Tuner: Removed Sklearn + scipy dependency

python/setup.py

@@ -112,7 +112,8 @@ def main():
     library_dirs = [config['lib'] for config in [opencl_config, cuda_config] if config is not None]
 
     #Include directories
-    include =' src/include src/lib/external'.split() + ['external/boost/', 'external/boost/boost/', os.path.join(find_module("numpy")[1], "core", "include")]
+    numpy_include = os.path.join(find_module("numpy")[1], "core", "include")
+    include =' src/include src/lib/external'.split() + ['external/boost/', 'external/boost/boost/', numpy_include]
 
     #Source files
     src =  'src/lib/symbolic/preset.cpp src/lib/symbolic/execute.cpp src/lib/symbolic/io.cpp src/lib/symbolic/expression.cpp src/lib/array.cpp src/lib/value_scalar.cpp src/lib/driver/backend.cpp src/lib/driver/device.cpp src/lib/driver/kernel.cpp src/lib/driver/buffer.cpp src/lib/driver/platform.cpp src/lib/driver/check.cpp src/lib/driver/program.cpp src/lib/driver/command_queue.cpp src/lib/driver/program_cache.cpp src/lib/driver/context.cpp src/lib/driver/event.cpp src/lib/driver/ndrange.cpp src/lib/driver/handle.cpp src/lib/exception/unknown_datatype.cpp src/lib/exception/operation_not_supported.cpp src/lib/profiles/presets.cpp src/lib/profiles/profiles.cpp src/lib/profiles/predictors/random_forest.cpp src/lib/kernels/templates/gemv.cpp src/lib/kernels/templates/axpy.cpp src/lib/kernels/templates/gemm.cpp src/lib/kernels/templates/ger.cpp src/lib/kernels/templates/dot.cpp src/lib/kernels/templates/base.cpp src/lib/kernels/mapped_object.cpp src/lib/kernels/stream.cpp src/lib/kernels/parse.cpp src/lib/kernels/keywords.cpp src/lib/kernels/binder.cpp src/lib/wrap/clBLAS.cpp '.split() + [os.path.join('src', 'bind', sf)  for sf in ['_isaac.cpp', 'core.cpp', 'driver.cpp', 'kernels.cpp', 'exceptions.cpp']]
@@ -129,16 +130,10 @@ def main():
     src= [f for f in src  if not f.endswith("once_atomic.cpp")]
 
     #Setup
-    setup(
-                name='isaac',
-                version='1.0',
-                description="Input-specific architecture-aware computations",
-                author='Philippe Tillet',
-                author_email='ptillet@g.harvard.edu',
-                license='MPL 2.0',
-                packages=["isaac"],
-                ext_package="isaac",
-                ext_modules=[Extension(
+    extensions = []
+    
+    #ISAAC
+    extensions += [Extension(
                     '_isaac',src,
                     extra_compile_args= backend_defines + ['-std=c++11', '-Wno-unused-function', '-Wno-unused-local-typedefs',  '-Wno-sign-compare', '-Wno-attributes', '-DBOOST_PYTHON_SOURCE '],
                     extra_link_args=['-Wl,-soname=_isaac.so'],
@@ -146,7 +141,24 @@ def main():
                     include_dirs=include,
                     library_dirs=library_dirs,
                     libraries=libraries
-                )],
+                    )]
+    
+    #External
+    extensions += [Extension('external._tree',
+                            ['isaac/external/_tree.c'], 
+                           extra_compile_args = ["-O3", "-Wall"], 
+                           include_dirs = ['.', numpy_include])]
+    
+    setup(
+                name='isaac',
+                version='1.0',
+                description="Input-specific architecture-aware computations",
+                author='Philippe Tillet',
+                author_email='ptillet@g.harvard.edu',
+                license='MPL 2.0',
+                packages=["isaac", "isaac.external"],
+                ext_package="isaac",
+                ext_modules=extensions,
                 cmdclass={'build_py': build_py, 'build_ext': build_ext_subclass},
                 classifiers=[
                     'Environment :: Console',

tune/model.py

@@ -1,4 +1,4 @@
-from sklearn import ensemble
+from isaac.external.forest import RandomForestRegressor
 import numpy as np
 
 def gmean(a, axis=0, dtype=None):
@@ -39,7 +39,7 @@ def train(X, Y, profiles):
     nrmses = {}
     for N in range(1,min(M+1,20)):
         for depth in range(1,min(M+1,20)):
-            clf = ensemble.RandomForestRegressor(N, max_depth=depth).fit(XTr, YTr)
+            clf = RandomForestRegressor(N, max_depth=depth).fit(XTr, YTr)
             t = np.argmax(clf.predict(XCv), axis = 1)
             y = np.array([YCv[i,t[i]] for i in range(t.size)])
             ground = np.max(YCv[:,:], axis=1)

tune/tune.py

@@ -3,8 +3,8 @@ from math import log, isinf
 from itertools import chain, product
 from numpy import argsort, argmax
 from operator import mul
-from sklearn import ensemble
 import isaac as sc
+from isaac.external.forest import RandomForestRegressor
 import optimize, tools, model
 
 from json import encoder
@@ -82,8 +82,8 @@ def tune(device, operation, json_path):
             if nparams==1:
                 predicted = profiles[0]
             else:
-                clf = ensemble.RandomForestRegressor(min(10, idx+1), max_depth=min(10, idx+1)).fit(X, Y)
-                #clf, nrmse = profile.train(X, Y, profiles)
+                clf = RandomForestRegressor(min(10, idx+1), max_depth=min(10, idx+1)).fit(X, Y)
+                #clf, nrmse = model.train(X, Y, profiles)
                 predperf = clf.predict(x)[0]
                 best = (-predperf).argsort()[:5]
                 perf = [performance(x, tools.benchmark(operation, profiles[b], tree)) for b in best]
@@ -130,7 +130,7 @@ def tune(device, operation, json_path):
     json_data[operation_name]['float32'] = {}
     D = json_data[operation_name]['float32']
     if len(profiles) > 1:
-        clf, nrmse = profile.train(X, Y, profiles)
+        clf, nrmse = model.train(X, Y, profiles)
         D['predictor'] = [{'children_left': e.tree_.children_left.tolist(),
                             'children_right': e.tree_.children_right.tolist(),
                             'threshold': e.tree_.threshold.astype('float64').tolist(),