Merge pull request janelia-flyem#32 from jni/tree-agglo

Tree-based agglomeration

.coveragerc

@@ -0,0 +1,11 @@
+[run]
+omit =
+    *test_package*
+    *test_*
+    *auto*
+    *segmentation_stitch*
+    *stack_np*
+    *stitch*
+    *valprob*
+    *setup*
+    *__init__*

.travis.yml

@@ -1,56 +1,38 @@
-# Adopted from networkx travis.yml
-
 language: python
 python:
   - "2.7"
 virtualenv:
   system_site_packages: true
-before_install:
-  # Current TravisCI VM is Ubuntu 12.04 which has Python 2.7 and 3.2.
-  # So apt-get will only apply to those virtualenvs.
-  - sudo apt-get install -qq python-numpy python-scipy python-matplotlib python-pydot python-gdal
-  - sudo apt-get install -qq libhdf5-serial-dev
+install:
+  # all installing is now handled by conda as it is faster and more robust
+  - sudo apt-get update
+  - pip install -U setuptools
+  - wget http://repo.continuum.io/miniconda/Miniconda-3.4.2-Linux-x86_64.sh -O miniconda.sh;
+  - bash miniconda.sh -b -p $HOME/miniconda
+  - export PATH="$HOME/miniconda/bin:$PATH"
+  - hash -r
+  - conda config --set always_yes yes --set changeps1 no
+  - conda update -q conda
+  - conda info -a
 
-  # Additionally, the TravisCI VM has numpy preinstalled on 2.6, 2.7 and 3.2.
-  # So for 3.3 and pypy, numpy is not preinstalled and must be pip-installed.
-  # Matplotlib, pydot, and gdal are not available for 3.2 via apt-get.
-  # And they are also not available for 2.6, 3.3, and pypy except via pip.
-  # We could install via pip, but...
-  #   skipping numpy (for 3.3 and pypy) since it takes too long.
-  #   skipping scipy (for 2.6, 3.3, and pypy) since it takes too long.
-  #   skipping matplotlib (for 2.6, 3.3, and pypy) since it takes too long.
-  #   skipping pydot (for all) since it is not 3.x compatible.
-  #   skipping gdal (for all) since it has errors while installing from pip.
+  - conda create -q -n test-environment python=$TRAVIS_PYTHON_VERSION numpy scipy matplotlib networkx cython h5py PIL nose scikit-image scikit-learn setuptools pip
+  - source activate test-environment
 
-  - pip install pyyaml --use-mirrors
-  - pip install pyparsing --use-mirrors
-  #- pip install coveralls --use-mirrors
-  - pip install networkx --use-mirrors
-  - pip install cython
-  - pip install h5py
-  - pip install PIL
-  - pip install nose
-  - pip install scikit-learn
-  - pip install scikit-image
+  # custom package not available from conda
+  - pip install viridis
+
+  # install testing and coverage packages
+  - pip install pytest pytest-cov
+  - pip install coveralls
 
-script:
   # Install gala
   - printenv PWD
-  - python setup.py install
+  # Build locally, pytest confuses local and global installs
+  - python setup.py develop
 
+script:
   # Run tests
-  - cd tests
-  - python -m nose test_agglo.py
-  - python -m nose test_features.py
-  - python -m nose test_watershed.py
-  - python -m nose test_gala.py
+  - py.test
 
 after_success:
-  # We must run coveralls from the git repo.
-  # But we ran tests from the installed directory.
-  # The .coverage file looks like it contains absolute paths,
-  # but coveralls does not seem to care.
-
-  #- cp .coverage $TRAVIS_BUILD_DIR
-  #- cd $TRAVIS_BUILD_DIR
-  #- if [[ $TRAVIS_PYTHON_VERSION == '2.7' ]]; then coveralls; fi
+  - coveralls

README.md

@@ -20,7 +20,7 @@ volumes...) and multiple channels per image.
 * HDF5 and h5py 1.5+
 * nose 1.3+
 * cython 0.17+
-* scikit-learn 0.10+, preferably 0.14+
+* scikit-learn 0.15
 * matplotlib 1.2+
 * scikit-image 0.9+
 
@@ -29,15 +29,19 @@ volumes...) and multiple channels per image.
 * progressbar 2.3-dev
 * [vigra/vigranumpy](hci.iwr.uni-heidelberg.de/vigra/) (1.9.0)
 
-For vigra, you are on your own. It is used for the random forest classifier,
-but if you don't install it you can use any of the scikit-learn classifiers,
+In its original incarnation, this project used Vigra for the random forest
+classifier. Installation is less simple than scikit-learn, which has emerged
+in the last year as a truly excellent implementation and is now recommended.
+Tests in the test suite expect scikit-learn rather than Vigra.
+You can also use any of the scikit-learn classifiers,
 including their newly-excellent random forest.
 
 ## Installation
 
 ### Installing gala
 
-Gala is a pure python library and can be installed in two ways:
+Gala is a Python library with limited Cython extensions and can be 
+installed in two ways:
 * Add the gala directory to your PYTHONPATH environment variable, or
 * Use distutils to install it into your preferred python environment:
 
@@ -47,19 +51,16 @@ $ python setup.py install
 
 ### Installing requirements
 
-You can install all the requirements yourself: most are available in
-the Python Package Index (PyPI) and can be installed with simple commands:
+Though you can install all the requirements yourself, as most are available in
+the Python Package Index (PyPI) and can be installed with simple commands,
+the easiest way to get up and running is to use the 
+[Continuum Anaconda](http://www.continuum.io/downloads) Python distribution.
+An example of installing all dependencies for gala on linux is shown in
+the `.travis.yml` test recipe.
 
-```bash
-$ pip install scikit-learn
-```
-
-Alternatively, a number of Python distributions include all the above
-dependencies and a bunch more for good measure. Two examples are
-[Continuum Anaconda](http://www.continuum.io/downloads) and
-[Enthought Canopy](https://www.enthought.com/products/canopy/).
+### Installing with Buildem
 
-Finally, you can use Janelia's own
+Alternatively, you can use Janelia's own
 [buildem system](http://github.com/janelia-flyem/buildem#readme) to
 automatically download, compile, test, and install requirements into a
 specified buildem prefix directory. (You will need CMake.) 
@@ -86,6 +87,7 @@ $ cd tests
 $ python test_agglo.py
 $ python test_features.py
 $ python test_watershed.py
+$ python test_optimized.py
 $ python test_gala.py
 ```