Operations¶

Operation classes¶

class Op(fn, *, name=None)[source]¶: Bases: multipledispatch.dispatcher.Dispatcher

class TransformOp(fn, *, name=None)[source]¶

Bases: funsor.ops.op.Op

set_inv(fn)[source]¶

Parameters:	fn (callable) – A function that inputs an arg `y` and outputs a value `x` such that `y=self(x)`.

set_log_abs_det_jacobian(fn)[source]¶

Parameters:	fn (callable) – A function that inputs two args `x, y`, where `y=self(x)`, and returns `log(abs(det(dy/dx)))`.

static inv(x)[source]¶

static log_abs_det_jacobian(x, y)[source]¶

Builtin operations¶

abs = ops.abs¶

add = ops.add¶

and_ = ops.and_¶

eq = ops.eq¶

exp = ops.exp[source]¶

ge = ops.ge¶

getitem = ops.GetitemOp(0)¶: Op encoding an index into one dimension, e.g. x[:,:,y] for offset of 2.

gt = ops.gt¶

invert = ops.invert¶

le = ops.le¶

log = ops.log[source]¶

log1p = ops.log1p¶

lt = ops.lt¶

matmul = ops.matmul¶

max = ops.max[source]¶

min = ops.min[source]¶

mul = ops.mul¶

ne = ops.ne¶

neg = ops.neg¶

nullop = ops.nullop[source]¶: Placeholder associative op that unifies with any other op

or_ = ops.or_¶

pow = ops.pow¶

reciprocal = ops.reciprocal[source]¶

safediv = ops.safediv[source]¶

safesub = ops.safesub[source]¶

sigmoid = ops.sigmoid¶

sqrt = ops.sqrt¶

sub = ops.sub¶

truediv = ops.truediv¶

xor = ops.xor¶

class AssociativeOp(fn, *, name=None)[source]¶: Bases: funsor.ops.op.Op

class AddOp(fn, *, name=None)[source]¶: Bases: funsor.ops.builtin.AssociativeOp

class MulOp(fn, *, name=None)[source]¶: Bases: funsor.ops.builtin.AssociativeOp

class MatmulOp(fn, *, name=None)[source]¶: Bases: funsor.ops.op.Op

class SubOp(fn, *, name=None)[source]¶: Bases: funsor.ops.op.Op

class NegOp(fn, *, name=None)[source]¶: Bases: funsor.ops.op.Op

class DivOp(fn, *, name=None)[source]¶: Bases: funsor.ops.op.Op

class NullOp(fn, *, name=None)[source]¶

Bases: funsor.ops.builtin.AssociativeOp

Placeholder associative op that unifies with any other op

class GetitemOp(offset)[source]¶

Bases: funsor.ops.op.Op

Op encoding an index into one dimension, e.g. x[:,:,y] for offset of 2.

class ExpOp(fn, *, name=None)[source]¶: Bases: funsor.ops.op.TransformOp

class LogOp(fn, *, name=None)[source]¶: Bases: funsor.ops.op.TransformOp

class ReciprocalOp(fn, *, name=None)[source]¶: Bases: funsor.ops.op.Op

Array operations¶

all = ops.all¶

amax = ops.amax¶

amin = ops.amin¶

any = ops.any¶

astype = ops.astype¶

cat = ops.cat¶

cholesky = ops.cholesky¶

cholesky_inverse = ops.cholesky_inverse¶

cholesky_solve = ops.cholesky_solve¶

clamp = ops.clamp¶

detach = ops.detach¶

diagonal = ops.diagonal¶

einsum = ops.einsum¶

expand = ops.expand¶

finfo = ops.finfo¶

full_like = ops.full_like¶

is_numeric_array = ops.is_numeric_array¶

logaddexp = ops.logaddexp¶

logsumexp = ops.logsumexp¶

new_arange = ops.new_arange¶

new_eye = ops.new_eye¶

new_zeros = ops.new_zeros¶

permute = ops.permute¶

prod = ops.prod¶

sample = ops.sample¶

stack = ops.stack¶

sum = ops.sum¶

transpose = ops.transpose¶

triangular_solve = ops.triangular_solve¶

unsqueeze = ops.unsqueeze¶

class LogAddExpOp(fn, *, name=None)[source]¶: Bases: funsor.ops.builtin.AssociativeOp

class SampleOp(fn, *, name=None)[source]¶: Bases: funsor.ops.array.LogAddExpOp

class ReshapeOp(shape)[source]¶: Bases: funsor.ops.op.Op