Merge pull request #2159 from nouiz/permutation

doc/library/tensor/nlinalg.txt

@@ -11,6 +11,10 @@
    :synopsis: Linear Algebra Ops Using Numpy
 .. moduleauthor:: LISA
 
+.. note::
+
+   This module is not imported by default. You need to import it to use it.
+
 API
 ===
 

doc/library/tensor/slinalg.txt

@@ -11,6 +11,10 @@
    :synopsis: Linear Algebra Ops Using Scipy
 .. moduleauthor:: LISA
 
+.. note::
+
+   This module is not imported by default. You need to import it to use it.
+
 API
 ===
 

theano/configparser.py

@@ -4,6 +4,7 @@
 
 import logging
 import os
+import shlex
 import sys
 import warnings
 
@@ -39,7 +40,10 @@ def parse_config_string(config_string, issue_warnings=True):
     Parses a config string (comma-separated key=value components) into a dict.
     """
     config_dict = {}
-    for kv_pair in config_string.split(','):
+    my_splitter = shlex.shlex(config_string, posix=True)
+    my_splitter.whitespace = ','
+    my_splitter.whitespace_split = True
+    for kv_pair in my_splitter:
         kv_pair = kv_pair.strip()
         if not kv_pair:
             continue

theano/gof/type.py

@@ -558,7 +558,7 @@ if (%(name)s == NULL) {
   py_%(name)s = PyCapsule_New((void *)%(name)s, NULL,
                               _py3_destructor);
   if (py_%(name)s != NULL) {
-    if (PyCaspule_SetContext(py_%(name)s, (void *)%(freefunc)s) != 0) {
+    if (PyCapsule_SetContext(py_%(name)s, (void *)%(freefunc)s) != 0) {
       /* This won't trigger a call to freefunc since it could not be
          set. The error case below will do it. */
       Py_DECREF(py_%(name)s);

theano/sandbox/cuda/opt.py

@@ -649,6 +649,10 @@ def local_gpu_careduce(node):
         if isinstance(node.op.scalar_op, (scal.Add, scal.Mul,
                                           scal.Maximum, scal.Minimum)):
             x, = node.inputs
+            # Otherwise, is some corner case, we will try to move it
+            # to the GPU later and this cause not wanted user warning.
+            if x.dtype != 'float32':
+                return
             replace = False
             if x.owner and isinstance(x.owner.op, HostFromGpu):
                 replace = True
@@ -666,7 +670,7 @@ def local_gpu_careduce(node):
                 # don't introduce a bigger transfer. It is hard to
                 # know if after all optimization we will do the bigger
                 # transfer or not. I'm guessing an heuristic to find
-                # that. I suppose that if the input of the recution is
+                # that. I suppose that if the input of the reduction is
                 # generated by an op that we can in some cases move to
                 # the GPU, that we will move it. If some CPU ops are
                 # supported only in some cases on the GPU, this will

theano/tensor/basic.py

@@ -632,22 +632,38 @@ def get_scalar_constant_value(orig_v, elemwise=True):
                 # test_sharedvar.py:test_shared_options.test_specify_shape_partial
                 if (v.owner.inputs[0].owner and
                     isinstance(v.owner.inputs[0].owner.op, Join) and
+                    len(v.owner.op.idx_list) == 1):
                     # Ensure the Join is joining only scalar variables (so that
                     # the constant value can be found at the same index as the one
                     # used in the sub-tensor).
-                    python_all(var.ndim == 0 for var in
-                               v.owner.inputs[0].owner.inputs) and
-                    len(v.owner.op.idx_list) == 1):
-
-                    idx = v.owner.op.idx_list[0]
-                    if isinstance(idx, gof.Type):
-                        idx = get_scalar_constant_value(v.owner.inputs[1])
-                    # Note the '+ 1' is because the first argument to Join is the
-                    # axis.
-                    ret = v.owner.inputs[0].owner.inputs[idx + 1]
-                    ret = get_scalar_constant_value(ret)
-                    # join can cast implicitly its input in some case.
-                    return theano._asarray(ret, dtype=v.type.dtype)
+                    if python_all(var.ndim == 0 for var in
+                                  v.owner.inputs[0].owner.inputs[1:]):
+                        idx = v.owner.op.idx_list[0]
+                        if isinstance(idx, gof.Type):
+                            idx = get_scalar_constant_value(v.owner.inputs[1])
+                        # Note the '+ 1' is because the first argument to Join is the
+                        # axis.
+                        ret = v.owner.inputs[0].owner.inputs[idx + 1]
+                        ret = get_scalar_constant_value(ret)
+                        # join can cast implicitly its input in some case.
+                        return theano._asarray(ret, dtype=v.type.dtype)
+                    if python_all(var.ndim == 1 for var in
+                                  v.owner.inputs[0].owner.inputs[1:]):
+                        idx = v.owner.op.idx_list[0]
+                        if isinstance(idx, gof.Type):
+                            idx = get_scalar_constant_value(v.owner.inputs[1])
+                        try:
+                            #TODO: assert joined axis is 0.
+                            length = 0
+                            for joined in v.owner.inputs[0].owner.inputs[1:]:
+                                ll = get_vector_length(joined)
+                                if idx < length + ll:
+                                    return get_scalar_constant_value(joined[idx-length])
+                                length += ll
+                        except TypeError:
+                            pass
+                        except ValueError:
+                            pass
 
                 elif (v.owner.inputs[0].owner and
                     isinstance(v.owner.inputs[0].owner.op,
@@ -2035,6 +2051,8 @@ class Nonzero(gof.Op):
         flattened input array.
 
     """
+    __props__ = ()
+
     def make_node(self, a):
         a = as_tensor_variable(a)
         if a.ndim == 0:
@@ -3663,6 +3681,15 @@ def get_vector_length(v):
         return len(v.owner.inputs)
     if v.owner and isinstance(v.owner.op, Shape):
         return v.owner.inputs[0].type.ndim
+    # If we take this slice: var[:0], we know it will have 0 elements.
+    if (v.owner and
+        isinstance(v.owner.op, theano.tensor.subtensor.Subtensor) and
+        isinstance(v.owner.op.idx_list[0], slice) and
+        v.owner.op.idx_list[0].start in [None, 0]):
+        stop = theano.tensor.subtensor.get_idx_list(
+            v.owner.inputs, v.owner.op.idx_list)[0].stop
+        if extract_constant(stop) == 0:
+            return 0
     raise ValueError("length not known")
 
 

theano/tensor/nnet/conv.py

@@ -58,7 +58,8 @@ def conv2d(input, filters, image_shape=None, filter_shape=None,
                  of shape: image_shape + filter_shape - 1
 
     :type subsample: tuple of len 2
-    :param subsample: factor by which to subsample the output
+    :param subsample: factor by which to subsample the output.
+                      Also called strides elsewhere.
 
     :type image_shape: None, tuple/list of len 4 of int or Constant variable
     :param image_shape: The shape of the input parameter.

theano/tensor/raw_random.py

@@ -183,7 +183,11 @@ class RandomFunction(gof.Op):
         draw.
 
         """
-        shape = tensor.as_tensor_variable(shape, ndim=1)
+        shape_ = tensor.as_tensor_variable(shape, ndim=1)
+        if shape == ():
+            shape = shape_.astype('int32')
+        else:
+            shape = shape_
         assert shape.type.ndim == 1
         assert (shape.type.dtype == 'int64') or (shape.type.dtype == 'int32')
         if not isinstance(r.type, RandomStateType):
@@ -700,7 +704,15 @@ def permutation(random_state, size=None, n=1, ndim=None, dtype='int64'):
     :note:
         Note that the output will then be of dimension ndim+1.
     """
-    ndim, size, bcast = _infer_ndim_bcast(ndim, size)
+    if size is None or size == ():
+        if not(ndim is None or ndim == 1):
+            raise TypeError(
+                "You asked for just one permutation but asked for more then 1 dimensions.")
+        ndim = 1
+        size = ()
+        bcast = ()
+    else:
+        ndim, size, bcast = _infer_ndim_bcast(ndim, size)
     #print "NDIM", ndim, size
     op = RandomFunction(permutation_helper,
             tensor.TensorType(dtype=dtype, broadcastable=bcast + (False,)),

theano/tensor/subtensor.py

@@ -2097,7 +2097,12 @@ def take(a, indices, axis=None, mode='raise'):
         shape = indices.shape
         ndim = indices.ndim
     else:
-        shape = theano.tensor.concatenate(
-            [a.shape[:axis], indices.shape, a.shape[axis + 1:]])
+        # If axis is 0, don't generate a useless concatenation.
+        if axis == 0:
+            shape = theano.tensor.concatenate(
+                [indices.shape, a.shape[axis + 1:]])
+        else:
+            shape = theano.tensor.concatenate(
+                [a.shape[:axis], indices.shape, a.shape[axis + 1:]])
         ndim = a.ndim + indices.ndim - 1
     return take(a, indices.flatten(), axis, mode).reshape(shape, ndim)

theano/tensor/tests/test_opt.py

@@ -3889,7 +3889,8 @@ class T_local_reduce(unittest.TestCase):
             assert isinstance(topo[-1].op, T.Elemwise), out
 
         # Test different axis for the join and the reduction
-        A = theano.shared(numpy.array([1, 2, 3, 4, 5]))
+        # We must force the dtype, of otherwise, this tests will fail in 32 bit system
+        A = theano.shared(numpy.array([1, 2, 3, 4, 5], dtype='int64'))
 
         f = theano.function([], T.sum(T.stack(A, A), axis=0), mode=self.mode)
         assert numpy.allclose(f(), [2, 4, 6, 8, 10])

theano/tensor/tests/test_raw_random.py

@@ -230,7 +230,7 @@ class T_random_function(utt.InferShapeTester):
         rng_R = random_state_type()
 
         # No shape, no args -> TypeError
-        self.assertRaises(TypeError, permutation, rng_R, size=None, ndim=2)
+        self.assertRaises(TypeError, poisson, rng_R, size=None, ndim=2)
 
     def test_random_function_ndim_added(self):
         """Test that random_function helper function accepts ndim_added as
@@ -561,6 +561,19 @@ class T_random_function(utt.InferShapeTester):
         self.assertTrue(numpy.all(val0 == numpy_val0))
         self.assertTrue(numpy.all(val1 == numpy_val1))
 
+        # Test that we can generate a list: have size=None or ().
+        for ndim in [1, None]:
+            post_r, out = permutation(rng_R, n=10, size=None, ndim=ndim)
+            inp = compile.In(rng_R,
+                             value=numpy.random.RandomState(utt.fetch_seed()),
+                             update=post_r, mutable=True)
+            f = theano.function([inp], out)
+            o = f()
+            assert o.shape == (10,)
+            assert (numpy.sort(o) == numpy.arange(10)).all()
+        # Wrong number of dimensions asked
+        self.assertRaises(TypeError, permutation, rng_R, size=None, ndim=2)
+
     def test_multinomial(self):
         """Test that raw_random.multinomial generates the same
         results as numpy."""

theano/tensor/tests/test_subtensor.py

@@ -27,7 +27,7 @@ from theano.tensor.subtensor import (inc_subtensor, set_subtensor,
 from theano.tensor import (as_tensor_variable, _shared,
                            NotScalarConstantError,
                            fscalar, iscalar, dscalar, cscalar,
-                           vector, dvector, fvector, lvector,
+                           vector, dvector, fvector, lvector, lrow,
                            fmatrix, dmatrix, lmatrix, matrix,
                            ctensor3, dtensor4)
 from theano.tensor.tests.test_basic import rand, randint_ranged, inplace_func
@@ -1140,6 +1140,7 @@ class TestAdvancedSubtensor(unittest.TestCase):
         self.ix1 = lvector()  # advanced 1d query
         self.ix12 = lvector()
         self.ix2 = lmatrix()
+        self.ixr = lrow()
 
     def eval_output_and_check(self, t):
         f = inplace_func([], t, mode=self.mode)
@@ -1164,6 +1165,11 @@ class TestAdvancedSubtensor(unittest.TestCase):
         assert a.broadcastable == self.ix2.broadcastable, (
                 a.broadcastable, self.ix2.broadcastable)
 
+    def test_index_into_mat_w_row(self):
+        a = self.m[self.ixr]
+        assert a.dtype == self.m.dtype, (a.dtype, self.m.dtype)
+        assert a.broadcastable == (True, False, False)
+
     def test_index_w_int_and_vec(self):
         # like test_ok_list, but with a single index on the first one
         # data has to have at least 2 dimensions