System namespace (sys)

The system namespace contains the ID, name, description, and other basic metadata.

The sys namespace (sys) contains auto-logged sys information and basic metadata on the run object. This includes the ID, name, timestamps, and status information.

You cannot create new fields in the sys namespace, but some of their values can be edited manually (marked with * in the table below).

• Editable field

Identifier (ID)

EachRunobject has a unique string identifier within the project.

The project key is always the first part of a run ID.

Example: In a project called "Classification" with project key CLS, the third run will have the identifier CLS-3.

You can access the ID of your active run programmatically with the fetch() method:

>>> run = neptune.init_run(project="ml-team/classification")
>>> run["sys/id"].fetch()
CLS-23

You can use the ID to resume logging later, access the metadata in read-only mode, or trash the run.

Custom run identifier

The custom run identifier, if one was supplied to thecustom_run_idargument when the run was created.

import hashlib
import time

run = neptune.init_run(
 custom_run_id=hashlib.md5(str(time.time()).encode()).hexdigest(),
)

Note that it's not possible to use a custom run ID in offline mode.

Related

Set custom run ID

Name

Editable name of a run.

It doesn't need to be unique, but you can use it as a human-friendly ID.

You can set it in the following ways:

• During initialization, pass the name to the name parameter of the init_run() function: run = neptune.init_run(name="mellow-panda"). If you don't set a name at initialization, once the run is synchronized with the server, Neptune sets the auto-generated identifier (sys/id) as the name.
• Any time the run is active in your code, assign a string to the sys/name field:
  ```python

  run = neptune.init_run(with_id=...)
  run["sys/name"] = "sleepy-swallow"
  ```

• In the app, hover over the run, then click → Run information. You can edit the name in the modal that opens.

During initialization, pass the name to thenameparameter of theinit_run()function:

run = neptune.init_run(name="mellow-panda")

If you don't set a name at initialization, once the run is synchronized with the server, Neptune sets the auto-generated identifier (sys/id) as the name.

Any time the run is active in your code, assign a string to thesys/namefield:

>>> run = neptune.init_run(with_id=...)
>>> run["sys/name"] = "sleepy-swallow"

In the app, hover over the run, then click→Run information. You can edit the name in the modal that opens.

Related

Set custom run name

Description

Editable description of a run.

The description field can store a longer description or free-form note about the run.

You can add a description with thedescriptionparameter ofinit_run():

run = neptune.init_run(description="My longer description of the run")

You can also add or edit descriptions through the web app:

• Click the run menu ( ) next to the run ID → Run information .
• In experiments table, add the sys/description field as a column. You can then edit the contents of the cell.

Running time

Total running time of the object in seconds.

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

Ping time

Time of the last interaction with the Neptune client library. That is, when something was last logged by the API.

Includes date, time of day, and information about the time zone.

Owner

Neptune username of the user or service account that created the run.

Creation time

Creation time of the run.

Includes date, time of day, and information about the time zone.

Modification time

Time that the run was last modified. This could be any kind of edit, such as setting a description or removing a tag through the app.

Includes date, time of day, and information about the time zone.

Tags

The tags assigned to the run.

You can assign tags in the following ways:

• During initialization, pass them as as list to thetagsargument of theinit_run()function: run=neptune.init_run(tags=["maskRCNN","finetune"])
• After creating the run, with theadd()method:run["sys/tags"].add("best")
• Through the Neptune app, in the Experiments tab

During initialization, pass them as as list to thetagsargument of theinit_run()function:

run = neptune.init_run(tags=["maskRCNN", "finetune"])

After creating the run, with theadd()method:run["sys/tags"].add("best")

For more, seeAdd tags.

Group tags

Group tags assigned to the run.

You can assign group tags in the following ways:

• After creating the run, with the add() method: run["sys/group_tags"].add("high lr")
• Through the Neptune app, in the Experiments tab

For more, seeGroups.

State

See also

sys/failed

A run object can be in one of two states:

• Active
• Inactive

Activemeans that at least one process is connected to the run. This may be a process that is logging training metrics or monitoring performance.

The run automatically transitions to anInactivestate once there's been no activity for 20 seconds, typically after the script ended or you invokedstop().

Connecting to an inactive run inread-only modedoesn't change the state to active.

Info

"Active/inactive" doesn't correspond to "success/failure". The connection to a run can be paused and resumed multiple times, so Neptune doesn't know when the logging was a success.

Example:An inactive state can mean that the run didn't start, that the training or monitoring was paused, or that it did in fact finish training with a success.

When fetching the experiments table programmatically, you can filter the returned table by passing values ofsys/state`to thestateargument of thefetch_runs_table()`method:

import neptune

project = neptune.init_project(project="workspace-name/project-name")
runs_table_df = project.fetch_runs_table(state="active").to_pandas()

Customization tip

To make the states match your workflow better, you can set a custom status for each run. For example:

• run["info/state"]="Success"
• run["info/state"]="Queued"

You can also defineBooleanfields:

• run["info/success"]=True
• run["info/queued"]=False

Size

Size of the run object in bytes.

>>> project = neptune.init_project(project="ml-team/nlu", mode="read-only")
>>> large_runs_df = project.fetch_runs_table(sort_by="sys/size").to_pandas()

Failed

See also

sys/state

Whether the run failed. You can set it manually to mark a run as failed, or reset it toFalsein case of resuming a failed run.

If there is a crash during monitoring, Neptune automatically sets theFailedstatus toTrue. You can override this behavior by setting thefail_on_exceptionparameter ininit_run()toFalse.

In both cases, the traceback is captured and appended to the"monitoring/traceback"field. (If you provided a custom monitoring namespace forinit_run(), the"traceback"field goes under that custom namespace.)

If you performed a remote abort, theFailedstatus will also be set toTrue. To learn more, seeStop or abort a run remotely.

This page is originally sourced from the legacy docs.