Osprey is written in Python, and can be installed with standard python machinery
# grab the latest version from github $ pip install git+git://github.com/pandegroup/osprey.git
# or clone the repo yourself and run `setup.py` $ git clone https://github.com/pandegroup/osprey.git $ cd osprey && python setup.py install
Currently, we recommend that you use the development version, since things are
moving fast. However, release versions from PyPI can be installed using
# grab the release version from PyPI $ pip install osprey
hyperopt(recommended, required for
MOE(recommended, required for
scipy(optional, for testing)
nose(optional, for testing)
You can grab most of them with conda.
$ conda install six pyyaml numpy scikit-learn sqlalchemy nose
Hyperopt can be installed with pip.
$ pip install hyperopt
To use the MOE search strategy,
osprey can call MOE via two interfaces
- MOE’s REST API, over HTTP
- MOE’s python API
Using the MOE REST API requires that you set up a MOE server somewhere. The recommended way to do this is via the MOE docker image. See the MOE documentation for more information.
To use the MOE python API, you must install MOE on the machines you use to run osprey. The MOE documentation has some information on how to do this, but it can be tricky. An easier alternative is to use the conda binary packages that we compiled for 64-bit linux (otherwise, sorry, you’re on your own).
conda install -c https://conda.binstar.org/rmcgibbo moe
See the github repo for more info on the compilation of these binaries.