quantlab.tracking.base.TrackingRun

class quantlab.tracking.base.TrackingRun

Bases: ABC

One record a tracker keeps: a training, a CV fold, a CV summary or a backtest.

Opened by Tracker.start_run and finished when its block is left. The public methods normalise their arguments and hand them to one hook each, which an adapter implements.

Examples

>>> with NullTracker().start_run(project="P", group=None, name="n", config={}) as run:
...     run.log({"train_loss": 0.5}, step=0)
...     run.summarize({"test": {"ic": 0.04}})

log(metrics, step)

Record values against a step, drawn as curves.

Parameters:

metricsMapping[str, float]

Metric names to numbers; each is converted to float.

stepint

The epoch or boosting round the values belong to.

Examples

>>> run = NullRun()
>>> run.log({"train_loss": 0.5, "val_loss": 0.7}, step=3)

log_file(path)

Attach a file to the run, such as a backtest’s HTML report.

Parameters:

pathstr or Path

The file to attach.

Raises:

FileNotFoundError

If path is not an existing file.

Examples

>>> import tempfile
>>> report = Path(tempfile.mkdtemp()) / "notes.html"
>>> _ = report.write_text("<p>report</p>")
>>> NullRun().log_file(report)

log_table(name, columns, rows, *, top_bars=None)

Record a table.

Parameters:

namestr

The table’s name in the run.

columnsSequence[str]

Column names.

rowsSequence[Sequence]

One sequence of cells per row, as many cells as columns.

top_barsint, optional

Also draw a bar chart of the first top_bars rows, labelled by the first column and sized by the second, where the tracker can draw one cheaply; a tracker that cannot ignores it.

Raises:

ValueError

If a row’s length differs from the number of columns, or top_bars is given for fewer than two columns.

Examples

>>> run = NullRun()
>>> run.log_table("importance/gain", ["factor", "importance"], [["mom", 0.6], ["rev", 0.4]])

summarize(metrics)

Set the run’s final values, one per name.

Nested mappings are flattened to outer/inner keys and every non-finite or non-numeric leaf is dropped (flatten_metrics). Writing a name again replaces its value.

Parameters:

metricsMapping

Metric names to numbers or to further mappings.

Examples

>>> run = NullRun()
>>> run.summarize({"whole": {"sharpe": 1.2}, "test_ic": 0.04})

update_config(params)

Add or change config values resolved after the run opened.

Parameters:

paramsMapping[str, Any]

Parameter names to values, made JSON-safe before they are sent.

Examples

>>> run = NullRun()
>>> run.update_config({"n_estimators": 412})