AttributeFilter

Python package: neptune-fetcher

Filter to apply to attributes when fetching runs or experiments.

Use to select specific metrics or other metadata based on various criteria.

Related

• Construct filters with Fetcher API

Parameters

name_eq

str | list of str

required

default: None

An attribute name or list of names to match exactly. If None, this filter is not applied.

type_in

list[Literal["float", "int", "string", "bool", "datetime", "float_series", "string_set", "string_series"]]

required

default: all available types

A list of allowed attribute types.

name_matches_all

str | list of str

required

default: None

A regular expression or list of expressions that the attribute name must match. If None, this filter is not applied.

name_matches_none

str | list of str

required

default: None

A regular expression or list of expressions that the attribute names mustn't match. Attributes matching any of the regexes are excluded. If None, this filter is not applied.

aggregations

list[Literal["last", "min", "max", "average", "variance"]]

required

default: "last"

List of aggregation functions to apply when fetching metrics of type FloatSeries or StringSeries.

Examples

Import the needed classes:

import neptune_fetcher.alpha as npt
from neptune_fetcher.alpha.filters import AttributeFilter

Constructing a filter

Select metrics not matching regexes ^test or loss$ and pick the "average" and "variance" aggregations:

AttributeFilter(
    name_matches_none=[r"^test", r"loss$"],
    aggregations=["average", "variance"],
)

Combine multiple filters with the pipe character:

filter_1 = AttributeFilter(...)
filter_2 = AttributeFilter(...)
filter_3 = AttributeFilter(...)
alternatives = filter_1 | filter_2 | filter_3

Using a filter

To use an attribute filter, pass it to the attributes argument of a fetching or listing method:

npt.fetch_experiments_table(
    experiments=["daring-kittiwake_week-41", "discreet-kittiwake_week-42"],
    attributes=AttributeFilter(name_eq="configs/learning_rate"),
)

You can also pass a list of attribute names directly to the attributes argument. Example using the runs API:

from neptune_fetcher.alpha import runs


runs.fetch_runs_table(
    runs=["spurious-kittiwake_025c425", "spurious-kittiwake_x56jjh2"],
    attributes=["configs/optimizer", "configs/batch_size"],
)