class ContactPointsOp(gof.COp):
num_inputs = 7
params_type = gof.ParamsType(tol=theano.scalar.float64, )
__props__ = ("tol", )
func_file = "./contact.cc"
func_name = "APPLY_SPECIFIC(contact)"
def __init__(self, tol=1e-10, **kwargs):
self.tol = float(tol)
super(ContactPointsOp, self).__init__(self.func_file, self.func_name)
def c_code_cache_version(self):
return get_cache_version()
def c_compile_args(self, compiler):
return get_compile_args(compiler)
def c_headers(self, compiler):
return ["theano_helpers.h", "contact_points.h"]
def c_header_dirs(self, compiler):
return [pkg_resources.resource_filename(__name__, "include")]
def make_node(self, *args):
if len(args) != self.num_inputs:
raise ValueError("expected {0} inputs".format(self.num_inputs))
dtype = theano.config.floatX
in_args = []
for a in args:
try:
a = tt.as_tensor_variable(a)
except tt.AsTensorError:
pass
else:
dtype = theano.scalar.upcast(dtype, a.dtype)
in_args.append(a)
ndim = in_args[0].ndim
out_args = [
tt.TensorType(dtype=dtype, broadcastable=[False] * ndim)(),
tt.TensorType(dtype=dtype, broadcastable=[False] * ndim)(),
tt.TensorType(dtype="int32", broadcastable=[False] * ndim)()
]
return gof.Apply(self, in_args, out_args)
def infer_shape(self, node, shapes):
return shapes[0], shapes[0], shapes[0]
class ContactPointsOp(gof.COp):
num_inputs = 7
params_type = gof.ParamsType(tol=theano.scalar.float64)
__props__ = ("tol", )
func_file = "./contact.cc"
func_name = "APPLY_SPECIFIC(contact)"
def __init__(self, tol=1e-10, **kwargs):
self.tol = float(tol)
super(ContactPointsOp, self).__init__(self.func_file, self.func_name)
def c_code_cache_version(self):
return get_cache_version()
def c_compile_args(self, compiler):
return get_compile_args(compiler)
def c_headers(self, compiler):
return ["exoplanet/theano_helpers.h", "exoplanet/contact_points.h"]
def c_header_dirs(self, compiler):
return get_header_dirs(eigen=False)
def make_node(self, *args):
if len(args) != self.num_inputs:
raise ValueError("expected {0} inputs".format(self.num_inputs))
in_args = [tt.as_tensor_variable(a) for a in args]
out_args = [
in_args[0].type(),
in_args[0].type(),
tt.zeros_like(in_args[0], dtype="int32").type(),
]
return gof.Apply(self, in_args, out_args)
def infer_shape(self, node, shapes):
return shapes[0], shapes[0], shapes[0]
def params_type(self):
return gof.ParamsType(i=theano.scalar.basic.int64)
class RegularGridOp(gof.COp):
params_type = gof.ParamsType(
ndim=theano.scalar.int64,
nout=theano.scalar.int64,
check_sorted=theano.scalar.bool,
bounds_error=theano.scalar.bool,
)
__props__ = ("ndim", "nout", "check_sorted", "bounds_error")
func_file = "./regular_grid.cc"
func_name = "APPLY_SPECIFIC(regular_grid)"
def __init__(self,
ndim,
nout=-1,
check_sorted=True,
bounds_error=True,
**kwargs):
self.ndim = int(ndim)
if not 0 < self.ndim <= 5:
raise ValueError("ndim must be less than or equal to 5")
self.nout = int(nout)
self.check_sorted = bool(check_sorted)
self.bounds_error = bool(bounds_error)
super(RegularGridOp, self).__init__(self.func_file, self.func_name)
def c_code_cache_version(self):
return get_cache_version()
def c_headers(self, compiler):
return ["theano_helpers.h"]
def c_header_dirs(self, compiler):
return [pkg_resources.resource_filename(__name__, "include")
] + get_header_dirs()
def c_compile_args(self, compiler):
args = get_compile_args(compiler)
args.append("-DREGULAR_GRID_NDIM={0}".format(self.ndim))
if self.ndim == 1:
args.append("-DREGULAR_GRID_NDIM_ORDER=Eigen::ColMajor")
else:
args.append("-DREGULAR_GRID_NDIM_ORDER=Eigen::RowMajor")
if 0 < self.nout <= 16:
args.append("-DREGULAR_GRID_NOUT={0}".format(self.nout))
if self.nout == 1:
args.append("-DREGULAR_GRID_NOUT_ORDER=Eigen::ColMajor")
else:
args.append("-DREGULAR_GRID_NOUT_ORDER=Eigen::RowMajor")
ndim_nout = self.ndim * self.nout
if 0 < ndim_nout <= 16:
args.append("-DREGULAR_GRID_NDIM_NOUT={0}".format(ndim_nout))
if ndim_nout == 1:
args.append("-DREGULAR_GRID_NDIM_NOUT_ORDER=Eigen::ColMajor")
else:
args.append("-DREGULAR_GRID_NDIM_NOUT_ORDER=Eigen::RowMajor")
for i in range(self.ndim):
args.append("-DREGULAR_GRID_{0}".format(i))
return args
def make_node(self, *args):
ndim = len(args) - 2
if ndim != self.ndim:
raise ValueError("number of input grids does not match ndim")
dtype = theano.config.floatX
in_args = []
for a in args:
try:
a = tt.as_tensor_variable(a)
except tt.AsTensorError:
pass
else:
dtype = theano.scalar.upcast(dtype, a.dtype)
in_args.append(a)
out_args = [
tt.TensorType(dtype=dtype, broadcastable=[False, False])(),
tt.TensorType(dtype=dtype, broadcastable=[False, False, False])(),
]
return gof.Apply(self, in_args, out_args)
def grad(self, inputs, gradients):
xi = inputs[0]
zi, dz = self(*inputs)
bz = gradients[0]
bx = tt.sum(tt.reshape(bz, (xi.shape[0], 1, zi.shape[1])) * dz,
axis=-1)
return tuple([bx] + [tt.zeros_like(i) for i in inputs[1:]])
def R_op(self, inputs, eval_points):
if eval_points[0] is None:
return eval_points
return self.grad(inputs, eval_points)
class IntegratedLimbDarkOp(StarryBaseOp):
params_type = gof.ParamsType(
tol=theano.scalar.float64,
min_depth=theano.scalar.int32,
max_depth=theano.scalar.int32,
Nc=theano.scalar.int32,
circular=theano.scalar.bool,
)
__props__ = ()
func_file = "./integrated_limbdark.cc"
func_name = "APPLY_SPECIFIC(integrated_limbdark)"
def __init__(
self,
tol=1e-5,
min_depth=0,
max_depth=50,
Nc=-1,
circular=False,
**kwargs
):
self.tol = float(tol)
self.min_depth = max(0, int(min_depth))
self.max_depth = max(self.min_depth + 1, int(max_depth))
self.Nc = int(Nc)
self.circular = bool(circular)
super(IntegratedLimbDarkOp, self).__init__()
def make_node(self, *args):
if len(args) != 11:
raise ValueError("wrong number of inputs")
in_args = [tt.as_tensor_variable(a) for a in args]
out_args = [
in_args[1].type(),
tt.lscalar().type(),
tt.TensorType(
dtype=theano.config.floatX,
broadcastable=[False] * (in_args[0].ndim + in_args[1].ndim),
)(),
in_args[1].type(),
in_args[1].type(),
in_args[1].type(),
in_args[1].type(),
in_args[1].type(),
]
if self.circular:
out_args += [
tt.lscalar().type(),
tt.lscalar().type(),
tt.lscalar().type(),
]
else:
out_args += [
in_args[1].type(),
in_args[1].type(),
in_args[1].type(),
]
return gof.Apply(self, in_args, out_args)
def infer_shape(self, node, shapes):
shape = shapes[1]
out_shapes = [
shape,
(),
list(shapes[0]) + list(shapes[1]),
shape,
shape,
shape,
shape,
shape,
]
if self.circular:
out_shapes += [(), (), ()]
else:
out_shapes += [shape, shape, shape]
return tuple(out_shapes)
def c_compile_args(self, compiler):
args = super(IntegratedLimbDarkOp, self).c_compile_args(compiler)
args.append("-DLIMBDARK_NC={0}".format(self.Nc))
if self.circular:
args.append("-DLIMBDARK_CIRCULAR")
return args
def grad(self, inputs, gradients):
c = inputs[0]
f, neval, dcl, dt, dr, dn, daome2, dcosi, de, dsinw, dcosw = self(
*inputs
)
bf = gradients[0]
for i, g in enumerate(gradients[1:]):
if not isinstance(g.type, theano.gradient.DisconnectedType):
raise ValueError(
"can't propagate gradients wrt parameter {0}".format(i + 1)
)
bc = tt.sum(
tt.reshape(bf, (1, bf.size)) * tt.reshape(dcl, (c.size, bf.size)),
axis=-1,
)
results = [
tt.reshape(bc, inputs[0].shape),
tt.zeros_like(inputs[1]),
tt.reshape(bf * dt, inputs[2].shape),
tt.reshape(bf * dr, inputs[3].shape),
tt.reshape(bf * dn, inputs[4].shape),
tt.reshape(bf * daome2, inputs[5].shape),
tt.zeros_like(inputs[6]),
tt.reshape(bf * dcosi, inputs[7].shape),
]
if self.circular:
results += [
tt.zeros_like(inputs[8]),
tt.zeros_like(inputs[9]),
tt.zeros_like(inputs[10]),
]
else:
results += [
tt.reshape(bf * de, inputs[8].shape),
tt.reshape(bf * dsinw, inputs[9].shape),
tt.reshape(bf * dcosw, inputs[10].shape),
]
return tuple(results)
def R_op(self, inputs, eval_points):
if eval_points[0] is None:
return eval_points
return self.grad(inputs, eval_points)
class KeplerOp(gof.COp):
params_type = gof.ParamsType(tol=theano.scalar.float64, )
__props__ = ("tol", )
func_file = "./solver.cc"
func_name = "APPLY_SPECIFIC(solver)"
def __init__(self, tol=1e-12, **kwargs):
self.tol = float(tol)
super(KeplerOp, self).__init__(self.func_file, self.func_name)
def c_code_cache_version(self):
return get_cache_version()
def c_headers(self, compiler):
return ["theano_helpers.h", "solver.h"]
def c_header_dirs(self, compiler):
return [pkg_resources.resource_filename(__name__, "include")]
def make_node(self, mean_anom, eccen):
in_args = [
tt.as_tensor_variable(mean_anom),
tt.as_tensor_variable(eccen)
]
return gof.Apply(self, in_args, [in_args[0].type(), in_args[0].type()])
def infer_shape(self, node, shapes):
return shapes[0], shapes[0]
def grad(self, inputs, gradients):
M, e = inputs
E, f = self(M, e)
bM = tt.zeros_like(M)
be = tt.zeros_like(M)
ecosE = e * tt.cos(E)
if not isinstance(gradients[0].type, theano.gradient.DisconnectedType):
# Backpropagate E_bar
bM = gradients[0] / (1 - ecosE)
be = tt.sin(E) * bM
if not isinstance(gradients[1].type, theano.gradient.DisconnectedType):
# Backpropagate f_bar
sinf2 = tt.sin(0.5 * f)
cosf2 = tt.cos(0.5 * f)
tanf2 = sinf2 / cosf2
e2 = e**2
ome2 = 1 - e2
ome = 1 - e
ope = 1 + e
cosf22 = cosf2**2
twoecosf22 = 2 * e * cosf22
factor = tt.sqrt(ope / ome)
inner = (twoecosf22 + ome) * tt.as_tensor_variable(gradients[1])
bM += factor * (ome * tanf2**2 + ope) * inner * cosf22 / (ope *
ome2)
be += -2 * cosf22 * tanf2 / ome2**2 * inner * (ecosE - 2 + e2)
return [bM, be]
def R_op(self, inputs, eval_points):
if eval_points[0] is None:
return eval_points
return self.grad(inputs, eval_points)
class IntegratedLimbDarkOp(StarryBaseOp):
params_type = gof.ParamsType(
tol=theano.scalar.float64,
min_depth=theano.scalar.int32,
max_depth=theano.scalar.int32,
Nc=theano.scalar.int32,
include_contacts=theano.scalar.bool,
)
__props__ = ()
func_file = "./integrated_limbdark.cc"
func_name = "APPLY_SPECIFIC(integrated_limbdark)"
def __init__(self,
tol=1e-6,
min_depth=0,
max_depth=50,
Nc=-1,
include_contacts=False,
**kwargs):
self.tol = float(tol)
self.min_depth = max(0, int(min_depth))
self.max_depth = max(self.min_depth + 1, int(max_depth))
self.Nc = int(Nc)
self.include_contacts = bool(include_contacts)
super(IntegratedLimbDarkOp, self).__init__()
def make_node(self, *args):
if len(args) != 11:
raise ValueError("wrong number of inputs")
dtype = theano.config.floatX
in_args = [tt.as_tensor_variable(a) for a in args]
out_args = [
in_args[1].type(),
tt.TensorType(dtype=theano.config.floatX,
broadcastable=[False] * (in_args[1].ndim + 1))(),
in_args[1].type(),
in_args[2].type(),
in_args[3].type(),
in_args[4].type(),
in_args[5].type(),
in_args[6].type(),
in_args[7].type(),
tt.lscalar().type(),
]
return gof.Apply(self, in_args, out_args)
def infer_shape(self, node, shapes):
return (shapes[1], list(shapes[0]) + list(shapes[1]), shapes[1],
shapes[2], shapes[3], shapes[4], shapes[5], shapes[6],
shapes[7], ())
def grad(self, inputs, gradients):
c, r, x, xt, xtt, y, yt, ytt, z, zt, dt = inputs
f, dfdcl, dfdr, dfdx, dfdxt, dfdxtt, dfdy, dfdyt, dfdytt, neval \
= self(*inputs)
bf = gradients[0]
for i, g in enumerate(gradients[1:]):
if not isinstance(g.type, theano.gradient.DisconnectedType):
raise ValueError(
"can't propagate gradients wrt parameter {0}".format(i +
1))
bc = tt.sum(tt.reshape(bf, (1, bf.size)) *
tt.reshape(dfdcl, (c.size, bf.size)),
axis=-1)
br = bf * dfdr
bx = bf * dfdx
bxt = bf * dfdxt
bxtt = bf * dfdxtt
by = bf * dfdy
byt = bf * dfdyt
bytt = bf * dfdytt
return (bc, br, bx, bxt, bxtt, by, byt, bytt, tt.zeros_like(z),
tt.zeros_like(zt), tt.zeros_like(dt))
def R_op(self, inputs, eval_points):
if eval_points[0] is None:
return eval_points
return self.grad(inputs, eval_points)
class IntegrateOp(gof.COp):
params_type = gof.ParamsType(
t=theano.scalar.float64,
dt=theano.scalar.float64,
integrator=theano.scalar.int32,
)
__props__ = ("t", "dt", "integrator")
func_file = "./integrate.cc"
func_name = "APPLY_SPECIFIC(integrate)"
_INTEGRATORS = {
"ias15": 0,
"whfast": 1,
"sei": 2,
"leapfrog": 4,
"none": 7,
"janus": 8,
"mercurius": 9,
}
def __init__(self, t=0.0, dt=0.1, integrator="ias15", **kwargs):
self.t = float(t)
self.dt = float(dt)
self.integrator = self._INTEGRATORS.get(integrator.lower(), None)
if self.integrator is None:
raise ValueError("unknown integrator {0}".format(integrator))
self.integrator = np.int32(self.integrator)
super(IntegrateOp, self).__init__(self.func_file, self.func_name)
def c_code_cache_version(self):
return get_cache_version()
def c_headers(self, compiler):
return [
"theano_helpers.h",
"rebound.h",
"vector",
"array",
"numeric",
"algorithm",
]
def c_header_dirs(self, compiler):
return [pkg_resources.resource_filename(__name__, "")
] + get_header_dirs()
def c_compile_args(self, compiler):
return get_compile_args(compiler)
def c_libraries(self, compiler):
return [get_librebound_name()]
def c_lib_dirs(self, compiler):
return [get_librebound_path()]
def make_node(self, masses, initial_coords, times):
in_args = [
tt.as_tensor_variable(masses),
tt.as_tensor_variable(initial_coords),
tt.as_tensor_variable(times),
]
dtype = theano.config.floatX
out_args = [
tt.TensorType(dtype=dtype, broadcastable=[False] * 3)(),
tt.TensorType(dtype=dtype, broadcastable=[False] * 5)(),
]
return gof.Apply(self, in_args, out_args)
def infer_shape(self, node, shapes):
return (
list(shapes[2]) + list(shapes[0]) + [6],
list(shapes[2]) + list(shapes[0]) + [7] + list(shapes[0]) + [6],
)
def grad(self, inputs, gradients):
masses, initial_coords, times = inputs
coords, jac = self(*inputs)
bcoords = gradients[0]
if not isinstance(gradients[1].type, theano.gradient.DisconnectedType):
raise ValueError(
"can't propagate gradients with respect to Jacobian")
# (time, num, 6) * (time, num, 7, num, 6) -> (num, 7)
grad = tt.sum(bcoords[:, None, None, :, :] * jac, axis=(0, 3, 4))
return grad[:, 0], grad[:, 1:], tt.zeros_like(times)
def R_op(self, inputs, eval_points):
if eval_points[0] is None:
return eval_points
return self.grad(inputs, eval_points)
