Installing spit

This document describes how to install the spit repository.

Installing Dependencies

There are a number of non-standard dependencies related to the CNN(s) that lie within SPIT.

In general, we recommend that you use Anaconda for the majority of these installations.

Detailed installation instructions are presented below:

Python Dependencies

specdb depends on the following list of Python packages.

We recommend that you use Anaconda to install and/or update these packages.

  • python versions 2.7, or 3.6 or later [2.7 will be phased out]
  • numpy version 1.13 or later
  • astropy version 3.0 or later
  • scipy version 0.19 or later
  • tensorflow version 1.4.1 or later
  • pillow version 4.2 or later
  • prettytensor version 0.7.4 or later [This dependency may be eventually eliminated]

If you are using Anaconda, you can check the presence of these packages with:

conda list "^python|numpy|astropy|scipy|tensorflow|pillow|prettytensor"

If the packages have been installed, this command should print out all the packages and their version numbers.

If any of these packages are missing you can install them with a command like:

pip install tensorflow

If any of the packages are out of date, they can be updated with a command like:

conda update scipy
# orL
pip update tensorflow

Installing spit

Presently, you must download the code from github:

#go to the directory where you would like to install specdb.
git clone https://github.com/PYPIT/spit.git

From there, you can build and install with

cd spit python setup.py install # or use develop

This should install the package and scripts. Make sure that your PATH includes the standard location for Python scripts (e.g. ~/anaconda/bin)

Installing Architectures

SPIT needs a trained CNN architecture to perform its `magic’. At present, there is only one (trained on Kast images).

Kast

Trained on a set of Kast images as described in Jankov & Prochaska (2018). Appears to work relatively well on other instruments (not extensively tested).

kast_original (unpacks to 811Mb)

  • Original CNN generated by Vik Jankov as described in Yankoff & Prochaska (2018)
  • Download the folder from this Google Drive
  • cd spit/data/checkpoints/
  • Unpack the downloaded zip file
  • It should be named kast_original