Remove shape_int

README.md

@@ -315,7 +315,6 @@ randn(ref)
 ### Shaping
 ```
 shape(a)
-shape_int(a)
 rank(a)
 length(a) (alias: size)
 isscalar(a)

benchmark.py

@@ -0,0 +1,49 @@
+from time import time
+
+import autograd.numpy as np
+
+import lab as B
+
+n = 20
+m = 1
+
+t = np.float64
+eps = B.cast(t, B.epsilon)
+
+
+def f1(x):
+    dists2 = (x - B.transpose(x)) ** 2
+    K = B.exp(-0.5 * dists2)
+    K = K + B.epsilon * B.eye(t, n)
+    L = B.cholesky(K)
+    return B.matmul(L, B.ones(t, n, m))
+
+
+def f2(x):
+    dists2 = (x - np.transpose(x)) ** 2
+    K = np.exp(-0.5 * dists2)
+    K = K + B.epsilon * np.eye(n, dtype=t)
+    L = np.linalg.cholesky(K)
+    return np.matmul(L, np.ones((n, m)))
+
+
+# Perform computation once.
+x = np.linspace(0, 1, n, dtype=t)[:, None]
+f1(x)
+f2(x)
+
+its = 10000
+
+s = time()
+for _ in range(its):
+    z = f2(x)
+us_native = (time() - s) / its * 1e6
+
+s = time()
+for _ in range(its):
+    z = f1(x)
+us_lab = (time() - s) / its * 1e6
+
+print('Overhead: {:.1f} us / {:.1f} %'
+      ''.format(us_lab - us_native,
+                100 * (us_lab / us_native - 1)))

lab/autograd/shaping.py

@@ -45,7 +45,7 @@ def vec_to_tril(a):
 def tril_to_vec(a):
     if B.rank(a) != 2:
         raise ValueError('Ianput must be rank 2.')
-    n, m = B.shape_int(a)
+    n, m = B.shape(a)
     if n != m:
         raise ValueError('Ianput must be square.')
     return a[anp.tril_indices(n)]

lab/shaping.py

@@ -7,7 +7,6 @@
 from .util import abstract
 
 __all__ = ['shape',
-           'shape_int',
            'rank',
            'length', 'size',
            'isscalar',
@@ -43,19 +42,6 @@ def shape(a):  # pragma: no cover
         return ()
 
 
-@dispatch(Numeric)
-def shape_int(a):  # pragma: no cover
-    """Get the shape of a tensor as a tuple of integers.
-
-    Args:
-        a (tensor): Tensor.
-
-    Returns:
-        tuple: Shape of `a` as a tuple of integers.
-    """
-    return shape(a)
-
-
 @dispatch(Numeric)
 def rank(a):  # pragma: no cover
     """Get the shape of a tensor.
@@ -173,7 +159,7 @@ def flatten(a):  # pragma: no cover
 
 
 def _vec_to_tril_shape(a):
-    n = shape_int(a)[0]
+    n = int(shape(a)[0])  # Dimensions are not necessarily integers!
     return int(((1 + 8 * n) ** .5 - 1) / 2)
 
 

lab/tensorflow/linear_algebra.py

@@ -42,8 +42,8 @@ def trace(a, axis1=0, axis2=1):
 
 @dispatch(TFNumeric, TFNumeric)
 def kron(a, b):
-    shape_a = B.shape_int(a)
-    shape_b = B.shape_int(b)
+    shape_a = B.shape(a)
+    shape_b = B.shape(b)
 
     # Check that ranks are equal.
     if len(shape_a) != len(shape_b):

lab/tensorflow/shaping.py

@@ -12,11 +12,6 @@
 __all__ = []
 
 
-@dispatch(TFNumeric)
-def shape_int(a):
-    return tuple(B.shape(a).as_list())
-
-
 @dispatch(TFNumeric)
 def length(a):
     return tf.size(a)
@@ -56,10 +51,10 @@ def vec_to_tril(a):
 def tril_to_vec(a):
     if B.rank(a) != 2:
         raise ValueError('Input must be rank 2.')
-    n, m = shape_int(a)
+    n, m = B.shape(a)
     if n != m:
         raise ValueError('Input must be square.')
-    return tf.gather_nd(a, list(zip(*np.tril_indices(n))))
+    return tf.gather_nd(a, list(zip(*np.tril_indices(int(n)))))
 
 
 @dispatch([TFNumeric])

lab/torch/linear_algebra.py

@@ -41,8 +41,8 @@ def trace(a, axis1=0, axis2=1):
 
 @dispatch(TorchNumeric, TorchNumeric)
 def kron(a, b):
-    shape_a = B.shape_int(a)
-    shape_b = B.shape_int(b)
+    shape_a = B.shape(a)
+    shape_b = B.shape(b)
 
     # Check that ranks are equal.
     if len(shape_a) != len(shape_b):

lab/torch/shaping.py

@@ -12,11 +12,6 @@
 __all__ = []
 
 
-@dispatch(TorchNumeric)
-def shape_int(a):
-    return tuple(B.shape(a))
-
-
 @dispatch(TorchNumeric)
 def length(a):
     return a.numel()
@@ -51,7 +46,7 @@ def vec_to_tril(a):
 def tril_to_vec(a):
     if B.rank(a) != 2:
         raise ValueError('Input must be rank 2.')
-    n, m = shape_int(a)
+    n, m = B.shape(a)
     if n != m:
         raise ValueError('Input must be square.')
     return a[np.tril_indices(n)]

tests/test_generic.py

@@ -37,8 +37,8 @@ def test_zeros_ones_eye():
                Value(3))
 
         # Check shape of calls.
-        yield eq, B.shape_int(f(2)), (2, 2) if f is B.eye else (2,)
-        yield eq, B.shape_int(f(2, 3)), (2, 3)
+        yield eq, B.shape(f(2)), (2, 2) if f is B.eye else (2,)
+        yield eq, B.shape(f(2, 3)), (2, 3)
 
         # Check shape type of calls.
         yield eq, B.dtype(f(2)), B.default_dtype
@@ -50,9 +50,9 @@ def test_zeros_ones_eye():
             ref = B.randn(t1, 4, 5)
 
             # Check shape of calls.
-            yield eq, B.shape_int(f(t2, 2)), (2, 2) if f is B.eye else (2,)
-            yield eq, B.shape_int(f(t2, 2, 3)), (2, 3)
-            yield eq, B.shape_int(f(ref)), (4, 5)
+            yield eq, B.shape(f(t2, 2)), (2, 2) if f is B.eye else (2,)
+            yield eq, B.shape(f(t2, 2, 3)), (2, 3)
+            yield eq, B.shape(f(ref)), (4, 5)
 
             # Check shape type of calls.
             yield eq, B.dtype(f(t2, 2)), t2

tests/test_linear_algebra.py

@@ -184,7 +184,7 @@ def test_pw_1d():
 
     # Check that we can feed both rank 1 and rank 2 tensors.
     for f, g in product(*([[lambda x: x, lambda x: x[:, 0]]] * 2)):
-        
+
         yield allclose, B.pw_dists2(f(a), g(b)), np.abs(a - b.T) ** 2
         yield allclose, B.pw_dists2(f(a)), np.abs(a - a.T) ** 2
         yield allclose, B.pw_dists(f(a), g(b)), np.abs(a - b.T)

tests/test_random.py

@@ -29,29 +29,29 @@ def test_random_generators():
     for f in [B.rand, B.randn]:
         # Test without specifying data type.
         yield deq, B.dtype(f()), B.default_dtype
-        yield allclose, B.shape_int(f()), (), False
+        yield eq, B.shape(f()), ()
         yield deq, B.dtype(f(2)), B.default_dtype
-        yield allclose, B.shape_int(f(2)), (2,)
+        yield allclose, B.shape(f(2)), (2,)
         yield deq, B.dtype(f(2, 3)), B.default_dtype
-        yield allclose, B.shape_int(f(2, 3)), (2, 3)
+        yield eq, B.shape(f(2, 3)), (2, 3)
 
         # Test with specifying data type.
         for t in [np.float32, tf.float32, torch.float32]:
             # Test direct specification.
             yield deq, B.dtype(f(t)), t
-            yield allclose, B.shape_int(f(t)), (), False
+            yield eq, B.shape(f(t)), ()
             yield deq, B.dtype(f(t, 2)), t
-            yield allclose, B.shape_int(f(t, 2)), (2,), False
+            yield eq, B.shape(f(t, 2)), (2,)
             yield deq, B.dtype(f(t, 2, 3)), t
-            yield allclose, B.shape_int(f(t, 2, 3)), (2, 3), False
+            yield eq, B.shape(f(t, 2, 3)), (2, 3)
 
             # Test reference specification.
             yield deq, B.dtype(f(f(t))), t
-            yield allclose, B.shape_int(f(f())), (), False
+            yield eq, B.shape(f(f())), ()
             yield deq, B.dtype(f(f(t, 2))), t
-            yield allclose, B.shape_int(f(f(t, 2))), (2,), False
+            yield eq, B.shape(f(f(t, 2))), (2,)
             yield deq, B.dtype(f(f(t, 2, 3))), t
-            yield allclose, B.shape_int(f(f(t, 2, 3))), (2, 3), False
+            yield eq, B.shape(f(f(t, 2, 3))), (2, 3)
 
 
 def test_conversion_warnings():

tests/test_shaping.py

@@ -10,7 +10,7 @@
 
 
 def test_sizing():
-    for f in [B.shape, B.shape_int, B.rank, B.length, B.size]:
+    for f in [B.shape, B.rank, B.length, B.size]:
         yield check_function, f, (Tensor(),), {}, False
         yield check_function, f, (Tensor(3, ),), {}, False
         yield check_function, f, (Tensor(3, 4),), {}, False