Project#

class skore.Project(name, *, mode='local', **kwargs)[source]#

API to manage a collection of key-report pairs.

Its constructor initializes a project by creating a new project or by loading an existing one.

The class main methods are put(), summarize(), get(), and sync(), respectively to insert a key-report pair into the project, obtain the metadata/metrics of the inserted reports, get a specific report by its id, and synchronize reports between projects.

Three mutually exclusive modes are available and can be configured using the mode parameter of the constructor:

Hub mode

The project is configured to communicate with skore hub.

In this mode, workspace is a skore hub concept that must be configured on the skore hub interface. It represents an isolated entity managing users, projects, and resources. It can be a company, organization, or team that operates independently within the system.

Note: Using Project in hub mode requires an account on skore hub, with access rights to the specified workspace. Authentication to skore hub is done by running skore.login() before instantiating the Project.

Local mode

Otherwise, the project is configured to the local mode to be persisted on the user machine in a directory called a workspace.

The workspace can be set using kwargs or the environment variable SKORE_WORKSPACE.
If not, it will be set to a default location:

  • we search the current working directory and its parents for a skore workspace: a directory named ‘skore’ containing a file named ‘.SKORE_WORKSPACE’. If found, use that.

  • otherwise if we are in a Git repository, create ‘skore’ at the root

    of the repository.

  • otherwise create ‘skore’ in the current working directory.

MLflow mode

In this mode, name is used as the MLflow experiment name. Reports are persisted as MLflow model artifacts in runs created under this experiment.

Refer to the Storing data science artifacts section of the user guide for more details.

Parameters:
namestr

The name of the project.

mode{“hub”, “local”, “mlflow”}

The mode of the project.

**kwargsdict

Extra keyword arguments passed to the project, depending on its mode.

workspace : str or Path-like, optional

  • If mode="hub", the Hub workspace name (required);

  • If mode="local", the local persistence directory (optional);

  • If mode="mlflow", ignored.

tracking_uristr, mode:mlflow only.

The URI of the MLflow tracking server.

Attributes:
namestr

The name of the project.

mode{“hub”, “local”, “mlflow”}

The mode of the project.

workspacePath or str or None

The workspace for local and hub modes; None otherwise.

tracking_uristr or None

The MLflow tracking URI for mlflow mode; None otherwise.

ml_taskMLTask

The ML task of the project; unset until a first report is put.

See also

Summary

Tabular view of metadata and metrics for persisted reports.

compare()

Compare reports side by side.

Project.summarize()

Create a summary view to investigate persisted reports’ metadata/metrics.

Examples

Construct reports.

>>> from sklearn.datasets import make_regression
>>> from sklearn.linear_model import LinearRegression
>>> from skore import evaluate
>>>
>>> X, y = make_regression(random_state=42)
>>> regressor = LinearRegression()
>>> regressor_report = evaluate(regressor, X, y, splitter=0.2)

Construct the project in local mode, persisted in a temporary directory.

>>> from pathlib import Path
>>> from tempfile import TemporaryDirectory
>>> from skore import Project
>>>
>>> tmpdir = TemporaryDirectory().name
>>> local_project = Project(name="my-xp", mode="local", workspace=Path(tmpdir))

Put reports in the project.

>>> local_project.put("my-simple-regression", regressor_report)

Investigate metadata/metrics to filter the best reports.

>>> summary = local_project.summarize()
>>> summary = summary.query("rmse < 67")
>>> reports = summary.compare()
static delete(name, *, mode='local', **kwargs)[source]#

Delete a project.

Parameters:
namestr

The name of the project.

mode{“hub”, “local”, “mlflow”}, default “local”

The mode of the project.

**kwargsdict

Extra keyword arguments passed to the project, depending on its mode.

workspacestr or Path-like, optional

See the Project class docstring for details.

tracking_uristr, mode:mlflow only.

The URI of the MLflow tracking server.

get(id)[source]#

Get a persisted report by its id.

Report IDs can be found via skore.Project.summarize(), which is also the preferred method of interacting with a skore.Project. The id passed here must match the id column returned by Project.summarize().

Parameters:
idstr

The id of a report already put in the project.

Returns:
reportEstimatorReport or CrossValidationReport

The report associated with id.

Examples

>>> from sklearn.datasets import make_regression
>>> from sklearn.linear_model import LinearRegression
>>> from pathlib import Path
>>> from tempfile import TemporaryDirectory
>>> from skore import Project, evaluate
>>> X, y = make_regression(random_state=42)
>>> report = evaluate(LinearRegression(), X, y, splitter=0.2)
>>> tmpdir = TemporaryDirectory()
>>> project = Project(name="my-xp", mode="local", workspace=Path(tmpdir.name))
>>> project.put("my-regression", report)
>>> summary = project.summarize()
>>> report_id = summary.frame().index.get_level_values("id")[0]
>>> retrieved = project.get(report_id)
>>> type(retrieved).__name__
'EstimatorReport'
>>> tmpdir.cleanup()
property mode#

The mode of the project.

property name#

The name of the project.

put(key, report)[source]#

Put a key-report pair to the project.

If the key already exists, its last report is modified to point to this new report, while keeping track of the report history.

Parameters:
keystr

The key to associate with report in the project. Name of the run for mode:mlflow

reportEstimatorReport | CrossValidationReport

The report to associate with key in the project.

Returns:
None

The report is persisted in the project backend.

Examples

>>> from sklearn.datasets import make_regression
>>> from sklearn.linear_model import LinearRegression
>>> from pathlib import Path
>>> from tempfile import TemporaryDirectory
>>> from skore import Project, evaluate
>>> X, y = make_regression(random_state=42)
>>> report = evaluate(LinearRegression(), X, y, splitter=0.2)
>>> tmpdir = TemporaryDirectory()
>>> project = Project(name="my-xp", mode="local", workspace=Path(tmpdir.name))
>>> project.put("my-regression", report)
>>> tmpdir.cleanup()
summarize()[source]#

Obtain metadata/metrics for all persisted reports.

Reports are returned in ascending order of their date field.

Returns:
summarySummary

Metadata and metrics for every report persisted in the project.

See also

Summary

Tabular view with interactive filtering in Jupyter.

compare()

Compare selected reports side by side.

sync(other, *, bidirectional=False, dry_run=False, **kwargs)[source]#

Copy missing reports to another project.

Reports are matched using the report_id column returned by Project.summarize().frame(). Set bidirectional=True to copy missing reports in both directions.

Parameters:
otherProject or {“hub”, “local”, “mlflow”}

Destination project. When a mode is given, build the destination with this project’s name and the mode-specific keyword arguments.

bidirectionalbool, default=False

If False, transfer reports from this project to other. If True, also transfer reports missing from this project.

dry_runbool, default=False

Return the planned operations without loading or storing reports.

**kwargsdict

Mode-specific arguments used to build the destination when other is a mode string. For example, pass workspace for "hub" or tracking_uri for "mlflow".

Returns:
resultpandas.DataFrame

Synchronization status indexed by report_id. The direction column is "outbound" from this project to other, "inbound" from other to this project, or missing for skipped reports. The status column is "planned", "transferred", or "skipped".

property tracking_uri#

The MLflow tracking URI for mlflow mode; None otherwise.

property workspace#

The workspace for local and hub modes; None otherwise.