experiment run

Execute an experiment locally. neptune run is an alias for neptune experiment run.

Without any arguments the command assumes that there is a main.py file
in the CWD (current working directory).

  neptune run
  neptune experiment run

You can execute any file within your code with:

  neptune run src/train_neural_net.py

You can pass command line arguments to your code after the Python script.

  neptune run src/train_neural_net.py --learning_rate 0.1

Read more about neptune experiment run options below to learn how to select
machine learning environment for your experiment, and much more.

Usage

neptune experiment run [-e ENVIRONMENT] [--log-channel PREFIX[:CHANNEL_NAME]]
  [--ml-framework ML_FRAMEWORK] [-x EXCLUDE] [-p PARAMETER]
  [-b FILE-OR-DIR] [--pip-requirements-file PIP-REQUIREMENTS-FILE]
  [--snapshot] [--offline] [--name NAME] [--description DESCRIPTION]
  [--tag TAG] [--property PROPERTY] [--project PROJECT]
  [--open-webbrowser [true|false]] [--disable-stdout-channel]
  [--disable-stderr-channel] [--config CONFIG-FILE] [-h] [-v]
  [--profile PROFILE] [--debug]
  [ENTRYPOINT [ARG [...]]]

Essential Options

Parameter Description
entrypoint Your Python script to run or any other executable (like bash): the main
entrypoint to your experiment. After the entrypoint, you can add any number of
arguments that should be passed to your entrypoint.
If no entrypoint is passed, Neptune will try to run [main.py](main.py) by default.
-e, --environment A machine learning environment. Visit
https://docs.neptune.ml/advanced-topics/environments/
for a list of environments supported by Neptune.
--log-channel, -l, --log-<br>channels List of prefixes based on which Neptune will build numeric channels.
Syntax:
   --log-channel prefix[:channel_name]
Example:
Let’s assume that an experiment prints:
   loss 0.5
   loss 0.6
   ...
In the example above using --log-channel loss will build a numeric loss
channel with values parsed from stdout/stderr. You can customize the channel’s
name with --log-channel loss:regularization_loss.
--ml-framework Integrate your experiment’s machine learning framework with Neptune.
Supported choices are: [keras, tensorflow].

Advanced Options

Parameter Description
-x, --exclude Files and directories that Neptune should leave out when uploading snapshot of
your code to the storage.
Wildcards are supported.
Example:
    --exclude .git --exclude service/*.log
-p, --parameter A parameter passed to your code (available via ctx.params) tracked by Neptune.
Explore more of parameters’ syntax here:
https://docs.neptune.ml/advanced-topics/experiments/#parameters
Read about how to use parameters to optimize your hyperparameter search here:
https://docs.neptune.ml/advanced-topics/hyperparameter-optimization/
Basic syntax:
    -p <name>:<value/values>:<description>
Example:
    -p 'learning_rate:0.1:Learning rate'
The parameter defined in the example above can be accessed from the code with
‘ctx.params.learning_rate’
-b, --backup Files and directories that should be uploaded to Neptune’s storage after the
experiment ends.
--pip-requirements-file A file containing a list of Python packages that your experiment needs.
Requirements file format with an example file can be found here:
http://pip.pypa.io/en/stable/reference/pip_install/#requirements-file-format
--snapshot If present, Neptune will make a snapshot of your sources in
~/.neptune/experiments/<experiment id> and run the experiment in this
directory. This option is useful if you want to change your code while running
an experiment.
--offline Execute the experiment locally, without connecting to Neptune.
This option is useful for early-stage debugging and when you don’t have access
to the Internet.

Experiment Characteristics

Parameter Description
--name Experiment’s name.
--description Experiment’s description.
Giving a detailed description is often extremely useful when coming back to the
experiment after some time.
--tag Experiment’s tags.
Use tags to quickly filter experiments.
--property Key-value pairs containing information you wish to associate with the
experiment.
Syntax:
    --properties key:value [key:value ...]
Example:
    --properties model-type:svm dataset:my-data-v3

Configuration

Parameter Description
--project Your project’s name.
You can set this globally using the neptune project activate command.
--open-webbrowser Automatically open browser when a new experiment is started.
--disable-stdout-channel Do not create text channel based on stdout.
--disable-stderr-channel Do not create text channel based on stderr.

Global Options

Parameter Description
--config Path to a Neptune CLI configuration file.
-h, --help Show this help message and exit.
-v, --version Show Neptune CLI version and exit.
--profile Set user profile to use.
--debug Run command with additional debug-level logging.