Creating a basic repository for your artifacts

In this tutorial, we will demonstrate how you can create a repository and log artifacts.

A repository is an organized structure that stores and versions your artifacts. It makes it easy to retrieve older versions of artifacts, log new ones or time travel.

import tempfile
import zoneinfo
from datetime import datetime
from pathlib import Path

import time_machine
from sklearn.datasets import make_classification
from sklearn.linear_model import LogisticRegression
from sklearn.svm import SVC

from lazyscribe import Repository

First, we’re going to train version 0 of our model that we want to deploy.

X, y = make_classification(n_samples=1000, n_features=10)
model = SVC(kernel="linear")
model.fit(X, y)

SVC(kernel='linear')

In a Jupyter environment, please rerun this cell to show the HTML representation or trust the notebook.
On GitHub, the HTML representation is unable to render, please try loading this page with nbviewer.org.

Next, we initialize our repository. In this tutorial we will use a temporary path, but you can use a local path or external filesystem.

tmpdir = Path(tempfile.mkdtemp())
repository = Repository(tmpdir / "repository.json", mode="w+")

Let’s log version 0 of our model to our repository alongside the list of input features. Remember to save! Nothing is actually written to file until you save.

Note

In this tutorial, you’ll see us use time-machine to specify datetimes. We are doing this because the tutorial code logs multiple versions of the same artifact within 1 second of each other, which can cause unexpected behaviour when calling lazyscribe.repository.Repository.save().

with time_machine.travel(
    datetime(2025, 10, 31, 17, 0, 0, tzinfo=zoneinfo.ZoneInfo("UTC")), tick=False
):
    repository.log_artifact("model", model, handler="pickle")
    repository.log_artifact("features", list(range(10)), handler="json", indent=4)
    repository.save()

Let’s simulate a deployment environment by initializing a read-only instance of lazyscribe.repository.Repository.

repository_read = Repository(tmpdir / "repository.json", mode="r")
repository_read.load_artifact("model")

SVC(kernel='linear')

In a Jupyter environment, please rerun this cell to show the HTML representation or trust the notebook.
On GitHub, the HTML representation is unable to render, please try loading this page with nbviewer.org.

Say we have trained a newer, better version of our model. We can log it as version 1 of our model artifact. Remember to save!

with time_machine.travel(
    datetime(2025, 11, 1, 17, 0, 0, tzinfo=zoneinfo.ZoneInfo("UTC")), tick=False
):
    modelv1 = LogisticRegression()
    modelv1.fit(X[:, 0:9], y)
    repository.log_artifact("model", modelv1, handler="pickle")
    repository.log_artifact("features", list(range(9)), handler="json", indent=4)
    repository.save()

lazyscribe.repository.Repository.load_artifact() will, by default, load the latest version of our model.

repository_read = Repository(tmpdir / "repository.json", mode="r")
repository_read.load_artifact("model")

LogisticRegression()

In a Jupyter environment, please rerun this cell to show the HTML representation or trust the notebook.
On GitHub, the HTML representation is unable to render, please try loading this page with nbviewer.org.

We can still time travel back to version 0 of our model artifact by specifying either the integer version, the datetime it was created, or a string version of the datetime.

repository_read.load_artifact("model", version=0)

SVC(kernel='linear')

In a Jupyter environment, please rerun this cell to show the HTML representation or trust the notebook.
On GitHub, the HTML representation is unable to render, please try loading this page with nbviewer.org.

We can also observe the differences between our features list using difflib. Here, we are comparing version 0 of the features artifact against the latest version.

print(repository_read.get_version_diff("features", version=0))

---

+++

@@ -7,6 +7,5 @@

     5,
     6,
     7,
-    8,
-    9
+    8
 ]