Install the Machine Learning Toolkit

The Machine Learning Toolkit (MLTK) enables users to create, validate, manage, and operationalize machine learning models through a guided user interface.

Use the following directions to install the MLTK on to your system(s).

Requirements

The current version of the Splunk Machine Learning Toolkit is 4.3.0. In order to successfully run this version, the following is required:

• Splunk Enterprise 7.0 or later or Splunk Cloud
• Installation of the Machine Learning Toolkit app from Splunkbase
• Installation of the Python for Scientific Computing add-on version 1.3 or 1.4

• Mac OS
• Linux 64-bit
• Linux 32-bit
• Windows 64-bit

In order to save models, users need the upload_lookup_files capability included in their role.

You cannot access new features in the MLTK without upgrading to the latest version. Versions 3.4.0 and above of the MLTK require upgrading to versions 1.3 or 1.4 of the PSC add-on. See the version dependencies table for the specific requirements between MLTK and PSC add-on versions.

Linux 32-bit support is not available should you upgrade to version 1.4 of the Python for Scientific Computing add-on.

Specific version dependencies

MLTK Version	PSC Version
4.3	1.3 or 1.4
4.2	1.3 or 1.4
4.1	1.3
4.0	1.3
3.4	1.3
3.3	1.2 or 1.3
3.2	1.2 or 1.3
3.1	1.2

If you have written any custom algorithms that rely on the PSC libraries, upgrading to version 1.3 or 1.4 the PSC library add-on will impact those algorithms. You will need to re-train any models (re-run the search that used the fit command) using those algorithms after you upgrade the PSC add-on.

Splunk Cloud deployments

Follow the appropriate directions below for your instance of self-service or managed Splunk Cloud.

Splunk Cloud trial and self-service Splunk Cloud

Install the Python for Scientific Computing add-on and the Machine Learning Toolkit app to your self-service instance of Splunk Cloud using the app browser in Splunk Cloud.

1. Log in to your Splunk Cloud instance.
2. From the Splunk Web home screen, click on the gear icon next to Apps in the left navigation bar.
3. Click Browse more apps.
4. Search for the Python for Scientific Computing add-on and install it.
5. Search for the Machine Learning Toolkit app and install it.

Managed Splunk Cloud

Open a ticket with support and request the Python for Scientific Computing add-on and the Machine Learning Tooklit app to be installed for you.

Splunk Enterprise deployments

Follow these directions for either single or distributed deployments.

Single instance deployment

Install the Python for Scientific Computing add-on and the Machine Learning Toolkit app onto your single instance Splunk Enterprise

1. Install the Python for Scientific Computing add-on first (required).
2. Install the Splunk Machine Learning Toolkit app.

Install an app or add-on in Splunk Web

1. In Splunk Web, click on the gear icon next to Apps in the left navigation bar.
2. On the Apps page, click Install app from file.
3. Click Choose File, navigate to and select the package file for the app or add-on, then click Open.
4. Click Upload.

Install an app or add-on from the command line

At the command line, enter the following content, depending on your operating system.

Unix/Linux:

./splunk install app <path/packagename>

Windows:

splunk install app <path\packagename>

Alternatively, unpack/unzip the file then copy the app directory to $SPLUNK_HOME/etc/apps on Unix based systems or %SPLUNK_HOME%\etc\apps on Windows systems.

Distributed deployment

Use the following tables to determine where and how to install the Machine Learning Toolkit and Python for Scientific Computing add-on in a distributed deployment of Splunk Enterprise. Depending on your environment, you may need to install the Machine Learning Toolkit and Python for Scientific Computing add-on in multiple places.

Where to install Splunk Machine Learning Toolkit and Python for Scientific Computing

This table provides a reference for installing the Machine Learning Toolkit and Python for Scientific Computing to a distributed deployment of Splunk Enterprise.

Splunk instance type	Supported	Required	Actions required / Comments
Search Heads	Yes	Yes	Install Python for Scientific Computing and the Splunk Machine Learning Toolkit to all search heads where the Splunk Machine Learning Toolkit is used. Search heads must be running Splunk Enterprise 6.6 or greater.
Indexers	Yes	Conditional	If you want to use the distributed apply feature of the Splunk Machine Learning Toolkit, install Python for Scientific Computing on all of your indexers. This feature is disabled by default. See Use your indexers to apply models for information. Indexers must be running Splunk Enterprise 6.6 or greater. The Splunk Machine Learning Toolkit does not need to be installed on the indexers to enable this functionality.
Heavy Forwarders	Yes	No	These apps do not contain a data collection component.
Universal Forwarders	Yes	No	These apps do not contain a data collection component.
Light Forwarders	Yes	No	These apps do not contain a data collection component.

Distributed deployment feature compatibility

This table describes the compatibility of the Splunk Machine Learning Toolkit and Python for Scientific Computing add-on with Splunk distributed deployment features.

Distributed deployment feature	Supported	Actions required
Search Head Clusters	Yes	Search heads must be running Splunk Enterprise 6.6 or greater.
Indexer Clusters	Yes	If you want to use the distributed apply feature of the Splunk Machine Learning Toolkit, install Python for Scientific Computing on the indexers in your cluster. This feature is disabled by default. See Use your indexers to apply models for information. Indexers must be running Splunk Enterprise 6.6 or greater. The Splunk Machine Learning Toolkit does not need to be installed on the indexers in your cluster to enable this functionality.

Use your indexers to apply models

If you have more than one Splunk indexer and want to take advantage of the parallel computing power available on your standalone Splunk indexers or Splunk indexing cluster, you can configure your indexers to run the apply command, a CPU-intensive task that applies machine-learning models.

Follow these steps to use your indexers to apply model:

1. Install the Python for Scientific Computing add-on on all of your indexers.
2. On each search head in your deployment, open the local mlspl.conf configuration file in a text editor.
   For Unix based systems: $SPLUNK_HOME/etc/apps/Splunk_ML_Toolkit/local/mlspl.conf
   For Windows systems: %SPLUNK_HOME%\etc\apps\Splunk_ML_Toolkit\local\mlspl.conf
   Create the mlspl.conf in the local directory if one does not exist.
3. Copy the [default] stanza from the default mlspl.conf configuration file to the local version of the configuration file if this stanza is not present.
   Location of default mlspl.conf file on Unix based systems: $SPLUNK_HOME/etc/apps/Splunk_ML_Toolkit/default/mlspl.conf
   Location of default mlspl.conf file on Windows systems: %SPLUNK_HOME%\etc\apps\Splunk_ML_Toolkit\default\mlspl.conf
4. Change the streaming_apply command to be true as follows: streaming_apply = true

Use the deployment methodology of your choice to make these configuration changes.

• To learn about updating search head cluster members, see Use the deployer to distribute apps and configuration updates in the Distributed Search manual.
• To learn about updating peers in an indexer cluster, see Manage app deployment across all peers in the Managing Indexers and Clusters of Indexers manual.

Bundle replication

Permanent model files, sometimes referred to as learned models or encoded lookups, are saved on disk. These files follow Splunk knowledge object rules, including permissions and bundle replication. Bundle replication is the process by which knowledge objects on the search head are distributed to the indexers.

The Machine Learning Toolkit includes a number of example model files that support the Showcase page. These examples are powered by .csv lookup files. To prevent performance issues, these .csv lookup files are not included in the toolkit's bundle replication process.