def test_class(self): class Test(object): def f(self, x): return x + 100.0 @classmethod def g(cls, x): return x + 100.0 @staticmethod def h(x): return x + 100.0 test = Test() t = Tracer() x = 1.0 o = t.trace(test.f, x) f = t.compile_function(x, o) assert(f(2.0) == 102.0) o = t.trace(test.g, x) f = t.compile_function(x, o) assert(f(2.0) == 102.0) o = t.trace(test.h, x) f = t.compile_function(x, o) assert(f(2.0) == 102.0)
def test_access_attribute(self): class Test(object): def __init__(self): self.x = np.arange(5.) - 10.0 def getx(self): return self.x test = Test() def f(x): return np.dot(x, test.x) x = np.arange(5.) t = Tracer() o = t.trace(f, x) g = t.compile_gradient(x, o, wrt=test.x) self.assertTrue(np.allclose(g(x), x))
def test_readme(self): """ the original README example""" # -- a vanilla function def f1(x): return x + 2 # -- a function referencing a global variable y = np.random.random(10) def f2(x): return x * y # -- a function with a local variable def f3(x): z = tag(np.ones(10), 'local_var') return (x + z) ** 2 # -- create a general symbolic tracer and apply # it to the three functions x = np.random.random(10) tracer = Tracer() out1 = tracer.trace(f1, x) out2 = tracer.trace(f2, out1) out3 = tracer.trace(f3, out2) # -- compile a function representing f(x, y, z) = out3 new_fn = tracer.compile_function(inputs=[x, y, 'local_var'], outputs=out3) # -- compile the gradient of f(x) = out3, with respect to y fn_grad = tracer.compile_gradient(inputs=x, outputs=out3, wrt=y, reduction=theano.tensor.sum) assert fn_grad # to stop flake error self.assertTrue(np.allclose(new_fn(x, y, np.ones(10)), f3(f2(f1(x)))))
def test_multiple_trace(self): def f1(x): return x + 1.0 def f2(x): return x * 2.0 def f3(x): return x ** 2 t = Tracer() x = np.random.random((3, 4)) o1 = t.trace(f1, x) o2 = t.trace(f2, o1) o3 = t.trace(f3, o2) # test function f = t.compile_function(x, o3) self.assertTrue(np.allclose(f(x), f3(f2(f1(x))))) # test gradient o4 = t.trace(lambda x: x.sum(), o3) g = t.compile_gradient(x, o4, wrt=x) self.assertTrue(np.allclose(g(x), 8 * (x+1)))