fetch_experiments_table()

Python package: neptune-query

Fetches a table of experiment metadata, with runs as rows and attributes as columns.

To narrow the results, define filters for experiments to search or attributes to include.

Related

• Fetch runs table

Parameters

project

str

optional

default: None

Path of the Neptune project, as WorkspaceName/ProjectName.

If not provided, the NEPTUNE_PROJECT environment variable is used.

experiments

str | list of str | Filter

optional

default: None

Filter specifying which experiments to include.

• If a string is provided, it's treated as a regex pattern that the experiment names must match.
• If a list of strings is provided, it's treated as exact experiment names to match.
• To provide a more complex condition on an arbitrary attribute value, pass a Filter object.

If no filter is specified, all experiments are returned.

attributes

str | list of str | AttributeFilter

optional

default: None

Filter specifying which attributes to include.

• If a string is provided, it's treated as a regex pattern that the attribute names must match.
• If a list of strings is provided, it's treated as exact attribute names to match.
• To provide a more complex condition, pass an AttributeFilter object.

sort_by

str | Attribute

optional

default: "sys/creation_time"

Name of the attribute to sort the table by.

Alternatively, an Attribute object that specifies the attribute type.

sort_direction

"asc" | "desc"

optional

default: "desc"

Sorting direction of the column specified by the sort_by parameter.

limit

int

optional

default: None

Maximum number of experiments to return. By default all experiments are returned.

type_suffix_in_column_names

bool

optional

default: False

If True, columns of the returned DataFrame are suffixed with :<type>. For example, "attribute1:float_series", "attribute1:string".

If set to False, the method throws an exception if there are multiple types under one path.

Returns

pandas.DataFrame – A DataFrame similar to the runs table in the web app.

The DataFrame has:

• a single-level index "experiment" with experiment names
• a single-level column index with attribute names

For series attributes, the last logged value is returned.

Raises

• AttributeTypeInferenceError – If the attribute type wasn't specified in a filter passed to the experiments argument, and the attribute has multiple types across the project's experiments.
• ConflictingAttributeTypes – If there are conflicting attribute types under the same path and the type_suffix_in_column_names argument is set to False.

Example

Fetch attributes matching loss or batch_size from four specific experiments:

import neptune_query as nq


nq.fetch_experiments_table(
    experiments=["seabird-1", "seabird-2", "seabird-3", "seabird-4"],
    attributes=r"loss | batch_size",
    type_suffix_in_column_names=True,
)

Sample output

            config/batch_size:float  config/batch_size:int  loss:float_series
experiment
seabird-4                      64.0                    NaN           0.181736
seabird-3                       NaN                   64.0           0.123372
seabird-2                       NaN                   32.0           0.224408
seabird-1                       NaN                   32.0           0.205908