Skip to main content

A library for deploying Numer.ai Prediction Nodes.

Project description

numerai-cli

CircleCI PyPI

Welcome to the Numerai CLI for the Numerai Tournament. This CLI runs on your local computer to configure a Numerai Prediction Node in the cloud. This solution is architected to cost less than $5/mo on average, but actual costs may vary. It has been tested and found working on MacOS/OSX, Windows 8/10, and Ubuntu 18/20, but should theoretically work anywhere that Docker and Python 3 are available.

If you have any problems, questions, comments, concerns, or general feedback, please refer to the Troubleshooting and Feedback before posting anywhere.

Contents

Getting Started

  1. Sign up a Numerai Account, get your Numerai API Keys, and your first Model:

    1. Sign up at https://numer.ai/signup and log in to your new account
    2. Go to https://numer.ai/account > "Your API keys" section > click "Add"
    3. Name your key and check all boxes under "Scope this key will have"
    4. Enter your password and click "Confirm"
    5. Copy your secret public key and secret key somewhere safe
  2. Pick a Cloud Provider and follow the setup directions (Currently we only support AWS):

  3. Install Docker and Python for your Operating System (if you encounter errors or your OS is not supported, please read the troubleshooting wiki to install Python and Docker):

    • Mac Terminal (cmd + space, type terminal, select terminal.app):

      curl https://raw.githubusercontent.com/numerai/numerai-cli/master/scripts/setup-mac.sh | bash
      
    • Ubuntu 18/20 Terminal (ctrl + alt + t):

      sudo apt update && sudo apt install -y libcurl4 curl && sudo curl https://raw.githubusercontent.com/numerai/numerai-cli/master/scripts/setup-ubu.sh | sudo bash
      
    • Windows 10 Command Prompt (windows key, type cmd, select Command Prompt):

      powershell -command "$Script = Invoke-WebRequest 'https://raw.githubusercontent.com/numerai/numerai-cli/master/scripts/setup-win10.ps1'; $ScriptBlock = [ScriptBlock]::Create($Script.Content); Invoke-Command -ScriptBlock $ScriptBlock"
      
  4. After the setup script confirms Python and Docker, install numerai-cli via:

    pip3 install --upgrade numerai-cli --user
    

    NOTES:

    • This command will also work to update to new versions of the package in the future.
    • If you are using python venv then drop the --user option. If you don't know what that is, disregard this note.

Node Configuration Tutorial

If you know you have all the prerequisites and have your AWS and Numerai API Keys at hand, you can run these commands on your local computer to get an example node running in minutes:

numerai setup
numerai node config --example tournament-python3
numerai node deploy
numerai node test

If you want to use larger instances to generate your predictions first run numerai list-constants to list the vCPU/mem presets available, then you can configure a node to use one of the presets via:

numerai node config -s mem-lg

Your compute node is now setup and ready to run!

VERY IMPORTANT These commands have stored configuration files in $USER_HOME/.numerai/. If you lose this directory without destroying your nodes, you will have to manually (and painfully) delete every cloud resource by hand, so make sure you backup these files regularly! DO NOT USE THESE COMMANDS FROM A TEMPORARY INSTANCE!!!!

The example node will train and predict in the cloud, but this is inefficient. It's usually best to train your model locally (ensuring that the trained model is stored in the same folder as your code) and then deploy the trained model. When you make changes to your code or re-train your model, simply deploy and test your node again:

numerai node deploy
numerai node test

The nodes.json file also includes the url for your Node Trigger. This trigger is registered with whichever model you specified during configuration. Each trigger will be called Saturday morning right after a new round opens, and if the related job fails it will be triggered again around 24 hours later. You can opt into compute for daily tournaments via the website (go to numer.ai/models, click the 3 dots next to your model, click "compute", and click the switch to opt in).

NOTES:

  • The default example does not make stake changes; you will still have to do that manually. Please refer to the numerapi docs for the methods you must call to do this.
  • You can view resources and logs in the AWS Console (region us-east-1) for your ECS Cluster and other resources

NEXT: read the Prediction Nodes wiki to learn about Numerai Examples and how to customize Prediction Nodes.

List of Commands

Use the --help option on any command or sub-command to get a full description of it:

numerai
numerai --help
numerai [command] --help
numerai [command] [sub-command] --help

Each command or sub-command takes its own options, for example if you want to copy the numerai signals example and configure a node for a signals model with large memory requirements you'd use something like this (replacing [MODEL NAME] with the relevant signals model name):

numerai node -m [MODEL NAME] -s config -s mem-lg -e signals-python3

Here, the node command takes a model name with -m and a flag -s to detect if it's a signals model or numerai model. The config sub-command also takes a -s option to specify the size of the node to configure.

Troubleshooting and Feedback

Before messaging the Rocketchat channel or creating a Github issue, please read through the following (especially the "Troubleshooting" section in the wiki):

If you still cannot find a solution or answer, please join us on the RocketChat #support Channel and include the following information with your issue/message:

  • The commands you ran that caused the error (even previous commands)
  • Version information from running:
    • pip3 show numerai-cli
    • python -V
    • docker -v
  • System Information from running
    • Mac: system_profiler SPSoftwareDataType && system_profiler SPHardwareDataType
    • Linux: lsb_release -a && uname -a
    • Windows: powershell -command "Get-ComputerInfo"

If you do not include this information, we cannot help you.

Upgrading and Updating

Upgrading numerai-cli will always require you to update the package itself using pip:

pip install --upgrade numerai-cli --user

0.1/0.2 -> 0.3.0

CLI 0.3.0 uses a new configuration format that is incompatible with versions 0.1 and 0.2, but a command to migrate you configuration is provided for you. Run this in the directory you ran numerai setup from the previous version:

numerai upgrade

Beyond

Some updates will make changes to configuration files used by Numerai CLI. These will require you to re-run some commands to upgrade your nodes to the newest versions:

  • numerai setup will copy over changes to files in the $HOME/.numerai directory
  • numerai node config will apply those changes to a node

Uninstall

numerai uninstall

Contributions

Project details


Release history Release notifications | RSS feed

Download files

Download the file for your platform. If you're not sure which to choose, learn more about installing packages.

Source Distribution

numerai-cli-0.3.5.tar.gz (5.4 MB view details)

Uploaded Source

Built Distribution

numerai_cli-0.3.5-py3-none-any.whl (5.4 MB view details)

Uploaded Python 3

File details

Details for the file numerai-cli-0.3.5.tar.gz.

File metadata

  • Download URL: numerai-cli-0.3.5.tar.gz
  • Upload date:
  • Size: 5.4 MB
  • Tags: Source
  • Uploaded using Trusted Publishing? No
  • Uploaded via: twine/4.0.2 CPython/3.7.16

File hashes

Hashes for numerai-cli-0.3.5.tar.gz
Algorithm Hash digest
SHA256 069235dbf24f7e2af6aab59ca984e6e647a1228348c7a6b2d20d206a1b2520d6
MD5 09e7b63799ca3e69360f197ba1e2103b
BLAKE2b-256 264fd60724a4d7e2858941fea3b203e3dd80703a4a48f4aa39cc9c9aeb9f2886

See more details on using hashes here.

Provenance

File details

Details for the file numerai_cli-0.3.5-py3-none-any.whl.

File metadata

  • Download URL: numerai_cli-0.3.5-py3-none-any.whl
  • Upload date:
  • Size: 5.4 MB
  • Tags: Python 3
  • Uploaded using Trusted Publishing? No
  • Uploaded via: twine/4.0.2 CPython/3.7.16

File hashes

Hashes for numerai_cli-0.3.5-py3-none-any.whl
Algorithm Hash digest
SHA256 3446daca098216d9eb8d116118e757327bd31fa8678f9820a3724958b9c66b30
MD5 ef713723b75095d835f39d9f27052dea
BLAKE2b-256 f49f40efcb28e6f9e9cbbf2261e4f2d80ce55a21f4a7278dbd0a60945435e2da

See more details on using hashes here.

Provenance

Supported by

AWS AWS Cloud computing and Security Sponsor Datadog Datadog Monitoring Fastly Fastly CDN Google Google Download Analytics Microsoft Microsoft PSF Sponsor Pingdom Pingdom Monitoring Sentry Sentry Error logging StatusPage StatusPage Status page