Python 3.6 migration, adding setup.py, and Travis integration (#1)

* Move everything in a subfolder and add a setup.pu

* Use some pycharm magik

* Fixing install_requires and switch to find_packages

* Move init to pymb

* Adding an __all__ to stuff

* Add scikit-sparse as deps for cholesky

* Fixing imports, and added scikit-sparse to reqs

* Some more formatting changes

* Change file( to open(

* Change xrange to range

* iteritems to items

* Adding native rpy2 calls to pick up R version

* Adding a Dockerfile to set up conda environment

* Adding a travis yml

* Use Miniconda docker

* Add exec permish to bin 01

* Add a print statement in 01

* Adding miniconda in travis

* Only keep py 3.6

* Remove -- frmo apt-get

* Move apt packages to block

* Add binaries for dependency installation

* remove intel packages

* Why is stage not continuing..

* Move conda installation to 01 shell

* Move all the non R conda installs to use conda-forge

* Remove before install block

* Adding R and Python versions in jobs

* Use R as travis base

* Add DESCRIPT

* Move 01 and 02 to install block and add a 03 tmb run check

* Remove R-base from Conda install

* Add more Python checks

* Just have checks for now

* Reset path thing?

* Move apt-get installation inside 01

* try install again ugh

* Remove second install true?

* Try double scripting

* Caching all directories in Travis

* brek cache

* break... more

* Bad yml, fixing

* Move back conda to before inst

* install in ~/miniconda3

* Add some false installs

* A lot of refactor

* Add yes flag

* - Adding build status on README
- Squash build and test in a single stage
- Move the R package installation in the install block

* Adding apt-get installs of gcc and build ess

* Oops missed the folder nav to PyMB

* Switch to sourcing python instead of ipython

* Move PyMB install to install block and make sure that rpy2 imports

* Bad R link

* Pip install rpy2

* Try and get libRblas to pick up

* Update ld to include properly

* Add rpy2.rinterface import

* Test if we can run example without activating env in shell file

* Adding RcppEigen dependency for TMB

* Trying to test where TMB is breaking

* Add libeigen3 as installing

* Add RcppEigen include flag in linreg.py

* let's build TMB from source and not include the Eigen flag

* Bad CD

* Run example inline

.travis.yml

@@ -0,0 +1,38 @@
+sudo: required
+
+language: r
+
+os:
+  - linux
+
+
+# Setup anaconda
+before_install:
+  - wget http://repo.continuum.io/miniconda/Miniconda3-latest-Linux-x86_64.sh -O miniconda.sh
+  - chmod +x miniconda.sh
+  - ./miniconda.sh -b -p /home/travis/miniconda3
+  - export PATH=/home/travis/miniconda3/bin:$PATH
+  - export LD_LIBRARY_PATH=/home/travis/R-bin/lib/R/lib:/home/travis/miniconda3/lib:$LD_LIBRARY_PATH
+  - R -e 'install.packages(c("Matrix", "RcppEigen", "numDeriv"),  repos="http://cran.us.r-project.org", dependencies=T)'
+
+## Cache R packages
+cache: packages
+
+## Set up conda environment, TMB and PyMB
+install:
+  - conda create -c conda-forge -n pymb_test -y python=3.6 numpy scipy libiconv libxml2 lxml scikit-sparse
+  - source activate pymb_test
+  - which -a python && pip install rpy2 && python -c "import rpy2.rinterface as rinterface"
+  - git clone https://github.com/kaskr/adcomp.git && cd adcomp && make install-metis-full && cd ..
+  - cd ../PyMB && python setup.py install
+
+jobs:
+  include:
+    - stage: Run PyMB tests
+      script:
+        - python Examples/linreg.py
+
+notifications:
+  email:
+    on_success: never
+    on_failure: never

DESCRIPTION

@@ -0,0 +1 @@
+Package: PyMB

Dockerfile

@@ -0,0 +1,44 @@
+FROM continuumio/miniconda3
+
+## Get all the Linux dependencies
+RUN apt-get update --fix-missing && \
+    apt-get install --yes \
+        wget \
+        curl \
+        bzip2 \
+        ca-certificates \
+        libglib2.0-0 \
+        libxext6 \
+        libsm6 \
+        libxrender1 \
+        git \
+        gcc \
+        g++ \
+        libreadline-dev \
+        libeigen3-dev \
+        tar \
+        build-essential && \
+    apt-get purge && \
+    apt-get clean
+
+## Install Python dependencies
+## NOTE: DO NOT install rpy2 from conda-forge, got issues
+RUN conda install -c conda-forge -y numpy scipy && \
+    scikit-sparse libiconv libxml2 lxml
+
+## Install R and rpy2
+RUN conda install -c r -y \
+    r-base=3.5.0 && \
+    pip install rpy2
+
+
+## Build TMB and it's dependencies
+RUN R -e 'install.packages(c("Matrix", "numDeriv", "RcppEigen"),  repos="http://cran.us.r-project.org", dependencies=T)' && \
+    cd /opt && \
+    git clone https://github.com/kaskr/adcomp.git && cd adcomp && \
+    make install-metis-full
+
+## Install PyMB
+RUN cd /opt && \
+    git clone https://github.com/kforeman/PyMB.git && \
+    cd PyMB && python setup.py install

Examples/linreg.py

@@ -0,0 +1,68 @@
+import numpy as np
+import os
+
+
+import PyMB
+import rpy2.robjects as ro
+import rpy2.rinterface as rin
+
+# create an empty model
+m = PyMB.model(name='linreg_code')
+
+# code for a simple linear regression
+linreg_code = '''
+#include <TMB.hpp>
+template<class Type>
+Type objective_function<Type>::operator() (){
+// DATA
+  DATA_VECTOR(Y);
+  DATA_VECTOR(x);
+
+// PARAMETERS
+  PARAMETER(alpha);
+  PARAMETER(Beta);
+  PARAMETER(logSigma);
+
+// MODEL
+  vector<Type> Y_hat = alpha + Beta*x;
+  REPORT(Y_hat);
+  Type nll = -sum(dnorm(Y, Y_hat, exp(logSigma), true));
+  return nll;
+}
+'''
+
+# Get the necessary paths to compile TMB code
+tmbinclude = ro.r('paste0(find.package("TMB"), "/include")')[0]
+# eigeninclude = ro.r('paste0(find.package("RcppEigen"), "/include")')[0]
+# tmbinclude = tmbinclude + ' -I' + eigeninclude
+rinclude = rin.R_HOME + "/include"
+rlib = rin.R_HOME + "/lib"
+
+# rpy2.rinterface.R_HOME
+
+# compile the model
+m.compile(codestr=linreg_code,
+          cc='g++',
+          R=rinclude,
+          TMB=tmbinclude,
+          LR=rlib,
+          verbose=True,
+          load=True)
+
+# simulate data
+N = 100
+m.data['x'] = np.arange(N)
+m.data['Y'] = m.data['x'] + 0.5 + np.random.rand(N)
+
+# set initial parameter values
+m.init['alpha'] = 0.
+m.init['Beta'] = 0.
+m.init['logSigma'] = 0.
+
+# set random parameters
+m.random = ['alpha', 'Beta']
+
+# fit the model
+m.optimize(draws=500)
+print(m.report('Y_hat'))
+m.print_parameters()

PyMB/__init__.py

@@ -0,0 +1 @@
+from PyMB.model import *

magic/PyMB_magic.py → PyMB/magic/PyMB_magic.py

@@ -100,7 +100,8 @@ def PyMB(self, line, cell):
                 code += '}\n'
 
             # compile model
-            model.compile(codestr=code, output_dir=args.OUTPUT_DIR, cc=args.CCOMPILER, R=args.R_DIR, TMB=args.TMB_DIR, LR=args.R_LIB, verbose=args.VERBOSE)
+            model.compile(codestr=code, output_dir=args.OUTPUT_DIR, cc=args.CCOMPILER, R=args.R_DIR,
+                          TMB=args.TMB_DIR, LR=args.R_LIB, verbose=args.VERBOSE)
 
             # print if verbose
             if args.VERBOSE == True: