App version: 3.4.11

System namespace

When you create a Neptune run or experiment, the system namespace (sys) is automatically created in the Run object. It contains basic metadata about the run, lineage, and environment.

This section contains a complete reference of namespaces and attributes in the sys namespace.

`diagnostics` namespace

The diagnostics namespace helps you identify large runs and monitor growth in metadata.

note

This namespace is available only for runs created after the 3.4.11 release.

`tracker_client`

Attribute path	Type
`sys/diagnostics/tracker_client`	`String`

The version of the Neptune Python client library that was installed when the run was created.

`attributes` namespace

The attributes namespace indicates the number of attributes per type as well as the total number of attributes.

Attribute types

`file_ref_count`

Attribute path	Type
`sys/diagnostics/attributes/file_ref_count`	`Integer`

The total number of File attributes.

`float_count`

Attribute path	Type
`sys/diagnostics/attributes/float_count`	`Integer`

The total number of Float attributes.

`float_series_count`

Attribute path	Type
`sys/diagnostics/attributes/float_series_count`	`Integer`

The total number of FloatSeries attributes.

`int_count`

Attribute path	Type
`sys/diagnostics/attributes/int_count`	`Integer`

The total number of Integer attributes.

`string_count`

Attribute path	Type
`sys/diagnostics/attributes/string_count`	`Integer`

The total number of String attributes.

`string_series_count`

Attribute path	Type
`sys/diagnostics/attributes/string_series_count`	`Integer`

The total number of StringSeries attributes.

`string_set_count`

Attribute path	Type
`sys/diagnostics/attributes/string_set_count`	`Integer`

The total number of StringSet attributes.

`total_count`

Attribute path	Type
`sys/diagnostics/attributes/total_count`	`Integer`

The total number of attributes in the run.

`total_series_datapoints`

Attribute path	Type
`sys/diagnostics/attributes/total_series_datapoints`	`Integer`

The combined number of data points logged to all of the run's series attributes. Each numerical, string, or file value in a series counts as a data point.

The following aren't counted:

Preview points
Points from before the fork step, if the run was forked

`experiment` namespace

`is_head`

Attribute path	Type
`sys/experiment/is_head`	`Boolean`

If True: The run is the experiment's most recently updated run.

The experiment is defined by the name attribute.
If False: The run is no longer the experiment head.

`name`

Attribute path	Type
`sys/experiment/name`	`String`

The experiment that the run is part of.

To make the name easy to read in the app, ensure that it's at most 190 characters long.

`running_time_seconds`

Attribute path	Type
`sys/experiment/running_time_seconds`	`Float`

Total running time of the constituent run objects in seconds. Running time counts as activity that involves logging metrics.

This number takes into account any additional resumes of runs.

See also sys/running_time_seconds for individual runs.

`forking` namespace

The forking namespace contains information about the fork point and parent run.

Fork an experiment

`depth`

Attribute path	Type
`sys/forking/depth`	`Integer`

Levels of descent from the experiment's root run. For example:

Depth value	Description
`0`	The root run
`1`	Child run of the root run
`2`	Run that forks off the child run

`parent`

Attribute path	Type
`sys/forking/parent`	`String`

Identifier of the parent run. That is, the run that the current run was forked from.

`requested_parent`

This attribute can be used for troubleshooting purposes: If there's a requested parent but no actual parent run logged, there may be an issue in the training process, such as a race condition or unhandled error.

`step`

Attribute path	Type
`sys/forking/step`	`Integer`

Step where the run forks off its parent run.

`creation_time`

Attribute path	Type
`sys/creation_time`	`Datetime`

Creation time of the run. The value is determined by the creation_time argument of the Run constructor.

`custom_run_id`

Attribute path	Type
`sys/custom_run_id`	`String`

Custom identifier of the run. The value is determined by the run_id argument of the Run constructor.

`description`

Attribute path	Type
`sys/description`	`String`

When you add a description via the web app or API, it's stored as a string in the description attribute.

To learn more, see Log metadata: Run description.

`failed`

This attribute is experimental and can be ignored.

`family`

This attribute can be ignored.

`group_tags`

Attribute path	Type
`sys/group_tags`	`StringSet`

When you apply tags with add_tags(group_tags=True), the tags are stored under the attribute sys/group_tags.

To learn more, see Groups and Manage tags.

`id`

Attribute path	Type
`sys/id`	`String`

The auto-generated Neptune identifier, which consists of the project key and a counter.

`modification_time`

Attribute path	Type
`sys/modification_time`	`Datetime`

When the run was last modified. This also takes into account actions via the web interface, such as managing tags.

`monitoring_time`

This attribute is deprecated and can be ignored.

`name`

Attribute path	Type
`sys/name`	`String`

Name of the run. Same as the experiment name.

To make the name easy to read in the app, ensure that it's at most 190 characters long.

`owner`

Attribute path	Type
`sys/owner`	`String`

The account which created the run.

`ping_time`

Attribute path	Type
`sys/ping_time`	`Datetime`

When the Python client last interacted with the run.

`relative_creation_time_ms`

Timestamp of the last logged metric. When forking experiments, Neptune uses this timestamp as the indicator of when the parent run ended.

`running_time_seconds`

Attribute path	Type
`sys/running_time_seconds`	`Float`

Total running time of the run object in seconds. Running time counts as activity that involves logging metrics.

This number takes into account any additional resumes of the run.

See also sys/experiment/running_time_seconds for experiments.

`size`

This attribute is experimental and can be ignored.

`state`

Attribute path	Type
`sys/state`	`ObjectState`

A run object can be in one of two states: active or inactive.

active means that at least one logging process is connected to the run.
The run state automatically changes to inactive once there's been no activity for 20 seconds, typically after the script ended or you called close().

Resuming an inactive run doesn't change its state to active if no metadata is added or modified.

`tags`

Attribute path	Type
`sys/tags`	`StringSet`

Tags applied to the run, either via API or the web app.

To learn more, see Manage tags.

`trashed`

This attribute is experimental and can be ignored.

diagnostics namespace​

tracker_client​

attributes namespace​

file_ref_count​

float_count​

float_series_count​

int_count​

string_count​

string_series_count​

string_set_count​

total_count​

total_series_datapoints​

experiment namespace​

is_head​

name​

running_time_seconds​

forking namespace​

depth​

parent​

requested_parent​

step​

creation_time​

custom_run_id​

description​

failed​

family​

group_tags​

id​

modification_time​

monitoring_time​

name​

owner​

ping_time​

relative_creation_time_ms​

running_time_seconds​

size​

state​

tags​

trashed​

`diagnostics` namespace

`tracker_client`

`attributes` namespace

`file_ref_count`

`float_count`

`float_series_count`

`int_count`

`string_count`

`string_series_count`

`string_set_count`

`total_count`

`total_series_datapoints`

`experiment` namespace

`is_head`

`name`

`running_time_seconds`

`forking` namespace

`depth`

`parent`

`requested_parent`

`step`

`creation_time`

`custom_run_id`

`description`

`failed`

`family`

`group_tags`

`id`

`modification_time`

`monitoring_time`

`name`

`owner`

`ping_time`

`relative_creation_time_ms`

`running_time_seconds`

`size`

`state`

`tags`

`trashed`