Replaced cxfreeze with pyinstaller. Works better.

python/autotune/CMakeLists.txt

@@ -1,13 +1,18 @@
-set(SETUP_PY_IN "${CMAKE_CURRENT_SOURCE_DIR}/setup.py")
+find_program(PYINSTALLER pyinstaller)
-set(SETUP_PY    "${CMAKE_CURRENT_BINARY_DIR}/setup.py")
+
+if(PYINSTALLER)
+
+set(SPEC_IN "${CMAKE_CURRENT_SOURCE_DIR}/pyinstaller_build.spec")
+set(SPEC    "${CMAKE_CURRENT_BINARY_DIR}/pyinstaller_build.spec")
 set(OUTPUT      "${CMAKE_CURRENT_BINARY_DIR}/build/timestamp")
 file(GLOB DEPS RELATIVE "${CMAKE_CURRENT_SOURCE_DIR}" "${CMAKE_CURRENT_SOURCE_DIR}/pysrc/*.py")
-LIST(APPEND DEPS "${CMAKE_CURRENT_SOURCE_DIR}/setup.py")
+LIST(APPEND DEPS "${CMAKE_CURRENT_SOURCE_DIR}/pyinstaller_build.spec")
 
-configure_file(${SETUP_PY_IN} ${SETUP_PY})
+configure_file(${SPEC_IN} ${SPEC})
 add_custom_command(OUTPUT ${OUTPUT}
-                   COMMAND ${PYTHON} ${SETUP_PY} build
+                   COMMAND ${PYINSTALLER} ${SPEC_IN} ${CMAKE_CURRENT_SOURCE_DIR}
                    COMMAND ${CMAKE_COMMAND} -E touch ${OUTPUT}
-                   DEPENDS ${DEPS})
+                   DEPENDS ${DEPS} pyatidlas)
 add_custom_target(autotune ALL DEPENDS ${OUTPUT})
-install(CODE "execute_process(COMMAND ${PYTHON} ${SETUP_PY} install)")
+
+endif()

python/autotune/pyinstaller_build.spec

@@ -0,0 +1,34 @@
+#!/usr/bin/env
+import os, sys
+
+prefix = sys.argv[2]
+
+sys.path.append('/home/philippe/Development/ATIDLAS/build/python/pyatidlas/build/lib.linux-x86_64-2.7/')
+sys.path.append('/home/philippe/Development/pyviennacl-dev/build/lib.linux-x86_64-2.7/')
+sys.path.append(os.path.join(prefix, 'pysrc'))
+
+a = Analysis([os.path.join(prefix, 'pysrc','autotune.py')],
+         hiddenimports=['scipy.sparse.csgraph._validation',
+                         'scipy.special._ufuncs_cxx',
+                         'scipy.sparse.linalg.dsolve.umfpack',
+                         'scipy.integrate.vode',
+                         'scipy.integrate.lsoda',
+                         'sklearn.utils.sparsetools._graph_validation',
+                         'sklearn.utils.sparsetools._graph_tools',
+                         'sklearn.utils.lgamma'],
+         hookspath=None,
+         excludes=['scipy.io.matlab','matplotlib','PyQt4'],
+         runtime_hooks=None)
+dict_tree = Tree(os.path.join(prefix, 'external', 'pyopencl-2014.1-py2.7.egg-info'), prefix = 'pyopencl-2014.1-py2.7.egg-info')
+a.datas += dict_tree
+pyz = PYZ(a.pure)
+exe = EXE(pyz,
+      a.scripts,
+      a.binaries,
+      a.zipfiles,
+      a.datas,
+      name='autotune',
+      debug=False,
+      strip=None,
+      upx=True,
+      console=True )

python/autotune/pysrc/model.py

@@ -1,63 +1,20 @@
 from sklearn import tree
 from sklearn import ensemble
-
 from numpy import array, bincount, mean, std, max, argmax, min, argmin, median
-from scipy.stats import gmean
 
 
-# def random_forest(Xtr, Ytr):
+def gmean(a, axis=0, dtype=None):
-#     clf = ensemble.RandomForestRegressor(10, max_depth=7).fit(Xtr,Ytr)
+    if not isinstance(a, np.ndarray):  # if not an ndarray object attempt to convert it
-#
+        log_a = np.log(np.array(a, dtype=dtype))
-#     def predict_tree(tree, x):
+    elif dtype:  # Must change the default dtype allowing array type
-#         tree_ = tree.tree_
+        if isinstance(a,np.ma.MaskedArray):
-#         children_left = tree_.children_left
+            log_a = np.log(np.ma.asarray(a, dtype=dtype))
-#         children_right = tree_.children_right
+        else:
-#         threshold = tree_.threshold
+            log_a = np.log(np.asarray(a, dtype=dtype))
-#         feature = tree_.feature
+    else:
-#         value = tree_.value
+        log_a = np.log(a)
-#         idx = 0
+    return np.exp(log_a.mean(axis=axis))
-#         while children_left[idx]!=-1:
+    
-#             if x[0, feature[idx]] <= threshold[idx]:
-#                 idx = children_left[idx]
-#             else:
-#                 idx = children_right[idx]
-#         return value[[idx],:,:][:,:,0]
-#
-#     s = 0
-#     for e in clf.estimators_:
-#         tree_ = e.tree_
-#         children_left = tree_.children_left
-#         children_right = tree_.children_right
-#         threshold = tree_.threshold
-#         feature = tree_.feature
-#         value = tree_.value
-#         s = s + value.size + feature.size + threshold.size + children_right.size + children_left.size
-#     print s*4*1e-3
-#     return clf, clf.predict
-#
-# def train_nn(layer_sizes, XTr, YTr, XTe, YTe):
-#     optimizer = HF(open(os.devnull, 'w'), 15)
-#     optimizer.doCGBacktracking = True
-#     net = FeedforwardNeuralNet(layer_sizes, [Act.Tanh() for i in range(len(layer_sizes)-2)], Act.Linear(), 1e-5)
-#
-#     nbatch=10
-#     bsize = XTr.shape[0]/nbatch
-#     data = ((XTr[(i%nbatch)*bsize:(i%nbatch+1)*bsize,:], YTr[(i%nbatch)*bsize:(i%nbatch+1)*bsize,:]) for i in range(nbatch))
-#     data = HFDataSource(data, bsize, gradBatchSize = nbatch*bsize, curvatureBatchSize = bsize, lineSearchBatchSize =nbatch*bsize, gradBatchIsTrainingSet=True)
-#     iters = optimizer.optimize(HFModel(net), data, 300, otherPrecondDampingTerm=net.L2Cost)
-#     bestte = collections.deque([float("inf")]*5, maxlen=5)
-#     for i,w in enumerate(iters):
-#         Diffte = YTe - net.predictions(XTe).as_numpy_array()
-#         Difftr = YTr - net.predictions(XTr).as_numpy_array()
-#         Ete = np.sum(Diffte**2)
-#         Etr = np.sum(Difftr**2)
-#         bestte.append(min(min(bestte),Ete))
-#         if min(bestte)==max(bestte):
-#             print 'Final test error: ', Ete
-#             return net, net.predictions
-#         print 'Iteration %d | Test error = %.2f | Train error = %.2f'%(i, Ete, Etr)
-#     return net, net.predictions
-
 def train_model(X, Y, profiles, metric):
     print("Building the model...")
 

python/autotune/setup.py

@@ -1,33 +0,0 @@
-import os
-import sys
-from setuptools import Extension, setup
-from cx_Freeze import setup, Executable
-
-def main():
-    sys.path.append('/home/philippe/Development/ATIDLAS/build/python/pyatidlas/build/lib.linux-x86_64-2.7/')
-    sys.path.append('/home/philippe/Development/pyviennacl-dev/build/lib.linux-x86_64-2.7/')
-    sys.path.append(os.path.join('${CMAKE_CURRENT_SOURCE_DIR}','pysrc'))
-    extdir = os.path.join('${CMAKE_CURRENT_SOURCE_DIR}','external')
-    
-    buildOptions = dict(packages = ['scipy.sparse.csgraph._validation',
-                                    'scipy.special._ufuncs_cxx',
-                                    'scipy.sparse.linalg.dsolve.umfpack',
-                                    'scipy.integrate.vode',
-                                    'scipy.integrate.lsoda',
-                                    'sklearn.utils.sparsetools._graph_validation',
-                                    'sklearn.utils.lgamma'],
-                        excludes = ['matplotlib'],
-                        bin_path_includes = ['/usr/lib/x86_64-linux-gnu/'],
-                        include_files = [(os.path.abspath(os.path.join(extdir, x)), x) for x in os.listdir(extdir)])
-    base = 'Console'
-    executables = [
-        Executable(os.path.join('${CMAKE_CURRENT_SOURCE_DIR}','pysrc','autotune.py'), base=base)
-    ]
-    setup(name='atidlas-tune',
-          version = '1.0',
-          description = 'Auto-tuner for ATIDLAS',
-          options = dict(build_exe = buildOptions),
-          executables = executables)
-
-if __name__ == "__main__":
-    main()