neptune-fetcher 0.5.0rc0

Neptune Fetcher

Neptune Fetcher

[!NOTE] This package is experimental.

Neptune Fetcher is designed to separate data retrieval capabilities from the regular neptune package. This separation makes data fetching more efficient and improves performance.

Installation

pip install neptune-fetcher

Example usage

Listing runs of a project

from neptune_fetcher import ReadOnlyProject

project = ReadOnlyProject("workspace/project")

for run in project.list_runs():
    print(run)  # dicts with Neptune ID and custom ID

Listing experiments of a project

from neptune_fetcher import ReadOnlyProject

project = ReadOnlyProject("workspace/project")

for experiment in project.list_experiments():
    print(experiment)

Fetching runs data frame with specific columns

from neptune_fetcher import ReadOnlyProject

project = ReadOnlyProject("workspace/project")

runs_df = project.fetch_runs_df(
    columns=["sys/custom_run_id", "sys/modification_time"],
    columns_regex="tree/.*",  # added to columns specified with the "columns" parameter
)

Fetching data from specified runs

from neptune_fetcher import ReadOnlyProject

project = ReadOnlyProject("workspace/project")

for run in project.fetch_read_only_runs(with_ids=["RUN-1", "RUN-2"]):
    run.prefetch(["parameters/optimizer", "parameters/init_lr"])

    print(run["parameters/optimizer"].fetch())
    print(run["parameters/init_lr"].fetch())

Fetching data from a single run

from neptune_fetcher import ReadOnlyProject, ReadOnlyRun

project = ReadOnlyProject("workspace/project")
run = ReadOnlyRun(project, with_id="TES-1")

run.prefetch(["parameters/optimizer", "parameters/init_lr"])
run.prefetch_series_values(["metrics/loss", "metrics/accuracy"], use_threads=True)

print(run["parameters/optimizer"].fetch())
print(run["parameters/init_lr"].fetch())

API reference

ReadOnlyProject

Representation of a Neptune project in a limited read-only mode.

Initialization

Initialize with the ReadOnlyProject class constructor:

project = ReadOnlyProject("workspace/project", api_token="...")

[!TIP] Find your API token in your user menu, in the bottom-left corner of the Neptune app.

Parameters:

Name	Type	Default	Description
project	str, optional	None	Name of a project in the form workspace-name/project-name. If None, the value of the NEPTUNE_PROJECT environment variable is used.
api_token	str, optional	None	Your Neptune API token (or a service account's API token). If None, the value of the NEPTUNE_API_TOKEN environment variable is used. To keep your token secure, avoid placing it in source code. Instead, save it as an environment variable.
proxies	dict, optional	None	Dictionary of proxy settings, if needed. This argument is passed to HTTP calls made via the Requests library. For details on proxies, see the Requests documentation.

---

list_runs()

Lists all runs of a project.

Each run is identified by Neptune ID (sys/id) and, if set, custom ID (sys/custom_run_id).

Returns: Iterator of dictionaries with Neptune run identifiers and custom identifiers.

Example:

project = ReadOnlyProject()

for run in project.list_runs():
    print(run)

---

list_experiments()

Lists all experiments of a project.

Each experiment is identified by:

• Neptune ID: sys/id
• (If set) Custom ID: sys/custom_run_id
• Name: sys/name

Example:

for experiment in project.list_experiments():
    print(experiment)

Returns: Iterator of dictionaries with Neptune experiment identifiers, custom identifiers and names.

---

fetch_runs()

Fetches a table containing Neptune IDs and custom run IDs of runs in the project.

Returns: pandas.DataFrame with two columns (sys/id and sys/custom_run_id) and one row for each run.

Example:

project = ReadOnlyProject()
df = project.fetch_runs()

---

fetch_experiments()

Fetches a table containing Neptune IDs, custom IDs and names of experiments in the project.

Example:

df = project.fetch_experiments()

Returns: pandas.DataFrame with three columns (sys/id, sys/custom_run_id, sys/name) and one row for each experiment.

---

fetch_runs_df()

Fetches the runs' metadata and returns them as a pandas DataFrame.

Parameters:

Name	Type	Default	Description
columns	List[str], optional	None	Names of columns to include in the table, as a list of field names. The Neptune ID (sys/id) is included automatically. If None, all the columns of the experiments table are included. Note: When using one or both of the columns and columns_regex parameters, the total number of matched columns must not exceed 100.
columns_regex	str, optional	None	A regex pattern to filter columns by name. Use this parameter to include columns in addition to the ones specified by the columns parameter. Note: When using one or both of the columns and columns_regex parameters, the total number of matched columns must not exceed 100.
custom_id_regex	str, optional	None	A regex pattern to filter the runs by custom ID. When applied, the total number of matched runs must not exceed 100.
with_ids	List[str], optional	None	List of multiple Neptune IDs. Example: ["NLU-1", "NLU-2"]. Matching any element of the list is sufficient to pass the criterion.
custom_ids	List[str], optional	None	List of multiple custom IDs. Example: ["nostalgic_shockley", "high_albattani"]. Matching any element of the list is sufficient to pass the criterion.
states	List[str], optional	None	List of states. Possible values: "inactive", "active". "Active" means that at least one process is connected to the run. Matching any element of the list is sufficient to pass the criterion.
owners	List[str], optional	None	List of multiple owners. Example: ["frederic", "josh"]. The owner is the user who created the run. Matching any element of the list is sufficient to pass the criterion.
tags	List[str], optional	None	A list of tags. Example: "lightGBM" or ["pytorch", "cycleLR"]. Note: Only runs that have all specified tags will pass this criterion.
trashed	bool, optional	False	Whether to retrieve trashed runs. If True, only trashed runs are retrieved. If False, only non-trashed runs are retrieved. If None or left empty, all run objects are retrieved, including trashed ones.
limit	int, optional	None	Maximum number of runs to fetch. If None, all runs are fetched.
sort_by	str, optional	sys/creation_time	Name of the field to sort the results by. The field must represent a simple type (string, float, integer).
ascending	bool, optional	False	Whether to sort the entries in ascending order of the sorting column values.
progress_bar	bool, Type[ProgressBarCallback], optional	None	Set to False to disable the download progress bar, or pass a type of ProgressBarCallback to use your own progress bar. If set to None or True, the default tqdm-based progress bar will be used.

Returns: pandas.DataFrame: A pandas DataFrame containing metadata of the fetched runs.

[!IMPORTANT] When using a regular expression to filter runs or columns, the total number of matched entries must not exceed 100.

Specifically, you can fetch a data frame with a maximum of:

• 100 columns, when using columns or columns_regex to filter columns.
• 100 runs, when using custom_id_regex to filter runs.

Examples:

Fetch all runs, with specific columns:

project = ReadOnlyProject()

runs_df = project.fetch_runs_df(
    columns=["sys/custom_run_id", "sys/modification_time", "training/lr"]
)

Fetch all runs, with specific columns and extra columns that match a regex pattern:

runs_df = project.fetch_runs_df(
    columns=["sys/custom_run_id", "sys/modification_time"],
    columns_regex="tree/.*",
)

Fetch runs by specific ID:

specific_runs_df = my_project.fetch_runs_df(custom_ids=["nostalgic_shockley", "high_albattani"])

---

fetch_experiments_df()

Fetches the experiments' metadata and returns them as a pandas DataFrame.

Parameters:

Name	Type	Default	Description
columns	List[str], optional	None	Names of columns to include in the table, as a list of field names. The Neptune ID ("sys/id") is included automatically. If None, all the columns of the experiments table are included. Note: When using one or both of the columns and columns_regex parameters, the total number of matched columns must not exceed 100.
columns_regex	str, optional	None	A regex pattern to filter columns by name. Use this parameter to include columns in addition to the ones specified by the columns parameter. Note: When using one or both of the columns and columns_regex parameters, the total number of matched columns must not exceed 100.
names_regex	str, optional	None	A regex pattern to filter the experiments by name. When applied, it needs to limit the number of experiments to 100 or fewer.
custom_id_regex	str, optional	None	A regex pattern to filter the experiments by custom ID. When applied, it needs to limit the number of experiments to 100 or fewer.
with_ids	List[str], optional	None	List of multiple Neptune IDs. Example: ["NLU-1", "NLU-2"]. Matching any element of the list is sufficient to pass the criterion.
custom_ids	List[str], optional	None	List of multiple custom IDs. Example: ["nostalgic_shockley", "high_albattani"]. Matching any element of the list is sufficient to pass the criterion.
states	List[str], optional	None	List of states. Possible values: "inactive", "active". "Active" means that at least one process is connected to the experiment. Matching any element of the list is sufficient to pass the criterion.
owners	List[str], optional	None	List of multiple owners. Example: ["frederic", "josh"]. The owner is the user who created the experiement. Matching any element of the list is sufficient to pass the criterion.
tags	List[str], optional	None	A list of tags. Example: "lightGBM" or ["pytorch", "cycleLR"]. Note: Only experiments that have all specified tags will pass this criterion.
trashed	bool, optional	False	Whether to retrieve trashed experiments. If True, only trashed experiments are retrieved. If False, only non-trashed experiments are retrieved. If None or left empty, all experiment objects are retrieved, including trashed ones.
limit	int, optional	None	Maximum number of experiments to fetch. If None, all experiments are fetched.
sort_by	str, optional	sys/creation_time	Name of the field to sort the results by. The field must represent a simple type (string, float, integer).
ascending	bool, optional	False	Whether to sort the entries in ascending order of the sorting column values.
progress_bar	bool, Type[ProgressBarCallback], optional	None	Set to False to disable the download progress bar, or pass a type of ProgressBarCallback to use your own progress bar. If set to None or True, the default tqdm-based progress bar will be used.

Returns: pandas.DataFrame: A pandas DataFrame containing metadata of the fetched experiments.

[!IMPORTANT] When using a regular expression to filter experiments or columns, the total number of matched entries must not exceed 100.

Specifically, you can fetch a data frame with a maximum of:

• 100 columns, when using columns or columns_regex to filter columns.
• 100 experiments, when using names_regex or custom_id_regex to filter experiments.

Examples:

Fetch all experiments with specific columns:

experiments_df = project.fetch_experiments_df(
    columns=["sys/custom_run_id", "sys/modification_time", "training/lr"]
)

Fetch all experiments with specific columns and extra columns that match a regex pattern:

experiments_df = project.fetch_experiments_df(
    columns=["sys/custom_run_id", "sys/modification_time"],
    columns_regex="tree/.*",
)

Fetch experiments by specific IDs:

specific_experiments_df = my_project.fetch_experiments_df(
    custom_ids=["nostalgic_shockley", "high_albattani"]
)

---

fetch_read_only_runs()

List runs of the project in the form of ReadOnlyRun.

Parameters:

Name	Type	Default	Description
with_ids	Optional[List[str]]	None	List of Neptune run IDs to fetch.
custom_ids	Optional[List[str]]	None	List of custom run IDs to fetch.

Returns: Iterator of ReadOnlyRun objects.

Example:

project = ReadOnlyProject()

for run in project.fetch_read_only_runs(custom_ids=["nostalgic_shockley", "high_albattani"]):
    ...

---

ReadOnlyRun

Representation of a Neptune run in a limited read-only mode.

Initialization

Can be created

• with the class constructor:

  project = ReadOnlyProject()
  run = ReadOnlyRun(project, with_id="TES-1")
  

• or as a result of the fetch_read_only_runs() method of the ReadOnlyProject class:

  for run in project.fetch_read_only_runs(
      custom_ids=["nostalgic_shockley", "high_albattani"]):
      ...
  

Parameters:

Name	Type	Default	Description
read_only_project	ReadOnlyProject	-	Source project from which run will be fetched.
with_id	Optional[str]	None	ID of the Neptune run to fetch. Example: RUN-1. Exclusive with the custom_id parameter.
custom_id	Optional[str]	None	Custom ID of the Neptune run to fetch. Example: high_albattani. Exclusive with the with_id parameter.

Example:

from neptune_fetcher import ReadOnlyProject, ReadOnlyRun

project = ReadOnlyProject("workspace-name/project-name", api_token="...")
run = ReadOnlyRun(project, custom_id="high_albattani")

---

.field_names

List of run field names.

A field is the location where a piece of metadata is stored in the run.

Returns: Iterator of run fields as strings.

Example:

for run in project.fetch_read_only_runs(custom_ids=["nostalgic_shockley", ...]):
    print(list(run.field_names))

---

Field lookup: run[field_name]

Used to access a specific field of a run. See Available types.

Returns: An internal object used to operate on a specific field.

Example:

run = ReadOnlyRun(...)
custom_id = run["sys/custom_run_id"].fetch()

---

prefetch()

Pre-fetches a batch of fields to the internal cache.

Improves the performance of access to consecutive field values.

Supported Neptune field types:

• Boolean
• Datetime
• Float
• FloatSeries
• Integer
• ObjectState
• String
• StringSet

Parameters:

Name	Type	Default	Description
paths	List[str]	-	List of field paths to fetch to the cache.

Example:

run = ReadOnlyRun(...)
run.prefetch(["parameters/optimizer", "parameter/init_lr"])
# No more calls to the API
print(run["parameters/optimizer"].fetch())
print(run["parameter/init_lr"].fetch())

prefetch_series_values()

Prefetches a batch of series to the internal cache.

Improves the performance of access to consecutive field values. Works only for series (FloatSeries).

To speed up the fetching process, this method can use Python's ThreadPoolExecutor. To enable it, set the use_threads parameter to True. By default, the maximum number of workers is 10. You can change this number by setting the NEPTUNE_FETCHER_MAX_WORKERS environment variable.

Parameters:

Name	Type	Default	Description
paths	List[str], required	None	List of paths to prefetch to the internal cache.
use_threads	bool, optional	False	Whether to use threads to fetch the data.

Example:

run.prefetch_series_values(["metrics/loss", "metrics/accuracy"])
# No more calls to the API
print(run["metrics/loss"].fetch_values())
print(run["metrics/accuracy"].fetch_values())

Available types

This section lists the available field types and data retrieval operations.

---

Boolean

fetch()

Retrieves a bool value either from the internal cache (see prefetch()) or from the API.

Example:

status = run["sys/failed"].fetch()

---

Datetime

fetch()

Retrieves a datetime.datetime value either from the internal cache (see prefetch()) or from the API.

Example:

created_at = run["sys/creation_time"].fetch()

---

Float

fetch()

Retrieves a float value either from the internal cache (see prefetch()) or from the API.

Example:

f1 = run["scores/f1"].fetch()

---

FloatSeries

fetch() or fetch_last()

Retrieves the last value of a series, either from the internal cache (see prefetch()) or from the API.

Returns: Optional[float]

Example:

loss = run["loss"].fetch_last()

fetch_values()

Retrieves all series values either from the internal cache (see prefetch_series_values()) or from the API.

Parameters:

Name	Type	Default	Description
include_timestamp	bool	True	Whether the fetched data should include the timestamp field.

Returns: pandas.DataFrame

Example:

values = run["loss"].fetch_values()

---

Integer

fetch()

Retrieves an int value either from the internal cache (see prefetch()) or from the API.

Example:

batch_size = run["batch_size"].fetch()

---

ObjectState

fetch()

Retrieves the state of a run either from the internal cache (see prefetch()) or from the API.

Returns: str

[!NOTE] The state can be active or inactive. It refers to whether new data was recently logged to the run. To learn more about this field, see System namespace: State in the Neptune docs.

Example:

state = run["sys/state"].fetch()

---

String

fetch()

Retrieves a str value either from the internal cache (see prefetch()) or from the API.

Example:

token = run["token"].fetch()

---

StringSet

fetch()

Retrieves a dict of str values either from the internal cache (see prefetch()) or from the API.

Example:

groups = run["sys/group_tags"].fetch()

License

This project is licensed under the Apache License Version 2.0. For more details, see Apache License Version 2.0.