Marvin does not work well with the system Python in OSX. Please, make sure you are using a supported Python installation before following these instructions. Good installations include Anaconda, Miniconda, or homebrew. After installing one of these distribution, make sure you are actually using it by running which python and which pip. These commands should print a path to the Python and pip executables within your virtual environment.


Quick Install

To install Marvin, use:

pip install sdss-marvin

If you are using an Anaconda distribution of Python, you may use the following available conda environment, here. Once downloaded, set up the virtual environment with:

conda env create -f marvin_2.7.1.yml

To create a clean, isolated, conda virtual environment from scratch, and install marvin into it, do the following:

conda create -n marvin python=3.8 ipython
conda activate marvin
pip install sdss-marvin

For troubleshooting installation issues, please see Install and Runtime Issues or the Marvin Github Issues page.

Developer’s Installation

To develop for marvin, follow these instructions:

git clone https://github.com/sdss/marvin
cd marvin
pip install -e .

Use this installation for development and testing, i.e. if you wish to fix bugs or add new features. This will checkout the repository, install marvin into your python path using pip’s editable mode. Afterwards, you can start developing for Marvin, without having to re-install it every time your code changes.

Access and Authentication

Public Access

By default Marvin is set up to run in public access mode, with the latest SDSS public data release, e.g. DR15. You can check your access from within an iPython terminal.

In your terminal, run:

$ ipython

Within your ipython session, type:

from marvin import config

A config.access of public means you are set up for public access only. In this mode, you only have access to publically available data. A config.access of collab indicates you are set up for SDSS collaboration proprietary data access.

SDSS Collaboration Access

For SDSS collaboration members, authentication is required to access proprietary collaboration data, and Marvin must have config.access set to collab. See more here. To set up authentication for Marvin, you must perform the following:

Set up your netrc

SDSS uses .netrc authentication to access data content on many domains. To set this up, create and edit a file in your home called .netrc and copy these lines inside:

machine api.sdss.org
   login <username>
   password <password>

machine data.sdss.org
   login <username>
   password <password>

and replace <username> and <password> with your login credentials. The default SDSS username and password is also acceptable for anonymous access. Finally, run chmod 600 ~/.netrc to make the file only accessible to the user.

API Token Authentication

Marvin requires token authentication to grant access and use of its API. Marvin uses the standard JSON Web Tokens for token authentication. To receive a valid token, you must login with your valid SDSS credentials, via the .netrc. With your netrc access in place, you will receive a valid API token. Tokens remain valid for 300 days.:

# login to receive a token

# see token

You will need a token every time you start a new python session. To change this, follow the steps in Automatically Logging In.

Automatically Logging In

As the default mode of marvin is public, you will need to authenticate and change to collab access inside every new iPython session. To simplify this process, marvin can be configured to automatically perform the access and authentication checks. To configure marvin, you must set up a custom marvin configuration file. Inside a ~/.marvin/marvin.yml file, set the following lines:

check_access: True
use_token: [token]

You can replace [token] with your authenticated API JSON token (without any string quotes). Upon import of marvin, Marvin will check for valid credentials and automatically set up your collaboration access.

Marvin Environment

Marvin requires a certain environment structure to access and (optionally) download data. By default, marvin will look for data files in a directory structure that mirrors the Science Archive Server. Data downloaded via marvin will also be stored according to that structure. The root of this directory structure is defined by the environment variable $SAS_BASE_DIR. For example, if marvin needs to use the drpall file for DR15, it will try to find it in $SAS_BASE_DIR/dr15/manga/spectro/redux/v2_4_3/drpall-v2_4_3.fits.

The Marvin environment structure is as follows:

======================   ==============================================   ======
Environment Variable     Default Path                                     Access
======================   ==============================================   ======
SAS_BASE_DIR             $HOME/sas
MANGA_SPECTRO_REDUX      $SAS_BASE_DIR/dr15/manga/spectro/redux           DR15
MANGA_SPECTRO_ANALYSIS   $SAS_BASE_DIR/dr15/manga/spectro/analysis        DR15

MANGA_SPECTRO_REDUX      $SAS_BASE_DIR/mangawork/manga/spectro/redux      collab
MANGA_SPECTRO_ANALYSIS   $SAS_BASE_DIR/mangawork/manga/spectro/analysis   collab
======================   ==============================================   ======

Marvin will check for these environment variables in your local system. If the above environment variables are not already defined, Marvin will use the specifed default paths. Otherwise Marvin will adopt your custom paths. If you wish to define custom paths, you can update the environment variable paths in your .bashrc or .cshrc file. As a general advice, if you are not using other products that require setting those environment variables, you should only define $SAS_BASE_DIR (or not define it and let Marvin configure itself).

Dependencies on SDSS software

Marvin depends on three pieces of SDSS-wide software:

  • marvin_brain: contains some core functionality, such as the API call framework, the basic web server, etc.

  • tree: defines the structure of the Science Archive Sever, relative paths to data products, etc.

  • sdss_access: tools for efficiently accessing data files, rsyncing data, etc.

For convenience, Marvin includes these products as external libraries, and you should not need to worry about them. However, if any of these libraries, except the tree product, are already installed in your system (i.e., you have defined $MARVIN_BRAIN_DIR, or $SDSS_ACCESS_DIR), Marvin will use the system wide products instead of its own versions. This is useful for development but note that it can lead to confusion about what version Marvin is using.

Install and Runtime Issues


We can use your help to expand this section. If you have encountered an issue or have questions that should be addressed here, please submit an issue.

Uwsgi package installation failure

For Python versions 3.8 and above, the uwsgi package fails to compile during standard pip installation. Alternatively, before installing Marvin, you can install the uwsgi package using conda with:

conda install -c conda-forge uwsgi

Afterwards installaton, marvin should install successfully.

Pip Failure with Python-Memcache

If pip fails while installing python-memcached, make sure that you have the latest version of setuptools by running pip install -U setuptools. Then, try running pip install sdss-marvin again.

How do I update marvin?

To upgrade an existing Marvin installation, run:

pip install -U sdss-marvin

By default, pip will update any underlying package on which marvin depends. If you want to prevent that you can upgrade marvin with pip install -U --no-deps sdss-marvin. This could, however, make marvin not work correctly. Instead, you can try pip install -U --upgrade-strategy only-if-needed sdss-marvin, which will upgrade a dependency only if needed.

Permissions Error

If your Marvin installation fails at any point during the pip install process with permissions problems, try running sudo pip install sdss-marvin. Note that an Anaconda or Homebrew distribution will not require permissions when pip installing things, so if you are receiving permissions errors, you may want to check that you are not using the Mac OSX system version of Python.

If you receive a permissions error regarding pip attempting to install a package in a different directory other than the Anaconda one, e.g. /lib/python3.6, try following the solution indicated in Marvin Issue 373

How to test that marvin has been installed correctly

Marvin is designed to get you started with minimal configuration on your part. However, this means that Marvin may import correctly but may not have all features immediately available. Here are a few commands you can try that will inform you if there are problems with your installation.

From a terminal window, type:


This will perform a variety of checks with Marvin and output the results to the terminal. We may ask you for this output when diagnosing any installation issues. After installing marvin, start a python/ipython session and run:

import marvin

If you get a dictionary filled with url routes, marvin is connecting correctly to the API server at Utah and you can use the remote features. If you get None, you may want to check the steps in Set up your netrc. If you get an error message such as

BrainError: Requests Timeout Error: HTTPSConnectionPool(host='api.sdss.org', port=443): Read timed out.
Your request took longer than 5 minutes and timed out. Please try again or simplify your request.

this means the servers at Utah have timed out and may possibly be down. Simply wait and try again later.

Marvin Remote Access Problems

If the above urlmap test crashes, or you attempt to use a Marvin Tool remotely, and you see this error:

AttributeError: 'Extensions' object has no attribute 'get_extension_for_class'

This is an issue with the Urllib and Requests python package. See this Issue for an ongoing discussion if this problem has been solved.

Lots of Warnings Upon import

If you see lots of warnings upon import of marvin, from /_bootstrap.py and referencing numpy.ufunc size changed, may indicate binary incompatibility, such as

import marvin
/anaconda3/envs/marvin_public/lib/python3.6/importlib/_bootstrap.py:219: RuntimeWarning: numpy.ufunc size changed, may indicate binary incompatibility. Expected 192 from C header, got 216 from PyObject
  return f(*args, **kwds)
/anaconda3/envs/marvin_public/lib/python3.6/importlib/_bootstrap.py:219: RuntimeWarning: numpy.ufunc size changed, may indicate binary incompatibility. Expected 192 from C header, got 216 from PyObject
  return f(*args, **kwds)

this arises when a Python package that uses Cython is compiled against a different version of numpy than is actually installed. See this article for more information. The consensus is that these warnings are fairly harmless and benign.

Matplotlib backend problems

Some users have reported that after installing marvin they get an error such as:

Python is not installed as a framework. The Mac OS X backend will not be able to function correctly if Python is not installed as a framework.

This problem is caused by matplotlib not being able to use the MacOS backend if you are using Anaconda. You need to switch your matplolib backend to Agg or TkAgg. Follow these instructions to fix the problem. If you do want to use the MacOS backend, consider installing Python using homebrew.

Web Browser Oddities

If the MPL dropdown list in the top menu bar is blank, or other elements appear to disappear, this is an indication your browser cache is creating conflicts. The solution is to clear your browser cache, close and restart your browser from scratch. You can also clear your browser cookies.

As a reminder, we recommend these browsers for the best Marvin web experience:

  • Google Chrome 53+ or higher

  • Mozilla Firefox 50+ or higher

  • Safari 10+ or Safari Technology Preview

Using IPython

If you plan to work with Marvin interactively, from the Python terminal, we recommend you use IPython, which provides many nice features such as autocompletion, between history, color coding, etc. It’s also especially useful if you plan to use Matplotlib, as IPython comes with default interactive plotting. If you installed Python via the Anaconda or Miniconda distributions, then you already have IPython installed. Just run ipython in your terminal. If you need to install it, do pip install jupyter.

Marvin on Windows

Marvin was originally designed to work on Mac or Linux operating systems. However it is possible at the moment to get Marvin working on Windows machines. The following guidelines have been tested on a Windows 10 machine running Python 3.6.

  • Install a Python version for Windows. Make sure to check the box to include Python in your environment variable Paths. If you are using Anaconda to install Python, make sure to check both the “Add Anaconda to my PATH environment variable” and “Register Anaconda as my default Python 3.6”

  • Marvin expects a HOME directory. Add this snippet of code before any of use of Marvin.

import os
os.environ['HOME'] = '/path/you/want/as/marvin/home/directory'
os.environ['SAS_BASE_DIR'] = os.path.join(os.getenv("HOME"), 'sas')
To add a permanent HOME path, follow these instructions.
  • open File Explorer, right click “This PC” on the left scroll bar and click Properties

  • on the left, click ‘Advanced System Settings’. You need Admin Privileges to do this.

  • on the bottom, there should be an ‘Environment Variables’ box. Below the User Variables column, click New.

  • add a new HOME environment variable that points to /path/you/want/as/marvin/home/directory.

  • Create the .netrc file and place it the directory you designated as HOME. You will need to modify the permissons of this file to match the expected chmod 600 permissions for Mac/Linux users. When creating the file, you can name it as anything but can rename it to .netrc from the command prompt.

With this, you should be able to run Marvin in windows. You can test it with import marvin. Currently, Marvin cannot download files due to issues with forward slashes in sdss-access but this will be fixed soon. We will continue to update these guidelines as we make further progress on a Windows-Marvin installation.