def call_parameters(self, arr_shape):
substitution_dict = {
sym: value
for sym, value in zip(self._symbolic_shape, arr_shape)
if sym is not None
}
widths = [
end - start for start, end in zip(
_get_start_from_slice(self._iterationSlice),
_get_end_from_slice(self._iterationSlice, arr_shape))
]
widths = sp.Matrix(widths).subs(substitution_dict)
extend_bs = (1, ) * (3 - len(self._block_size))
block_size = self._block_size + extend_bs
if not self._compile_time_block_size:
assert len(block_size) == 3
adapted_block_size = []
for i in range(len(widths)):
factor = div_floor(prod(block_size[:i]),
prod(adapted_block_size))
adapted_block_size.append(
sp.Min(block_size[i] * factor, widths[i]))
block_size = tuple(adapted_block_size) + extend_bs
block_size = tuple(
sp.Min(bs, max_bs)
for bs, max_bs in zip(block_size, self._maximum_block_size))
grid = tuple(
div_ceil(length, block_size)
for length, block_size in zip(widths, block_size))
extend_gr = (1, ) * (3 - len(grid))
return {'block': block_size, 'grid': grid + extend_gr}
def cylinder_grasp_affordance(self, gripper, obj_input):
frame = obj_input.get_frame()
shape = obj_input.get_dimensions()
cylinder_z = frame.col(2)
cylinder_pos = pos_of(frame)
gripper_x = gripper.frame.col(0)
gripper_z = gripper.frame.col(2)
gripper_pos = pos_of(gripper.frame)
c_to_g = gripper_pos - cylinder_pos
zz_align = sp.Abs(gripper_z.dot(cylinder_z))
xz_align = gripper_x.dot(cylinder_z)
dist_z = cylinder_z.dot(c_to_g)
border_z = (shape[2] - gripper.height) * 0.5
cap_dist_normalized_signed = dist_z / border_z
cap_dist_normalized = sp.Abs(cap_dist_normalized_signed)
cap_top_grasp = 1 - sp.Max(-xz_align * sp.Min(cap_dist_normalized_signed, 1), 0)
cap_bottom_grasp = 1 - sp.Min(xz_align * sp.Max(cap_dist_normalized_signed, -1), 0)
dist_z_center_normalized = sp.Max(1 - cap_dist_normalized, 0)
dist_ax = sp.sqrt(frame.col(0).dot(c_to_g) ** 2 + frame.col(1).dot(c_to_g) ** 2)
center_grasp = (1 - dist_z_center_normalized - dist_ax) * zz_align
return sp.Max(center_grasp, cap_top_grasp, cap_bottom_grasp) * obj_input.get_class_probability()
def calc_set_union(set_a, set_b):
if isinstance(set_a, subsets.Indices) or isinstance(set_b, subsets.Indices):
raise NotImplementedError('Set union with indices is not implemented.')
if not (isinstance(set_a, subsets.Range) and isinstance(set_b, subsets.Range)):
raise TypeError('Can only compute the union of ranges.')
if len(set_a) != len(set_b):
raise ValueError('Range dimensions do not match')
union = []
for range_a, range_b in zip(set_a, set_b):
r_union = []
for i in range(3):
if isinstance(range_a[i], SymExpr):
a_exact = range_a[i].expr
a_approx = range_a[i].approx
else:
a_exact = range_a[i]
a_approx = range_a[i]
if isinstance(range_b[i], SymExpr):
b_exact = range_b[i].expr
b_approx = range_b[i].approx
else:
b_exact = range_b[i]
b_approx = range_b[i]
if i in {0, 2}:
r_union.append(SymExpr(sympy.Min(a_exact, b_exact), sympy.Min(a_approx, b_approx)))
else:
r_union.append(SymExpr(sympy.Max(a_exact, b_exact), sympy.Max(a_approx, b_approx)))
union.append(r_union)
# union.append([
# sympy.Min(range_a[0], range_b[0]),
# sympy.Max(range_a[1], range_b[1]),
# sympy.Min(range_a[2], range_b[2]),
# ])
return subsets.Range(union)
def _execute_controller(self):
rate = rospy.Rate(10)
P, I = sp.var('p i')
min_linear_vel = sp.var("minLinearVel")
min_angular_vel = sp.var("minAngularVel")
distErr, distVerticalErr, vertDiff, lastLinearOut = sp.var(
'distErr distVerticalErr vertDiff lastLinearOut')
anglErr, lastAngularOut, anglDiff = sp.var(
'anglErr lastAngularOut anglDiff')
self.linear_eq = sp.Piecewise(
(sp.Max(P * distErr + I * lastLinearOut, min_linear_vel),
sp.And(sp.Abs(anglErr) < anglDiff, distErr >= 0)),
(sp.Min(P * distErr + I * lastLinearOut, -min_linear_vel),
sp.And(sp.Abs(anglErr) < anglDiff, distErr < 0)),
(0, sp.Abs(anglErr) >= anglDiff))
self.linear_vertical_eq = sp.Piecewise(
(sp.Max(P * distVerticalErr + I * lastLinearOut,
min_linear_vel), distVerticalErr > vertDiff), (0, True))
angular_eq_temp = P * anglErr + I * lastAngularOut
self.angular_eq = sp.Piecewise(
(sp.Min(angular_eq_temp, -min_angular_vel), angular_eq_temp < 0),
(sp.Max(angular_eq_temp, min_angular_vel), angular_eq_temp >= 0))
constansts = {
P: self.current_config["p"],
I: self.current_config["i"],
min_linear_vel: self.current_config["min_linear_vel"],
min_angular_vel: self.current_config["min_angular_vel"],
anglDiff: self.current_config["angular_diff"],
vertDiff: self.current_config["vertical_diff"]
}
variables = [
distErr, distVerticalErr, lastLinearOut, anglErr, lastAngularOut
]
self.linear_eq = self.linear_eq.subs(constansts)
self.linear_vertical_eq = self.linear_vertical_eq.subs(constansts)
self.angular_eq = self.angular_eq.subs(constansts)
while not rospy.is_shutdown():
if self._goal != None and self._enabled:
self._command_to(self._goal, variables)
self._reach_publisher.publish(Bool(self._reach))
rate.sleep()
def f(self, yvec, params):
import sympy as sp
return NumSysLin.f(self, [
m * yi for m, yi in zip(
self.max_concs(params, min_=lambda x: sp.Min(*x),
dtype=object), yvec)
], params)
def detect_reduction_type(wcr_str, openmp=False):
""" Inspects a lambda function and tries to determine if it's one of the
built-in reductions that frameworks such as MPI can provide.
:param wcr_str: A Python string representation of the lambda function.
:param openmp: Detect additional OpenMP reduction types.
:return: dtypes.ReductionType if detected, dtypes.ReductionType.Custom
if not detected, or None if no reduction is found.
"""
if wcr_str == '' or wcr_str is None:
return None
# Get lambda function from string
wcr = eval(wcr_str)
wcr_ast = ast.parse(wcr_str).body[0].value.body
# Run function through symbolic math engine
a = sympy.Symbol('a')
b = sympy.Symbol('b')
try:
result = wcr(a, b)
except (TypeError, AttributeError,
NameError): # e.g., "Cannot determine truth value of relational"
result = None
# Check resulting value
if result == sympy.Max(a, b) or (isinstance(wcr_ast, ast.Call)
and isinstance(wcr_ast.func, ast.Name)
and wcr_ast.func.id == 'max'):
return dtypes.ReductionType.Max
elif result == sympy.Min(a, b) or (isinstance(wcr_ast, ast.Call)
and isinstance(wcr_ast.func, ast.Name)
and wcr_ast.func.id == 'min'):
return dtypes.ReductionType.Min
elif result == a + b:
return dtypes.ReductionType.Sum
elif result == a * b:
return dtypes.ReductionType.Product
elif result == a & b:
return dtypes.ReductionType.Bitwise_And
elif result == a | b:
return dtypes.ReductionType.Bitwise_Or
elif result == a ^ b:
return dtypes.ReductionType.Bitwise_Xor
elif isinstance(wcr_ast, ast.BoolOp) and isinstance(wcr_ast.op, ast.And):
return dtypes.ReductionType.Logical_And
elif isinstance(wcr_ast, ast.BoolOp) and isinstance(wcr_ast.op, ast.Or):
return dtypes.ReductionType.Logical_Or
elif (isinstance(wcr_ast, ast.Compare)
and isinstance(wcr_ast.ops[0], ast.NotEq)):
return dtypes.ReductionType.Logical_Xor
elif result == b:
return dtypes.ReductionType.Exchange
# OpenMP extensions
elif openmp and result == a - b:
return dtypes.ReductionType.Sub
elif openmp and result == a / b:
return dtypes.ReductionType.Div
return dtypes.ReductionType.Custom
def test_kappa_expressions():
Monomer('A', ['site'], {'site': ['u']})
Parameter('two', 2)
Parameter('kr', 0.1)
Parameter('num_A', 1000)
Expression('kf', 1e-5 / two)
Expression('test_sqrt', -1 + sympy.sqrt(1 + two))
Expression('test_pi', sympy.pi)
Expression('test_e', sympy.E)
Expression('test_log', sympy.log(two))
Expression('test_exp', sympy.exp(two))
Expression('test_sin', sympy.sin(two))
Expression('test_cos', sympy.cos(two))
Expression('test_tan', sympy.tan(two))
Expression('test_max', sympy.Max(two, kr, 2.0))
Expression('test_min', sympy.Min(two, kr, 2.0))
Expression('test_mod', sympy.Mod(10, two))
Expression('test_piecewise',
sympy.Piecewise((0.0, two < 400.0), (1.0, True)))
Initial(A(site=('u')), num_A)
Rule('dimerize_fwd',
A(site='u') + A(site='u') >> A(site=('u', 1)) % A(site=('u', 1)), kf)
Rule('dimerize_rev',
A(site=('u', 1)) % A(site=('u', 1)) >> A(site='u') + A(site='u'), kr)
# We need an arbitrary observable here to get a Kappa output file
Observable('A_obs', A())
# Accommodates Expression in kappa simulation
run_simulation(model, time=0)
Rule('degrade_dimer', A(site=('u', ANY)) >> None, kr)
Observable('dimer', A(site=('u', ANY)))
# Accommodates site with explicit state and arbitrary bond
run_simulation(model, time=0, seed=_KAPPA_SEED)
def transform_set(x, expr, sympy_set):
""" Transform a sympy_set by an expression
>>> x = sympy.Symbol('x')
>>> domain = sympy.Interval(-sympy.pi / 4, -sympy.pi / 6, False, True) | sympy.Interval(sympy.pi / 6, sympy.pi / 4, True, False)
>>> transform_set(x, -2 * x, domain)
[-pi/2, -pi/3) U (pi/3, pi/2]
"""
if isinstance(sympy_set, sympy.Union):
return sympy.Union(
transform_set(x, expr, arg) for arg in sympy_set.args)
if isinstance(sympy_set, sympy.Intersection):
return sympy.Intersection(
transform_set(x, expr, arg) for arg in sympy_set.args)
f = sympy.Lambda(x, expr)
if isinstance(sympy_set, sympy.Interval):
left, right = f(sympy_set.left), f(sympy_set.right)
if left < right:
new_left_open = sympy_set.left_open
new_right_open = sympy_set.right_open
else:
new_left_open = sympy_set.right_open
new_right_open = sympy_set.left_open
return sympy.Interval(sympy.Min(left, right), sympy.Max(left, right),
new_left_open, new_right_open)
if isinstance(sympy_set, sympy.FiniteSet):
return sympy.FiniteSet(list(map(f, sympy_set)))
def compile(self):
#code.interact(local=locals())
# define the render function
dist = float('inf')
for face in self.faces:
dist = sp.Min(dist, face.intersect(None, {}))
# print the render function
codegen(('distance', dist),
'C',
to_files=True,
prefix='dr_cube/distance',
header=True,
empty=True)
# print the derivitives
for var in self.get_syms():
deriv = sp.diff(dist, var)
name = 'diff_{0}'.format(var)
deriv = deriv.replace(sp.Heaviside, simp_heaviside)
codegen((name, deriv),
'C',
to_files=True,
prefix='dr_cube/' + name,
header=True,
empty=True)
def _create_strided_range(self, sdfg: SDFG, state: SDFGState,
map_entry: nodes.MapEntry):
map_exit = state.exit_node(map_entry)
dim_idx = self.dim_idx
new_dim_prefix = self.new_dim_prefix
tile_size = self.tile_size
divides_evenly = self.divides_evenly
tile_stride = self.tile_stride
if tile_stride == 0:
tile_stride = tile_size
if tile_stride != tile_size:
raise NotImplementedError
# Retrieve parameter and range of dimension to be strip-mined.
target_dim = map_entry.map.params[dim_idx]
td_from, td_to, td_step = map_entry.map.range[dim_idx]
new_dim = self._find_new_dim(sdfg, state, map_entry, new_dim_prefix,
target_dim)
new_dim_range = (td_from, td_to, tile_size)
new_map = nodes.Map(map_entry.map.label, [new_dim],
subsets.Range([new_dim_range]))
dimsym = dace.symbolic.pystr_to_symbolic(new_dim)
td_from_new = dimsym
if divides_evenly:
td_to_new = dimsym + tile_size - 1
else:
if isinstance(td_to, dace.symbolic.SymExpr):
td_to = td_to.expr
td_to_new = dace.symbolic.SymExpr(
sympy.Min(dimsym + tile_size - 1, td_to),
dimsym + tile_size - 1)
td_step_new = td_step
return new_dim, new_map, (td_from_new, td_to_new, td_step_new)
def calc_set_union(set_a, set_b):
if isinstance(set_a, subsets.Indices) or isinstance(
set_b, subsets.Indices):
raise NotImplementedError('Set union with indices is not implemented.')
if not (isinstance(set_a, subsets.Range)
and isinstance(set_b, subsets.Range)):
raise TypeError('Can only compute the union of ranges.')
if len(set_a) != len(set_b):
raise ValueError('Range dimensions do not match')
union = []
for range_a, range_b in zip(set_a, set_b):
union.append([
sympy.Min(range_a[0], range_b[0]),
sympy.Max(range_a[1], range_b[1]),
sympy.Min(range_a[2], range_b[2]),
])
return subsets.Range(union)
def cie_optical_depth_correction(self):
ciefudge = 1.0
mdensity = sympy.Symbol('mdensity')
tau = (mdensity / 1.96e16)**2.0
tau = sympy.Max(mdensity, 1e-5)
ciefudge = sympy.Min((1.0 - sympy.exp(-tau)) / tau, 1.0)
return ciefudge
def sort(fs):
fs = list(fs)
for i in range(len(fs) - 1, -1, -1):
for j in range(i, -1, -1):
Fi = fs[i]
Fj = fs[j]
fs[j] = sympy.Min(Fi, Fj)
fs[i] = sympy.Max(Fi, Fj)
return fs
def simplify_ext(expr):
"""
An extended version of simplification with expression fixes for sympy.
:param expr: A sympy expression.
:return: Simplified version of the expression.
"""
a = sympy.Wild('a')
b = sympy.Wild('b')
c = sympy.Wild('c')
# Push expressions into both sides of min/max.
# Example: Min(N, 4) + 1 => Min(N + 1, 5)
dic = expr.match(sympy.Min(a, b) + c)
if dic:
return sympy.Min(dic[a] + dic[c], dic[b] + dic[c])
dic = expr.match(sympy.Max(a, b) + c)
if dic:
return sympy.Max(dic[a] + dic[c], dic[b] + dic[c])
return expr
def _infer_binary_ops(self, node):
funcs = {
'Add': lambda l: l[0] + l[1],
'Div': lambda l: l[0] // l[1], # integer div in sympy
'Max': lambda l: sympy.Max(l[0], l[1]),
'Min': lambda l: sympy.Min(l[0], l[1]),
'Mul': lambda l: l[0] * l[1],
'Sub': lambda l: l[0] - l[1]
}
assert node.op_type in funcs
self._compute_on_sympy_data(node, funcs[node.op_type])
def intersect(self, other):
minCorner = [
sp.Max(self.minCorner[d], other.minCorner[d])
for d in range(self.dim)
]
maxCorner = [
sp.Max(minCorner[d],
sp.Min(self.maxCorner[d], other.maxCorner[d]))
for d in range(self.dim)
]
return AABB(minCorner, maxCorner)
def widen_bound(op: IComparisonOp[Any], old: sym.Number, new: sym.Number) -> sym.Number:
if op == operator.le:
mx = sym.Max(old, new) # type: ignore
assert isinstance(mx, sym.Number)
return mx
elif op == operator.ge:
mn = sym.Min(old, new) # type: ignore
assert isinstance(mn, sym.Number)
return mn
else:
raise Exception("unsupported operator")
def overapproximate(expr):
"""
Takes a sympy expression and returns its maximal possible value
in specific cases.
"""
if isinstance(expr, list):
return [overapproximate(elem) for elem in expr]
if isinstance(expr, SymExpr):
if expr.expr != expr.approx:
return expr.approx
else:
return overapproximate(expr.expr)
if not isinstance(expr, sympy.Basic):
return expr
a = sympy.Wild('a')
b = sympy.Wild('b')
c = sympy.Wild('c')
# If Min(x, N-y), return the non-symbolic of the two components
match = expr.match(sympy.Min(a, b) + c)
if match is not None and len(match) == 3:
# First, construct the min expression with "c" inline
newexpr = sympy.Min(match[a] + match[c], match[b] + match[c])
# Match again
match = newexpr.match(sympy.Min(a, b))
if match is not None and len(match) == 2:
if issymbolic(match[a]) and not issymbolic(match[b]):
return match[b]
if issymbolic(match[b]) and not issymbolic(match[a]):
return match[a]
# If ceiling((k * ((N - 1) / k))) + k), return N
a = sympy.Wild('a', properties=[lambda k: k.is_Symbol or k.is_Integer])
b = sympy.Wild('b', properties=[lambda k: k.is_Symbol or k.is_Integer])
int_floor = sympy.Function('int_floor')
match = expr.match(sympy.ceiling(b * int_floor(a - 1, b)) + b)
if match is not None and len(match) == 2:
return match[a]
return expr
def _create_from_tile_numbers(self, sdfg: SDFG, state: SDFGState,
map_entry: nodes.MapEntry):
map_exit = state.exit_node(map_entry)
# Retrieve transformation properties.
dim_idx = self.dim_idx
new_dim_prefix = self.new_dim_prefix
divides_evenly = self.divides_evenly
number_of_tiles = self.tile_size
tile_stride = self.tile_stride
number_of_tiles = dace.symbolic.pystr_to_symbolic(number_of_tiles)
# Retrieve parameter and range of dimension to be strip-mined.
target_dim = map_entry.map.params[dim_idx]
td_from, td_to, td_step = map_entry.map.range[dim_idx]
tile_size = map_entry.map.range.size_exact()[dim_idx] / number_of_tiles
if tile_stride == 0:
tile_stride = tile_size
if tile_stride != tile_size:
raise NotImplementedError
new_dim = self._find_new_dim(sdfg, state, map_entry, new_dim_prefix,
target_dim)
new_dim_range = (td_from, number_of_tiles - 1, 1)
new_map = nodes.Map(map_entry.map.label, [new_dim],
subsets.Range([new_dim_range]))
dimsym = dace.symbolic.pystr_to_symbolic(new_dim)
td_from_new = dimsym * tile_size
if divides_evenly:
td_to_new = (dimsym + 1) * tile_size - 1
else:
if isinstance(td_to, dace.symbolic.SymExpr):
td_to = td_to.expr
td_to_new = dace.symbolic.SymExpr(
sympy.Min((dimsym + 1) * tile_size - 1, td_to),
(dimsym + 1) * tile_size - 1)
td_step_new = td_step
return new_dim, new_map, (td_from_new, td_to_new, td_step_new)
def optimize(g):
"""
Find parameters by optimizing over the given table of estimated inequalities
"""
# Build the objective
g['Terms'] = g.apply(lambda row: sp.Min(row.g, 0)**2, axis=1)
objective = g.Terms.sum()
variables = list(objective.atoms(sp.Symbol))
# Turn the objective into a function
def function(values):
z = zip(variables, values)
return float(objective.subs(z))
# Create the intial guess
initial_guess = np.ones(len(variables))
# Optimize!
return [
opt.minimize(function, initial_guess, method='nelder-mead'), variables
]
def test_bng_printer():
# Constants
assert _bng_print(sympy.pi) == '_pi'
assert _bng_print(sympy.E) == '_e'
x, y = sympy.symbols('x y')
# Binary functions
assert _bng_print(sympy.sympify('x & y')) == 'x && y'
assert _bng_print(sympy.sympify('x | y')) == 'x || y'
# Trig functions
assert _bng_print(sympy.sin(x)) == 'sin(x)'
assert _bng_print(sympy.cos(x)) == 'cos(x)'
assert _bng_print(sympy.tan(x)) == 'tan(x)'
assert _bng_print(sympy.asin(x)) == 'asin(x)'
assert _bng_print(sympy.acos(x)) == 'acos(x)'
assert _bng_print(sympy.atan(x)) == 'atan(x)'
assert _bng_print(sympy.sinh(x)) == 'sinh(x)'
assert _bng_print(sympy.cosh(x)) == 'cosh(x)'
assert _bng_print(sympy.tanh(x)) == 'tanh(x)'
assert _bng_print(sympy.asinh(x)) == 'asinh(x)'
assert _bng_print(sympy.acosh(x)) == 'acosh(x)'
assert _bng_print(sympy.atanh(x)) == 'atanh(x)'
# Logs and powers
assert _bng_print(sympy.log(x)) == 'ln(x)'
assert _bng_print(sympy.exp(x)) == 'exp(x)'
assert _bng_print(sympy.sqrt(x)) == 'sqrt(x)'
# Rounding
assert _bng_print(sympy.Abs(x)) == 'abs(x)'
assert _bng_print(sympy.floor(x)) == 'rint(x - 0.5)'
assert _bng_print(sympy.ceiling(x)) == '(rint(x + 1) - 1)'
# Min/max
assert _bng_print(sympy.Min(x, y)) == 'min(x, y)'
assert _bng_print(sympy.Max(x, y)) == 'max(x, y)'
def __init__(self, epsilon, debug=False):
self.debug = debug
self.epsilon = epsilon
self.A = 'A'
self.A_bar = 'A_bar'
self.B = 'B'
self.B_bar = 'B_bar'
self.operators = [self.A]
self.operators_bar = [self.A_bar]
x_1, x_2, z_1, z_2 = sp.symbols('x_1 x_2 z_1 z_2')
eps_inv = 1. / epsilon
I_1_left_integrand = z_1 * z_2 * (1 - epsilon * (z_2 - z_1))**2
I_1_right_integrand = z_1 * z_2 * (1 - epsilon * (z_1 - z_2))**2
I_1_1_right = sp.integrate(I_1_right_integrand, (z_2, 0, z_1))
I_1_1_left = sp.integrate(I_1_left_integrand, (z_2, z_1, x_2))
I_1_1 = sp.integrate(I_1_1_left + I_1_1_right, (z_1, 0, x_1))
I_1_2_right = sp.integrate(I_1_right_integrand,
(z_2, 0, z_1 - eps_inv))
I_1_2 = sp.integrate(I_1_2_right, (z_1, eps_inv, x_1))
I_1_3_left = sp.integrate(I_1_left_integrand, (z_1, 0, z_2 - eps_inv))
I_1_3 = sp.integrate(I_1_3_left,
(z_2, eps_inv, sp.Min(x_1 + eps_inv, x_2)))
I_1_4_left = sp.integrate(I_1_left_integrand, (z_1, 0, x_1))
I_1_4 = sp.integrate(I_1_4_left, (z_2, x_1 + eps_inv, x_2))
self.I_1_1 = self.functionise_sympy([x_1, x_2], I_1_1)
self.I_1_2 = self.functionise_sympy([x_1, x_2], I_1_2)
self.I_1_3 = self.functionise_sympy([x_1, x_2], I_1_3)
self.I_1_4 = self.functionise_sympy([x_1, x_2], I_1_4)
I_2_left_integrand = z_1 * (z_2 - 1) * (1 - epsilon * (z_2 - z_1))**2
I_2_left_c1 = sp.integrate(I_2_left_integrand,
(z_2, x_2, z_1 + eps_inv))
I_2_c1 = sp.integrate(I_2_left_c1,
(z_1, sp.Max(x_2 - eps_inv, 0.), x_1))
I_2_left_c2 = sp.integrate(I_2_left_integrand,
(z_1, z_2 - eps_inv, x_1))
I_2_c2 = sp.integrate(I_2_left_c2,
(z_2, x_2, sp.Min(x_1 + eps_inv, 1.)))
self.I_2_c1 = self.functionise_sympy([x_1, x_2], I_2_c1)
self.I_2_c2 = self.functionise_sympy([x_1, x_2], I_2_c2)
I_3_right_integrand = (z_1 - 1) * z_2 * (1 - epsilon * (z_1 - z_2))**2
I_3_left_integrand = (z_1 - 1) * z_2 * (1 - epsilon * (z_2 - z_1))**2
# assuming x_2 + eps_inv < 1...
I_3_1_c1_right = sp.integrate(I_3_right_integrand,
(z_2, z_1 - eps_inv, z_1))
I_3_1_c1_left = sp.integrate(I_3_left_integrand,
(z_2, z_1, z_1 + eps_inv))
I_3_1_c1 = sp.integrate(I_3_1_c1_right + I_3_1_c1_left,
(z_1, x_1, x_2 - eps_inv))
# sp.Max(x_1, eps_inv)
I_3_2_c1_right = sp.integrate(I_3_right_integrand,
(z_2, z_1 - eps_inv, z_1))
I_3_2_c1_left = sp.integrate(I_3_left_integrand, (z_2, z_1, x_2))
I_3_2_c1 = sp.integrate(I_3_2_c1_right + I_3_2_c1_left,
(z_1, sp.Max(x_2 - eps_inv, x_1), x_2))
I_3_3_c1_right = sp.integrate(I_3_right_integrand,
(z_2, z_1 - eps_inv, x_2))
I_3_3_c1 = sp.integrate(I_3_3_c1_right,
(z_1, x_2, sp.Min(x_2 + eps_inv, 1.)))
I_3_4_c1_right = sp.integrate(I_3_right_integrand,
(z_2, z_1 - eps_inv, 0))
I_3_4_c1 = sp.integrate(I_3_4_c1_right, (z_1, x_1, eps_inv))
self.I_3_1_c1 = self.functionise_sympy([x_1, x_2], I_3_1_c1)
self.I_3_2_c1 = self.functionise_sympy([x_1, x_2], I_3_2_c1)
self.I_3_3_c1 = self.functionise_sympy([x_1, x_2], I_3_3_c1)
self.I_3_4 = self.functionise_sympy([x_1, x_2], I_3_4_c1)
I_4_right_integrand = (z_1 - 1) * (z_2 - 1) * (1 - epsilon *
(z_1 - z_2))**2
I_4_left_integrand = (z_1 - 1) * (z_2 - 1) * (1 - epsilon *
(z_2 - z_1))**2
I_4_1_left = sp.integrate(I_4_left_integrand, (z_1, x_1, z_2))
I_4_1_right = sp.integrate(I_4_right_integrand, (z_1, z_2, 1))
I_4_1 = sp.integrate(I_4_1_right + I_4_1_left, (z_2, x_2, 1))
I_4_2_right = sp.integrate(I_4_right_integrand,
(z_2, x_2, z_1 - eps_inv))
I_4_2 = sp.integrate(I_4_2_right, (z_1, x_2 + eps_inv, 1))
I_4_3_left = sp.integrate(
I_4_left_integrand,
(z_1, sp.Max(x_2 - eps_inv, x_1), z_2 - eps_inv))
I_4_3 = sp.integrate(I_4_3_left, (z_2, sp.Max(x_2, x_1 + eps_inv), 1))
I_4_4_left = sp.integrate(I_4_left_integrand, (z_2, x_2, 1))
I_4_4 = sp.integrate(I_4_4_left, (z_1, x_1, x_2 - eps_inv))
self.I_4_1 = self.functionise_sympy([x_1, x_2], I_4_1)
self.I_4_2 = self.functionise_sympy([x_1, x_2], I_4_2)
self.I_4_3 = self.functionise_sympy([x_1, x_2], I_4_3)
self.I_4_4 = self.functionise_sympy([x_1, x_2], I_4_4)
# x' < x - eps_inv
A_k_I_1a = sp.integrate(
x_2 * (z_2 - 1) * (1 - epsilon * (x_1 - z_2))**2,
(z_2, sp.Max(x_1 - eps_inv, 0), x_1))
A_k_I_1b = sp.integrate(
x_2 * (z_2 - 1) * (1 - epsilon * (z_2 - x_1))**2,
(z_2, x_1, sp.Min(x_1 + eps_inv, 1)))
# x' > x + eps_inv
A_k_I_2a = sp.integrate(
(x_2 - 1) * z_2 * (1 - epsilon * (x_1 - z_2))**2,
(z_2, sp.Max(x_1 - eps_inv, 0), x_1))
A_k_I_2b = sp.integrate(
(x_2 - 1) * z_2 * (1 - epsilon * (z_2 - x_1))**2,
(z_2, x_1, sp.Min(x_1 + eps_inv, 1)))
# x' < x, x' > x - eps_inv
A_k_I_4a = sp.integrate(
(x_2 - 1) * z_2 * (1 - epsilon * (x_1 - z_2))**2,
(z_2, sp.Max(x_1 - eps_inv, 0), x_2))
A_k_I_4b = sp.integrate(
x_2 * (z_2 - 1) * (1 - epsilon * (x_1 - z_2))**2, (z_2, x_2, x_1))
A_k_I_4c = A_k_I_1b
# x' > x, x' < x + eps_inv
A_k_I_5a = A_k_I_2a
A_k_I_5b = sp.integrate(
(x_2 - 1) * z_2 * (1 - epsilon * (z_2 - x_1))**2, (z_2, x_1, x_2))
A_k_I_5c = sp.integrate(
x_2 * (z_2 - 1) * (1 - epsilon * (z_2 - x_1))**2,
(z_2, x_2, sp.Min(x_1 + eps_inv, 1)))
self.A_k_I_1 = self.functionise_sympy([x_1, x_2], A_k_I_1a + A_k_I_1b)
self.A_k_I_2 = self.functionise_sympy([x_1, x_2], A_k_I_2a + A_k_I_2b)
self.A_k_I_3 = self.functionise_sympy([x_1, x_2], A_k_I_2a + A_k_I_1b)
self.A_k_I_4 = self.functionise_sympy([x_1, x_2],
A_k_I_4a + A_k_I_4b + A_k_I_4c)
self.A_k_I_5 = self.functionise_sympy([x_1, x_2],
A_k_I_5a + A_k_I_5b + A_k_I_5c)
def _print_Min(self, expr):
""" Adapted from sympy/printing/cxxcode.py """
if len(expr.args) == 1:
return self._print(expr.args[0])
return "[min] {} {}".format(self._print(expr.args[0]),
self._print(sympy.Min(*expr.args[1:])))
def forward_projection(volume: pystencils.Field,
projection: pystencils.Field,
projection_matrix,
step_size=1,
cubic_bspline_interpolation=False,
add_to_projector=False,
central_ray_point=None):
# is_projection_stack = projection.spatial_dimensions == volume.spatial_dimensions
interpolation_mode = 'cubic_spline' if cubic_bspline_interpolation else 'linear'
volume_texture = pystencils.interpolation_astnodes.Interpolator(
volume, interpolation_mode)
ndim = volume.spatial_dimensions
projection_matrix = pystencils_reco.ProjectiveMatrix(projection_matrix)
t = pystencils_reco.typed_symbols('_parametrization', 'float32')
texture_coordinates = sympy.Matrix(
pystencils_reco.typed_symbols(f'_t:{ndim}', 'float32'))
u = projection.physical_coordinates_staggered
x = volume.index_to_physical(texture_coordinates)
is_perspective = projection_matrix.matrix.cols == ndim + 1
if is_perspective:
eqn = projection_matrix @ sympy.Matrix([*x, 1]) - sympy.Matrix(
[*(t * u), t])
else:
# this also works for perspective/cone beam projection (but may lead to instable parametrization)
eqn = projection_matrix @ x - u
ray_equations = sympy.solve(eqn, texture_coordinates, rational=False)
if not is_perspective:
t = [t for t in texture_coordinates
if t not in ray_equations.keys()][0]
assert len(
ray_equations.keys()
) == ndim - 1, "projection_matrix does not appear to define a projection"
ray_equations = sympy.Matrix(
[ray_equations[s] if s != t else t for s in texture_coordinates])
projection_vector = sympy.diff(ray_equations, t)
projection_vector_norm = projection_vector.norm()
projection_vector /= projection_vector_norm
conditions = pystencils_reco._geometry.coordinate_in_field_conditions(
volume, ray_equations)
if not central_ray_point:
central_ray_point = [0] * projection.spatial_dimensions
central_ray = projection_vector.subs({
i: j
for i, j in zip(pystencils.x_vector(projection.spatial_dimensions),
central_ray_point)
})
intersection_candidates = []
for i in range(ndim):
solution_min = sympy.solve(ray_equations[i], t, rational=False)
solution_max = sympy.solve(ray_equations[i] - volume.spatial_shape[i],
t,
rational=False)
intersection_candidates.extend(solution_min + solution_max)
intersection_point1 = sympy.Piecewise(
*[(f, sympy.And(*conditions).subs({t: f}))
for f in intersection_candidates], (-0, True))
intersection_point2 = sympy.Piecewise(
*[(f, sympy.And(*conditions).subs({t: f}))
for f in reversed(intersection_candidates)], (-0, True))
assert intersection_point1 != intersection_point2, \
"The intersections are unconditionally equal, reconstruction volume is not in detector FOV!"
# perform a integer set analysis here?
# space = isl.Space.create_from_names(isl.DEFAULT_CONTEXT, set=[str(t) for t in texture_coordinates])
# ray_set = isl.BasicSet.universe(space)
# for i, t in enumerate(texture_coordinates):
# # dafaq?
# ray_set.add_constraint(isl.Constraint.ineq_from_names(space, {str(texture_coordinates): 1}))
# ray_set.add_constraint(isl.Constraint.ineq_from_names(space,
# # {1: -volume.shape[i],
# str(texture_coordinates): -1}))
# ray_set.add_constraint(isl.Constraint.eq_from_name(space, ray_equations[i].subs({ #TODO
min_t = sympy.Min(intersection_point1, intersection_point2)
max_t = sympy.Max(intersection_point1, intersection_point2)
# parametrization_dim = list(ray_equations).index(t)
# min_t = 0
# max_t = volume.spatial_shape[parametrization_dim]
line_integral, num_steps, min_t_tmp, max_t_tmp, intensity_weighting, step = pystencils.data_types.typed_symbols(
'line_integral, num_steps, min_t_tmp, max_t_tmp, intensity_weighting, step',
'float32')
i = pystencils.data_types.TypedSymbol('i', 'int32')
num_steps = pystencils.data_types.TypedSymbol('num_steps', 'int32')
# step = step_size / projection_vector_norm
# tex_coord = ray_equations.subs({t: min_t_tmp + i * step})
tex_coord = ray_equations.subs({t: min_t_tmp}) + projection_vector * i
if callable(volume.coordinate_transform):
intensity_weighting_sym = projection_vector.dot(central_ray)**2
else:
intensity_weighting_sym = projection_vector.dot(central_ray)**2
assignments = {
min_t_tmp:
min_t,
max_t_tmp:
max_t,
num_steps:
sympy.ceiling(
(max_t_tmp - min_t_tmp) / (step_size / projection_vector_norm)),
line_integral:
sympy.Sum(volume_texture.at(tex_coord), (i, 0, num_steps)),
intensity_weighting:
intensity_weighting_sym,
projection.center():
(line_integral * step_size * intensity_weighting) +
(projection.center() if add_to_projector else 0)
# projection.center(): (max_t_tmp - min_t_tmp) / step # Uncomment to get path length
}
# def create_autodiff(self, constant_fields=None):
# backward_assignments = backward_projection(AdjointField(projections),
# AdjointField(volume),
# projection_matrix,
# 1)
# self._autodiff = pystencils.autodiff.AutoDiffOp(
# assignments, "op", constant_fields=constant_fields, backward_assignments=backward_assignments)
# assignments._create_autodiff = types.MethodType(create_autodiff, assignments)
return assignments
def or_(truth0, truth1):
exp = sympy.Min(truth0.expression, truth1.expression)
sig = sympy.Max(truth0.sigma, truth1.sigma)
return Truth(exp, sig)
def _infer_Min(self, node):
self._compute_on_sympy_data(node, lambda l: sympy.Min(l[0], l[1]))
def diffusion_reaction(fluctuations: bool):
# parameters
L = (32, 32)
stencil_factor = np.sqrt(1 / (1 + np.sqrt(2)))
dh = ps.create_data_handling(domain_size=L,
periodicity=True,
default_target=ps.Target.CPU)
species = 2
n_fields = []
j_fields = []
r_flux_fields = []
for i in range(species):
n_fields.append(dh.add_array(f'n_{i}', values_per_cell=1))
j_fields.append(
dh.add_array(f'j_{i}',
values_per_cell=3**dh.dim // 2,
field_type=ps.FieldType.STAGGERED_FLUX))
r_flux_fields.append(dh.add_array(f'r_{i}', values_per_cell=1))
velocity_field = dh.add_array('v', values_per_cell=dh.dim)
D = 0.00666
time = 1000
r_order = [2.0, 0.0]
r_rate_const = 0.00001
r_coefs = [-2, 1]
def grad(f):
return sp.Matrix([ps.fd.diff(f, i) for i in range(dh.dim)])
flux_eq = -D * grad(n_fields[0])
fvm_eq = ps.fd.FVM1stOrder(n_fields[0], flux=flux_eq)
vof_adv = ps.fd.VOF(j_fields[0], velocity_field, n_fields[0])
continuity_assignments = fvm_eq.discrete_continuity(j_fields[0])
# merge calculation of advection and diffusion terms
flux = []
for adv, div in zip(vof_adv, fvm_eq.discrete_flux(j_fields[0])):
assert adv.lhs == div.lhs
flux.append(ps.Assignment(adv.lhs, adv.rhs + div.rhs))
flux = ps.AssignmentCollection(flux)
if (fluctuations):
rng_symbol_gen = random_symbol(flux.subexpressions, dim=dh.dim)
for i in range(len(flux.main_assignments)):
n = j_fields[0].staggered_stencil[i]
assert flux.main_assignments[i].lhs == j_fields[
0].staggered_access(n)
# calculate mean density
dens = (n_fields[0].neighbor_vector(n) +
n_fields[0].center_vector)[0] / 2
# multyply by smoothed haviside function so that fluctuation will not get bigger that the density
dens *= sp.Max(
0,
sp.Min(1.0, n_fields[0].neighbor_vector(n)[0]) *
sp.Min(1.0, n_fields[0].center_vector[0]))
# lenght of the vector
length = sp.sqrt(len(j_fields[0].staggered_stencil[i]))
# amplitude of the random fluctuations
fluct = sp.sqrt(2 * dens * D) * sp.sqrt(
1 / length) * stencil_factor
# add fluctuations
fluct *= 2 * (next(rng_symbol_gen) - 0.5) * sp.sqrt(3)
flux.main_assignments[i] = ps.Assignment(
flux.main_assignments[i].lhs,
flux.main_assignments[i].rhs + fluct)
# Add the folding to the flux, so that the random numbers persist through the ghostlayers.
fold = {
ps.astnodes.LoopOverCoordinate.get_loop_counter_symbol(i):
ps.astnodes.LoopOverCoordinate.get_loop_counter_symbol(i) % L[i]
for i in range(len(L))
}
flux.subs(fold)
r_flux = ps.AssignmentCollection(
[ps.Assignment(j_fields[i].center, 0) for i in range(species)])
reaction = r_rate_const
for i in range(species):
reaction *= sp.Pow(n_fields[i].center, r_order[i])
if (fluctuations):
rng_symbol_gen = random_symbol(r_flux.subexpressions, dim=dh.dim)
reaction_fluctuations = sp.sqrt(
sp.Abs(reaction)) * 2 * (next(rng_symbol_gen) - 0.5) * sp.sqrt(3)
reaction_fluctuations *= sp.Min(1, sp.Abs(reaction**2))
else:
reaction_fluctuations = 0.0
for i in range(species):
r_flux.main_assignments[i] = ps.Assignment(
r_flux_fields[i].center,
(reaction + reaction_fluctuations) * r_coefs[i])
continuity_assignments.append(
ps.Assignment(n_fields[0].center,
n_fields[0].center + r_flux_fields[0].center))
flux_kernel = ps.create_staggered_kernel(flux).compile()
reaction_kernel = ps.create_kernel(r_flux).compile()
pde_kernel = ps.create_kernel(continuity_assignments).compile()
sync_conc = dh.synchronization_function(
[n_fields[0].name, n_fields[1].name])
def f(t, r, n0, fac, fluctuations):
"""Calculates the amount of product created after a certain time of a reaction with form xA -> B
Args:
t: Time of the reation
r: Reaction rate constant
n0: Initial density of the
fac: Reaction order of A (this in most cases equals the stochometric coefficient x)
fluctuations: Boolian whether fluctuations were included during the reaction.
"""
if fluctuations:
return 1 / fac * (n0 + n0 / (n0 - (n0 + 1) * np.exp(fac * r * t)))
return 1 / fac * (n0 - (1 / (fac * r * t + (1 / n0))))
def run(density_init: float, velocity: np.ndarray, time: int):
for i in range(species):
dh.fill(n_fields[i].name,
np.nan,
ghost_layers=True,
inner_ghost_layers=True)
dh.fill(j_fields[i].name,
0.0,
ghost_layers=True,
inner_ghost_layers=True)
dh.fill(r_flux_fields[i].name,
0.0,
ghost_layers=True,
inner_ghost_layers=True)
# set initial values for velocity and density
for i in range(dh.dim):
dh.fill(velocity_field.name,
velocity[i],
i,
ghost_layers=True,
inner_ghost_layers=True)
dh.fill(n_fields[0].name, density_init)
dh.fill(n_fields[1].name, 0.0)
measurement_intervall = 10
data = []
sync_conc()
for i in range(time):
if (i % measurement_intervall == 0):
data.append([
i,
dh.gather_array(n_fields[1].name).mean(),
dh.gather_array(n_fields[0].name).mean()
])
dh.run_kernel(reaction_kernel, seed=41, time_step=i)
for s_idx in range(species):
flux_kernel(n_0=dh.cpu_arrays[n_fields[s_idx].name],
j_0=dh.cpu_arrays[j_fields[s_idx].name],
v=dh.cpu_arrays[velocity_field.name],
seed=42 + s_idx,
time_step=i)
pde_kernel(n_0=dh.cpu_arrays[n_fields[s_idx].name],
j_0=dh.cpu_arrays[j_fields[s_idx].name],
r_0=dh.cpu_arrays[r_flux_fields[s_idx].name])
sync_conc()
data = np.array(data).transpose()
x = data[0]
analytical_value = f(x, r_rate_const, density_init, abs(r_coefs[0]),
fluctuations)
# test mass conservation
np.testing.assert_almost_equal(
dh.gather_array(n_fields[0].name).mean() +
2 * dh.gather_array(n_fields[1].name).mean(), density_init)
r_tol = 2e-3
if fluctuations:
r_tol = 3e-2
np.testing.assert_allclose(data[1], analytical_value, rtol=r_tol)
return lambda density_init, v: run(density_init, np.array(v), time)
def advection_diffusion_fluctuations(dim: int):
# parameters
if dim == 2:
L = (32, 32)
stencil_factor = np.sqrt(1 / (1 + np.sqrt(2)))
elif dim == 3:
L = (16, 16, 16)
stencil_factor = np.sqrt(1 /
(1 + 2 * np.sqrt(2) + 4.0 / 3.0 * np.sqrt(3)))
dh = ps.create_data_handling(domain_size=L,
periodicity=True,
default_target=ps.Target.CPU)
n_field = dh.add_array('n', values_per_cell=1)
j_field = dh.add_array('j',
values_per_cell=3**dim // 2,
field_type=ps.FieldType.STAGGERED_FLUX)
velocity_field = dh.add_array('v', values_per_cell=dim)
D = 0.00666
time = 10000
def grad(f):
return sp.Matrix([ps.fd.diff(f, i) for i in range(dim)])
flux_eq = -D * grad(n_field)
fvm_eq = ps.fd.FVM1stOrder(n_field, flux=flux_eq)
vof_adv = ps.fd.VOF(j_field, velocity_field, n_field)
# merge calculation of advection and diffusion terms
flux = []
for adv, div in zip(vof_adv, fvm_eq.discrete_flux(j_field)):
assert adv.lhs == div.lhs
flux.append(ps.Assignment(adv.lhs, adv.rhs + div.rhs))
flux = ps.AssignmentCollection(flux)
rng_symbol_gen = random_symbol(flux.subexpressions, dim=dh.dim)
for i in range(len(flux.main_assignments)):
n = j_field.staggered_stencil[i]
assert flux.main_assignments[i].lhs == j_field.staggered_access(n)
# calculate mean density
dens = (n_field.neighbor_vector(n) + n_field.center_vector)[0] / 2
# multyply by smoothed haviside function so that fluctuation will not get bigger that the density
dens *= sp.Max(
0,
sp.Min(1.0,
n_field.neighbor_vector(n)[0]) *
sp.Min(1.0, n_field.center_vector[0]))
# lenght of the vector
length = sp.sqrt(len(j_field.staggered_stencil[i]))
# amplitude of the random fluctuations
fluct = sp.sqrt(2 * dens * D) * sp.sqrt(1 / length) * stencil_factor
# add fluctuations
fluct *= 2 * (next(rng_symbol_gen) - 0.5) * sp.sqrt(3)
flux.main_assignments[i] = ps.Assignment(
flux.main_assignments[i].lhs, flux.main_assignments[i].rhs + fluct)
# Add the folding to the flux, so that the random numbers persist through the ghostlayers.
fold = {
ps.astnodes.LoopOverCoordinate.get_loop_counter_symbol(i):
ps.astnodes.LoopOverCoordinate.get_loop_counter_symbol(i) % L[i]
for i in range(len(L))
}
flux.subs(fold)
flux_kernel = ps.create_staggered_kernel(flux).compile()
pde_kernel = ps.create_kernel(
fvm_eq.discrete_continuity(j_field)).compile()
sync_conc = dh.synchronization_function([n_field.name])
# analytical density distribution calculation
def P(rho, density_init):
res = []
for r in rho:
res.append(
np.power(density_init, r) * np.exp(-density_init) /
np.math.gamma(r + 1))
return np.array(res)
def run(density_init: float, velocity: np.ndarray, time: int):
dh.fill(n_field.name,
np.nan,
ghost_layers=True,
inner_ghost_layers=True)
dh.fill(j_field.name,
np.nan,
ghost_layers=True,
inner_ghost_layers=True)
# set initial values for velocity and density
for i in range(dim):
dh.fill(velocity_field.name,
velocity[i],
i,
ghost_layers=True,
inner_ghost_layers=True)
dh.fill(n_field.name, density_init)
measurement_intervall = 10
warm_up = 1000
data = []
sync_conc()
for i in range(warm_up):
dh.run_kernel(flux_kernel, seed=42, time_step=i)
dh.run_kernel(pde_kernel)
sync_conc()
for i in range(time):
dh.run_kernel(flux_kernel, seed=42, time_step=i + warm_up)
dh.run_kernel(pde_kernel)
sync_conc()
if (i % measurement_intervall == 0):
data = np.append(data,
dh.gather_array(n_field.name).ravel(), 0)
# test mass conservation
np.testing.assert_almost_equal(
dh.gather_array(n_field.name).mean(), density_init)
n_bins = 50
density_value, bins = np.histogram(data, density=True, bins=n_bins)
bins_mean = bins[:-1] + (bins[1:] - bins[:-1]) / 2
analytical_value = P(bins_mean, density_init)
print(density_value - analytical_value)
np.testing.assert_allclose(density_value, analytical_value, atol=2e-3)
return lambda density_init, v: run(density_init, np.array(v), time)
def __init__(self, epsilon):
x, y = sp.symbols('x y')
z = sp.Symbol('z')
eps_inv = 1. / epsilon
def phi_L(m, n):
return (1 - epsilon * (m - n))**4 * (4 * epsilon * (m - n) + 1)
def phi_R(m, n):
return phi_L(n, m)
I1_1 = sp.integrate(
phi_L(x, z) * phi_L(y, z), (z, sp.Max(y - eps_inv, 0), x))
I1_2 = sp.integrate(
phi_R(x, z) * phi_L(y, z),
(z, sp.Max(x, y - eps_inv), sp.Min(y, x + eps_inv)))
I1_3 = sp.integrate(
phi_R(x, z) * phi_R(y, z), (z, y, sp.Min(x + eps_inv, 1)))
self.I1_1 = self.functionise_sympy([x, y], I1_1)
self.I1_2 = self.functionise_sympy([x, y], I1_2)
self.I1_3 = self.functionise_sympy([x, y], I1_3)
self.epsilon = epsilon
self.eps_inv = eps_inv
def lap_phi_L(m, n):
return -20 * epsilon**2 * (1 - epsilon *
(m - n))**3 + 60 * epsilon**3 * (
m - n) * (1 - epsilon * (m - n))**2
def lap_phi_R(m, n):
return lap_phi_L(n, m)
I2_1 = sp.integrate(
lap_phi_L(x, z) * phi_L(y, z), (z, sp.Max(y - eps_inv, 0), x))
I2_2 = sp.integrate(
lap_phi_R(x, z) * phi_L(y, z),
(z, sp.Max(x, y - eps_inv), sp.Min(y, x + eps_inv)))
I2_3 = sp.integrate(
lap_phi_R(x, z) * phi_R(y, z), (z, y, sp.Min(x + eps_inv, 1)))
I3_1 = sp.integrate(
phi_L(x, z) * lap_phi_L(y, z), (z, sp.Max(y - eps_inv, 0), x))
I3_2 = sp.integrate(
phi_R(x, z) * lap_phi_L(y, z),
(z, sp.Max(x, y - eps_inv), sp.Min(y, x + eps_inv)))
I3_3 = sp.integrate(
phi_R(x, z) * lap_phi_R(y, z), (z, y, sp.Min(x + eps_inv, 1)))
self.I2_1 = self.functionise_sympy([x, y], I2_1)
self.I2_2 = self.functionise_sympy([x, y], I2_2)
self.I2_3 = self.functionise_sympy([x, y], I2_3)
self.I3_1 = self.functionise_sympy([x, y], I3_1)
self.I3_2 = self.functionise_sympy([x, y], I3_2)
self.I3_3 = self.functionise_sympy([x, y], I3_3)
I4_1 = sp.integrate(
lap_phi_L(x, z) * lap_phi_L(y, z), (z, sp.Max(y - eps_inv, 0), x))
I4_2 = sp.integrate(
lap_phi_R(x, z) * lap_phi_L(y, z),
(z, sp.Max(x, y - eps_inv), sp.Min(y, x + eps_inv)))
I4_3 = sp.integrate(
lap_phi_R(x, z) * lap_phi_R(y, z), (z, y, sp.Min(x + eps_inv, 1)))
self.I4_1 = self.functionise_sympy([x, y], I4_1)
self.I4_2 = self.functionise_sympy([x, y], I4_2)
self.I4_3 = self.functionise_sympy([x, y], I4_3)
self.A = 'A'
self.A_bar = 'A_bar'
self.B = 'B'
self.B_bar = 'B_bar'
self.operators = [self.A, self.B]
self.operators_bar = [self.A_bar, self.B_bar]
def _infer_Slice(self, node):
if get_opset(self.out_mp_) <= 9:
axes = get_attribute(node, 'axes')
starts = get_attribute(node, 'starts')
ends = get_attribute(node, 'ends')
steps = [1] * len(axes)
else:
starts = self._get_value(node, 1)
ends = self._get_value(node, 2)
assert starts is not None and ends is not None
axes = self._try_get_value(node, 3)
steps = self._try_get_value(node, 4)
if axes is None:
axes = list(range(0, len(starts)))
if steps is None:
steps = [1] * len(starts)
new_shape = self._get_sympy_shape(node, 0)
for i, s, e, t in zip(axes, starts, ends, steps):
# TODO: handle step
assert t == 1
idx = handle_negative_axis(i, len(new_shape))
if is_literal(e):
if e >= int(2**31 - 1): # max value of int32
e = new_shape[i]
elif is_literal(new_shape[i]):
e = min(e, new_shape[i])
else:
if e > 0:
e = sympy.Min(e, new_shape[i])
else:
e = new_shape[i] + e
else:
if is_literal(new_shape[i]):
e = sympy.Min(e, new_shape[i])
else:
try:
if e >= new_shape[i]:
e = new_shape[i]
except Exception:
print(
'Unable to determine if {} <= {}, treat as equal'.
format(e, new_shape[i]))
e = new_shape[i]
if is_literal(s) and int(s) < 0:
s = new_shape[i] + s
new_shape[idx] = e - s
vi = self.known_vi_[node.output[0]]
vi.CopyFrom(
helper.make_tensor_value_info(
node.output[0], vi.type.tensor_type.elem_type,
get_shape_from_sympy_shape(new_shape)))
if node.input[0] in self.sympy_data_:
assert [0] == axes
assert len(starts) == 1
assert len(ends) == 1
self.sympy_data_[node.output[0]] = self.sympy_data_[
node.input[0]][starts[0]:ends[0]]
