import, but Django supports about two dozen field comparators field__comparator=value.

and

ln.Artifact.filter(suffix=".jpg", created_by=user).df()
Hide code cell output
uid version is_latest description key suffix type size hash n_objects n_observations _hash_type _accessor visibility _key_is_virtual storage_id transform_id run_id created_at created_by_id
id
1 wJem73xED8zcmjFV0000 None True My image None .jpg None 29358 r4tnqmKI_SjrkdLzpuWp4g None None md5 None 1 True 1 None None 2024-10-22 09:24:40.487582+00:00 1

less than/ greater than

Or subset to artifacts smaller than 10kB. Here, we can’t use keyword arguments, but need an explicit where statement.

ln.Artifact.filter(created_by=user, size__lt=1e4).df()
Hide code cell output
uid version is_latest description key suffix type size hash n_objects n_observations _hash_type _accessor visibility _key_is_virtual storage_id transform_id run_id created_at created_by_id
id
2 v2MJDTEJ042Agknn0000 None True The iris collection None .parquet dataset 5629 5axHfO_2Pmlun2sJZHVuNg None None md5 DataFrame 1 True 1 None None 2024-10-22 09:24:45.652752+00:00 1
3 de6fDVsyNVop7WNs0000 None True My fastq None .fastq.gz None 20 hi7ZmAzz8sfMd3vIQr-57Q None None md5 None 1 True 1 None None 2024-10-22 09:24:45.661507+00:00 1

in

ln.Artifact.filter(suffix__in=[".jpg", ".fastq.gz"]).df()
Hide code cell output
uid version is_latest description key suffix type size hash n_objects n_observations _hash_type _accessor visibility _key_is_virtual storage_id transform_id run_id created_at created_by_id
id
1 wJem73xED8zcmjFV0000 None True My image None .jpg None 29358 r4tnqmKI_SjrkdLzpuWp4g None None md5 None 1 True 1 None None 2024-10-22 09:24:40.487582+00:00 1
3 de6fDVsyNVop7WNs0000 None True My fastq None .fastq.gz None 20 hi7ZmAzz8sfMd3vIQr-57Q None None md5 None 1 True 1 None None 2024-10-22 09:24:45.661507+00:00 1

order by

ln.Artifact.filter().order_by("-updated_at").df()
Hide code cell output
uid version is_latest description key suffix type size hash n_objects n_observations _hash_type _accessor visibility _key_is_virtual storage_id transform_id run_id created_at created_by_id
id
3 de6fDVsyNVop7WNs0000 None True My fastq None .fastq.gz None 20 hi7ZmAzz8sfMd3vIQr-57Q None None md5 None 1 True 1 None None 2024-10-22 09:24:45.661507+00:00 1
2 v2MJDTEJ042Agknn0000 None True The iris collection None .parquet dataset 5629 5axHfO_2Pmlun2sJZHVuNg None None md5 DataFrame 1 True 1 None None 2024-10-22 09:24:45.652752+00:00 1
1 wJem73xED8zcmjFV0000 None True My image None .jpg None 29358 r4tnqmKI_SjrkdLzpuWp4g None None md5 None 1 True 1 None None 2024-10-22 09:24:40.487582+00:00 1

contains

ln.Transform.filter(name__contains="search").df().head(5)
Hide code cell output
uid version is_latest name key description type source_code hash reference reference_type _source_code_artifact_id created_at created_by_id
id
16 OAOTJQryiilB0000 None True Research IgE IgG3. None None notebook None None None None None 2024-10-22 09:24:54.885132+00:00 1
30 rFcG9Ubv3jFy0000 None True Cluster intestinal candidate IgG research inte... None None notebook None None None None None 2024-10-22 09:24:54.886469+00:00 1
31 9jjmpbGBDvKp0000 None True Earlobe IgG IgG1 Adrenergic neural cells resea... None None notebook None None None None None 2024-10-22 09:24:54.886563+00:00 1
51 387oRaZfZrBT0000 None True Efficiency IgG3 research Adrenal glands. None None notebook None None None None None 2024-10-22 09:24:54.888437+00:00 1
59 wNgcUvlf3uJF0000 None True Igg4 IgG3 research IgA. None None notebook None None None None None 2024-10-22 09:24:54.889186+00:00 1

And case-insensitive:

ln.Transform.filter(name__icontains="Search").df().head(5)
Hide code cell output
uid version is_latest name key description type source_code hash reference reference_type _source_code_artifact_id created_at created_by_id
id
16 OAOTJQryiilB0000 None True Research IgE IgG3. None None notebook None None None None None 2024-10-22 09:24:54.885132+00:00 1
30 rFcG9Ubv3jFy0000 None True Cluster intestinal candidate IgG research inte... None None notebook None None None None None 2024-10-22 09:24:54.886469+00:00 1
31 9jjmpbGBDvKp0000 None True Earlobe IgG IgG1 Adrenergic neural cells resea... None None notebook None None None None None 2024-10-22 09:24:54.886563+00:00 1
51 387oRaZfZrBT0000 None True Efficiency IgG3 research Adrenal glands. None None notebook None None None None None 2024-10-22 09:24:54.888437+00:00 1
59 wNgcUvlf3uJF0000 None True Igg4 IgG3 research IgA. None None notebook None None None None None 2024-10-22 09:24:54.889186+00:00 1

startswith

ln.Transform.filter(name__startswith="Research").df()
Hide code cell output
uid version is_latest name key description type source_code hash reference reference_type _source_code_artifact_id created_at created_by_id
id
16 OAOTJQryiilB0000 None True Research IgE IgG3. None None notebook None None None None None 2024-10-22 09:24:54.885132+00:00 1
122 U4E0PVsoSdMu0000 None True Research IgE Mast cell Cortical Adrenergic neu... None None notebook None None None None None 2024-10-22 09:24:54.899153+00:00 1
157 UTqsHovY9VRp0000 None True Research Thymus efficiency IgA Speed cells IgG1. None None notebook None None None None None 2024-10-22 09:24:54.906393+00:00 1
326 dZjp7HyZtfNG0000 None True Research Earlobe intestinal Earlobe IgG1 IgG1 ... None None notebook None None None None None 2024-10-22 09:24:54.929127+00:00 1
327 hJe19erQghXL0000 None True Research Speed cells IgG study result. None None notebook None None None None None 2024-10-22 09:24:54.929219+00:00 1
334 gv3xyagHaSWY0000 None True Research Mast cell efficiency candidate study. None None notebook None None None None None 2024-10-22 09:24:54.933454+00:00 1
348 2wOookNWxY4k0000 None True Research IgG4 IgE intestinal Ciliary body. None None notebook None None None None None 2024-10-22 09:24:54.934755+00:00 1
388 fLW998dUk6Ss0000 None True Research IgE Adrenergic neural cells. None None notebook None None None None None 2024-10-22 09:24:54.938437+00:00 1
480 3ldrS7uSciJn0000 None True Research candidate Ciliary body investigate Sp... None None notebook None None None None None 2024-10-22 09:24:54.954324+00:00 1

or

ln.Artifact.filter(ln.Q(suffix=".jpg") | ln.Q(suffix=".fastq.gz")).df()
Hide code cell output
uid version is_latest description key suffix type size hash n_objects n_observations _hash_type _accessor visibility _key_is_virtual storage_id transform_id run_id created_at created_by_id
id
1 wJem73xED8zcmjFV0000 None True My image None .jpg None 29358 r4tnqmKI_SjrkdLzpuWp4g None None md5 None 1 True 1 None None 2024-10-22 09:24:40.487582+00:00 1
3 de6fDVsyNVop7WNs0000 None True My fastq None .fastq.gz None 20 hi7ZmAzz8sfMd3vIQr-57Q None None md5 None 1 True 1 None None 2024-10-22 09:24:45.661507+00:00 1

negate/ unequal

ln.Artifact.filter(~ln.Q(suffix=".jpg")).df()
Hide code cell output
uid version is_latest description key suffix type size hash n_objects n_observations _hash_type _accessor visibility _key_is_virtual storage_id transform_id run_id created_at created_by_id
id
2 v2MJDTEJ042Agknn0000 None True The iris collection None .parquet dataset 5629 5axHfO_2Pmlun2sJZHVuNg None None md5 DataFrame 1 True 1 None None 2024-10-22 09:24:45.652752+00:00 1
3 de6fDVsyNVop7WNs0000 None True My fastq None .fastq.gz None 20 hi7ZmAzz8sfMd3vIQr-57Q None None md5 None 1 True 1 None None 2024-10-22 09:24:45.661507+00:00 1

Clean up the test instance.

!rm -r ./test-registries
!lamin delete --force test-registries
Hide code cell output 
• deleting instance testuser1/test-registries