Metrics

Define and track training metrics.

See Also

• Monitoring Overview - All monitoring tools
• Reporters - Output metrics
• Callbacks - Training callbacks

---

datarax.monitoring.metrics

Metrics collection and tracking utilities for Datarax.

This module provides classes for collecting, tracking, and aggregating metrics during pipeline operation. These utilities enable monitoring of pipeline performance, throughput, and resource usage.

logger module-attribute

logger = getLogger(__name__)

MetricRecord dataclass

MetricRecord(name: str, value: float, timestamp: float, component: str, metadata: dict[str, Any] | None = None)

A single metric measurement.

Attributes:

Name	Type	Description
name	str	Name of the metric.
value	float	Measured value.
timestamp	float	Time when the metric was recorded.
component	str	Component that generated the metric.
metadata	dict[str, Any] | None	Optional additional metadata about the metric.

name instance-attribute

name: str

value instance-attribute

value: float

timestamp instance-attribute

timestamp: float

component instance-attribute

component: str

metadata class-attribute instance-attribute

metadata: dict[str, Any] | None = None

MetricsCollector

MetricsCollector(enabled: bool = True)

Collects and aggregates metrics during pipeline operations.

This class provides methods for tracking time-based metrics and recording arbitrary metrics during pipeline execution.

Attributes:

Name	Type	Description
enabled		Whether metrics collection is enabled.

Parameters:

Name	Type	Description	Default
enabled	bool	Whether metrics collection is enabled.	True

enabled instance-attribute

enabled = enabled

start_timer

start_timer(name: str, component: str = 'pipeline') -> None

Start timing an operation.

Parameters:

Name	Type	Description	Default
name	str	Name of the timer.	required
component	str	Component being timed.	'pipeline'

stop_timer

stop_timer(name: str, component: str = 'pipeline', metadata: dict[str, Any] | None = None) -> None

Stop timing and record elapsed time.

Parameters:

Name	Type	Description	Default
name	str	Name of the timer.	required
component	str	Component being timed.	'pipeline'
metadata	dict[str, Any] | None	Optional additional metadata about the timing.	None

record_metric

record_metric(name: str, value: float, component: str = 'pipeline', metadata: dict[str, Any] | None = None) -> None

Record a metric value.

Parameters:

Name	Type	Description	Default
name	str	Name of the metric.	required
value	float	Value to record.	required
component	str	Component generating the metric.	'pipeline'
metadata	dict[str, Any] | None	Optional additional metadata about the metric.	None

get_metrics

get_metrics() -> list[MetricRecord]

Get all collected metrics.

Returns:

Type	Description
list[MetricRecord]	List of collected metric records.

clear

clear() -> None

Clear all metrics and timers.

AggregatedMetrics

AggregatedMetrics()

Aggregates metrics over time.

This class provides methods for calculating aggregate statistics over collected metrics.

add_metrics

add_metrics(metrics: list[MetricRecord]) -> None

Add metrics to the aggregator.

Parameters:

Name	Type	Description	Default
metrics	list[MetricRecord]	List of metric records to add.	required

get_average

get_average(name: str, component: str = 'pipeline') -> float | None

Get the average value of a metric.

Parameters:

Name	Type	Description	Default
name	str	Name of the metric.	required
component	str	Component that generated the metric.	'pipeline'

Returns:

Type	Description
float | None	Average value, or None if no metrics were collected.

get_min

get_min(name: str, component: str = 'pipeline') -> float | None

Get the minimum value of a metric.

Parameters:

Name	Type	Description	Default
name	str	Name of the metric.	required
component	str	Component that generated the metric.	'pipeline'

Returns:

Type	Description
float | None	Minimum value, or None if no metrics were collected.

get_max

get_max(name: str, component: str = 'pipeline') -> float | None

Get the maximum value of a metric.

Parameters:

Name	Type	Description	Default
name	str	Name of the metric.	required
component	str	Component that generated the metric.	'pipeline'

Returns:

Type	Description
float | None	Maximum value, or None if no metrics were collected.

get_sum

get_sum(name: str, component: str = 'pipeline') -> float | None

Get the sum of values for a metric.

Parameters:

Name	Type	Description	Default
name	str	Name of the metric.	required
component	str	Component that generated the metric.	'pipeline'

Returns:

Type	Description
float | None	Sum of values, or None if no metrics were collected.

get_count

get_count(name: str, component: str = 'pipeline') -> int

Get the number of measurements for a metric.

Parameters:

Name	Type	Description	Default
name	str	Name of the metric.	required
component	str	Component that generated the metric.	'pipeline'

Returns:

Type	Description
int	Number of measurements.

get_rate

get_rate(name: str, component: str = 'pipeline') -> float | None

Get the rate of a metric (count / time period).

Parameters:

Name	Type	Description	Default
name	str	Name of the metric.	required
component	str	Component that generated the metric.	'pipeline'

Returns:

Type	Description
float | None	Rate (occurrences per second), or None if less than two measurements
float | None	exist.

clear

clear() -> None

Clear all aggregated metrics.