Merge pull request #955 from nouiz/small

.travis.yml

@@ -11,7 +11,7 @@ before_install:
   - sudo apt-get install -qq libatlas3gf-base libatlas-dev
 install:
 #If we don't install numpy before SciPy 0.10.1, the SciPy installations fails.
-  - "pip install -q numpy --use-mirrors"
+  - "pip install -q numpy==1.4.1 --use-mirrors"
 #We can't install SciPy as there is no BLAS installed.
   - "pip install . --no-deps --use-mirrors"
 # command to run tests

doc/tutorial/faq.txt

@@ -42,6 +42,25 @@ space to allow to reuse them during the next call to the same Theano
 function if they are of the good shape. The shape could change if the
 shape of the inputs change.
 
+Faster Small Theano function
+----------------------------
+
+.. note::
+
+   For Theano 0.6 and up.
+
+For Theano function that don't do much work like a regular logistic
+regression, the overhead of checking the input can be significant. You
+can disable it by setting f.trust_input to True to remove this
+check. Make sure you pass argument as what you said when compiling the
+Theano function.
+
+Also for small Theano function, you can remove more python overhead by
+making a Theano function that don't take any inputs. You can use shared
+variable to help you. Then you can call it like this: ``f.fn()`` or
+``f.fn(n_calls=N)`` to speed up. In the last case, only the last
+function output is returned.
+
 Related Projects
 ----------------
 

theano/scan_module/scan_perform_ext.py

@@ -51,7 +51,7 @@ except ImportError:
             loc = os.path.join(config.compiledir, dirname)
             if not os.path.exists(loc):
                 os.mkdir(loc)
-            preargs = ['-pthread', '-fwrapv', '-O2', '-fno-strict-aliasing']
+            preargs = ['-fwrapv', '-O2', '-fno-strict-aliasing']
             preargs += cmodule.GCC_compiler.compile_args()
             cmodule.GCC_compiler.compile_str(dirname, code, location=loc,
                                              preargs=preargs)

theano/tensor/extra_ops.py

@@ -91,6 +91,12 @@ class BinCountOp(theano.Op):
 
     def __init__(self, minlength=None):
         self.minlength = minlength
+        if minlength is not None:
+            numpy_ver = [int(n) for n in numpy.__version__.split('.')[:2]]
+            if not bool(numpy_ver >= [1, 6]):
+                raise NotImplementedError(
+                    "BinCountOp with minlength attribute"
+                    " need NumPy 1.6 or higher.")
 
     def __eq__(self, other):
         return (type(self) == type(other) and
@@ -145,8 +151,11 @@ class BinCountOp(theano.Op):
 
         if weights is not None and weights.shape != x.shape:
             raise TypeError("All inputs must have the same shape.")
-
-        z[0] = np.bincount(x, weights=weights, minlength=self.minlength)
+        #Needed for numpy 1.4.1 compatibility
+        if self.minlength:
+            z[0] = np.bincount(x, weights=weights, minlength=self.minlength)
+        else:
+            z[0] = np.bincount(x, weights=weights)
 
     def grad(self, inputs, outputs_gradients):
         output = self(*inputs)

theano/tensor/tests/test_extra_ops.py

@@ -10,6 +10,9 @@ from theano import tensor as T
 from theano import config, tensor, function
 
 
+numpy_ver = [int(n) for n in numpy.__version__.split('.')[:2]]
+numpy_16 = bool(numpy_ver >= [1, 6])
+
 class TestBinCountOp(utt.InferShapeTester):
     def setUp(self):
         super(TestBinCountOp, self).setUp()
@@ -39,12 +42,14 @@ class TestBinCountOp(utt.InferShapeTester):
 
                 f1 = theano.function([x], bincount(x))
                 f2 = theano.function([x, w], bincount(x, weights=w))
-                f3 = theano.function([x], bincount(x, minlength=23))
-                f4 = theano.function([x], bincount(x, minlength=5))
 
                 assert (np.bincount(a) == f1(a)).all()
                 assert np.allclose(np.bincount(a, weights=weights),
                                    f2(a, weights))
+                if not numpy_16:
+                    continue
+                f3 = theano.function([x], bincount(x, minlength=23))
+                f4 = theano.function([x], bincount(x, minlength=5))
                 assert (np.bincount(a, minlength=23) == f3(a)).all()
                 assert (np.bincount(a, minlength=5) == f4(a)).all()
 
@@ -79,6 +84,8 @@ class TestBinCountOp(utt.InferShapeTester):
                             50, size=(25,)).astype(dtype)],
                         self.op_class)
 
+                if not numpy_16:
+                    continue
                 self._compile_and_check(
                         [x],
                         [bincount(x, minlength=60)],

theano/tensor/tests/test_sort.py

 import unittest
-from numpy.testing import assert_allclose
 from theano.tests import unittest_tools as utt
 
 import numpy as np
@@ -22,7 +21,7 @@ class test_sort(unittest.TestCase):
         a = tensor.dmatrix()
         w = sort(a)
         f = theano.function([a], w)
-        assert_allclose(f(self.m_val), np.sort(self.m_val))
+        assert np.allclose(f(self.m_val), np.sort(self.m_val))
 
     def test2(self):
         a = tensor.dmatrix()
@@ -32,7 +31,7 @@ class test_sort(unittest.TestCase):
         for axis_val in 0, 1:
             gv = f(self.m_val, axis_val)
             gt = np.sort(self.m_val, axis_val)
-            assert_allclose(gv, gt)
+            assert np.allclose(gv, gt)
 
     def test3(self):
         a = tensor.dvector()
@@ -40,7 +39,7 @@ class test_sort(unittest.TestCase):
         f = theano.function([a], w2)
         gv = f(self.v_val)
         gt = np.sort(self.v_val)
-        assert_allclose(gv, gt)
+        assert np.allclose(gv, gt)
 
     def test4(self):
         a = tensor.dmatrix()
@@ -50,7 +49,7 @@ class test_sort(unittest.TestCase):
         for axis_val in 0, 1:
             gv = f(self.m_val, axis_val)
             gt = np.sort(self.m_val, axis_val)
-            assert_allclose(gv, gt)
+            assert np.allclose(gv, gt)
 
     def test5(self):
         a1 = SortOp("mergesort", [])
@@ -67,7 +66,7 @@ class test_sort(unittest.TestCase):
         f = theano.function([a], l)
         gv = f(self.m_val)
         gt = np.sort(self.m_val, None)
-        assert_allclose(gv, gt)
+        assert np.allclose(gv, gt)
 
 
 class TensorInferShapeTester(utt.InferShapeTester):
@@ -97,7 +96,7 @@ def test_argsort():
     f = theano.function([a], w)
     gv = f(m_val)
     gt = np.argsort(m_val)
-    assert_allclose(gv, gt)
+    assert np.allclose(gv, gt)
 
     #Example 2
     a = tensor.dmatrix()
@@ -107,7 +106,7 @@ def test_argsort():
     for axis_val in 0, 1:
         gv = f(m_val, axis_val)
         gt = np.argsort(m_val, axis_val)
-        assert_allclose(gv, gt)
+        assert np.allclose(gv, gt)
 
     #Example 3
     a = tensor.dvector()
@@ -115,7 +114,7 @@ def test_argsort():
     f = theano.function([a], w2)
     gv = f(v_val)
     gt = np.argsort(v_val)
-    assert_allclose(gv, gt)
+    assert np.allclose(gv, gt)
 
     #Example 4
     a = tensor.dmatrix()
@@ -125,7 +124,7 @@ def test_argsort():
     for axis_val in 0, 1:
         gv = f(m_val, axis_val)
         gt = np.argsort(m_val, axis_val)
-        assert_allclose(gv, gt)
+        assert np.allclose(gv, gt)
 
     #Example 5
     a = tensor.dmatrix()
@@ -143,4 +142,4 @@ def test_argsort():
     f = theano.function([a], w2)
     gv = f(m_val)
     gt = np.argsort(m_val, None)
-    assert_allclose(gv, gt)
+    assert np.allclose(gv, gt)