python - scipy ImportError on travis-ci -

The first time I am setting up Travis-CI, I standardly believe how I install:

  language: dragon dragon: - "Phle_ install installing the 2.7" # dependency: - sudo apt-get -qq update - sudo apt- install Get -qq Python numpy Python scipy Python opencv - install sudo apt-get -qq install libhdf5-serial-dev hdf5-tools: - "pip installation numexpr" - "pip installation cython" - "PIP setting r needs .txt --use-mirrors" # test Script f For Ane: nosetests   
 makes everything but when you start Nojaset, get me  
  importError: a module scipy.ndimage   
  Update:  Here is more direct performance of the problem  
  $ sudo apt-get install python-egg verse dragon-Speepi Python opencv $ python -c 'import scipy' Traceback (most recent call last): File "& lt; String & Gt; ", line 1, & lt; Module & gt; ImportError: the scipy named module "python -c 'import scipy' 'order failed and went out of 1 during the installation.   
 I tried installing Sipeepi using pus. I first tried to install gfortran ... any suggestions?  
  Another update : Since Travis, official document has been added to use Conde with Travis See the answer to Ostrocach.   
 
 
 Find two ways around the hairdresser:  
 
 
 As suggested by @unutbu, establish your own virtual environment and install PIP everything inside that environment. To get the build up, it is very slow to install the sip from this source.   
 
 After the approach used by the Pandas project, Travis is system-wide Site-package, and sample using an apt-hot Mr. and Speepi Install it is very sharp key lines  
  virtualenv: system_site_packages: true   
  before_install  before the group travis In .yml, after these open commands  
  SITE_PKG_DIR = $ VIRTUAL_ENV / lib / dragon $ TRAVIS_PYTHON_VERSION / site-package rm -f $ VIRTUAL_ENV / lib / dragon $ TRAVIS_PYTHON_VERSION / no-global site- Packages.txt   
 and then end  
  apt-get install dragon-numpy apt-get dragon-scipy  install  
 which will get when nosetests tries to import them.    
  Update   
 Now I prefer to build a conda-based, which is faster than either of the above I keep the strategies here on a project.