Expr
Namespace: SymTensor
expression module
Nested types and modules
| Type | Description |
| ArityT | arity of an op |
| BinaryOpT | ops with two exprs as arguments |
| ChannelT | a channel of a multi-channel op or loop |
| ExprRngSpecT | a simplified range specification of one dimension |
| ExprRngsSpecT | a simplified range specification of all dimensions |
| ExprT | an expression |
| FullExprRngSpecT | |
| FullExprRngsSpecT | |
| IOp | A mathematical operation in an expression. This models a mathematical function or operator that takes one or more tensors and returns one tensor. |
| LeafOpT | ops with no exprs as arguments |
| LoopInputT | a loop variable value specification |
| LoopSpecT | A loop specification. A loop provides iterative evaluation of one or multiple expresisons. A loop can slice over its arguments and reference values computed in previous loop iterations. A loop can compute multiple values at once. Each computed values is referred to as a channel. |
| LoopValueT | the value of a loop channel |
| MultiChannelOpT | an n-ary op with multiple output channels |
| NaryOpT | ops with an arbitrary exprs as arguments |
| PlusElems | start plus the specified number of (symbolic elements) |
| PreviousChannelT | references a loop channel of a previous iteration |
| SequenceArgSliceT | a slice of an argument to the loop |
| UnaryHeldOpT | an op that will expand into an expression once symbolic sizes have been substituted |
| UnaryOpT | ops with one expr as argument |
Functions and values
| Function or value | Description |
annotate ano a
Signature: ano:string -> a:ExprT -> ExprT
|
annotated expression |
arange size
Signature: size:SizeSpecT -> ExprT
Type parameters: 'T |
Vector counting from zero to given size minus one. |
arangeOfType shp typ
Signature: shp:SizeSpecT -> typ:Type -> ExprT
|
Vector counting from zero to given size minus one. |
argMaxAxis ax a
Signature: ax:int -> a:ExprT -> ExprT
|
index of maximum over given dimension |
argMaxKeepingAxis ax a
Signature: ax:int -> a:ExprT -> ExprT
|
index of maximum over given dimension, while keeping the axis with one (broadcastable) element |
argMinAxis ax a
Signature: ax:int -> a:ExprT -> ExprT
|
index of maximum over given dimension |
argMinKeepingAxis ax a
Signature: ax:int -> a:ExprT -> ExprT
|
index of maximum over given dimension, while keeping the axis with one (broadcastable) element |
assumeJacobian jac expr
Signature: jac:ExprT -> expr:ExprT -> ExprT
|
assumes the specified Jacobian when calculating derivatives |
assumeZeroDerivative expr
Signature: expr:ExprT -> ExprT
|
nullifies the Jacobian when calculating derivatives |
boxOption oo
Signature: oo:'?177823 option -> obj option
Type parameters: '?177823 |
boxes the contents of an option |
broadcast ss a
Signature: ss:ShapeSpecT -> a:ExprT -> ExprT
|
broadcast of SizeBroadcast dimensions |
broadcastIfNecessary ss expr
Signature: ss:ShapeSpecT -> expr:ExprT -> ExprT
|
Wraps the given op in a Broadcast op if its shape does not match ss. |
broadcastToSame a b
Signature: a:ExprT -> b:ExprT -> ExprT * ExprT
|
pads and broadcasts |
broadcastToSameMany es
Signature: es:ExprT list -> ExprT list
|
pads and broadcasts all arguments to same shape if possible |
broadcastToShape shp a
Signature: shp:ShapeSpecT -> a:ExprT -> ExprT
|
pads from the left and broadcasts the argument to the given shape if possible |
cage (minVal, maxVal) a
Signature: (minVal:'?178102 option * maxVal:'?178103 option) -> a:ExprT -> ExprT
Type parameters: '?178102, '?178103 |
Ensures that all elements are between Some minVal and Some maxVal. |
canEvalAllSymSizes expr
Signature: expr:ExprT -> bool
|
true if all shapes in the expression can be evaluated to numeric shapes |
check expr
Signature: expr:ExprT -> ExprT
|
Traverses the expression and checks ops' arguments for compatible shapes. |
checkAxis ax expr
Signature: ax:int -> expr:ExprT -> unit
|
checks that given axis is valid for specified expression |
checkedExprs
Signature: HashSet<ExprT>
|
expressions that were already checked for correctness |
checkExpr expr
Signature: expr:ExprT -> unit
|
Checks ops' arguments for compatible shapes. |
checkFinite name a
Signature: name:string -> a:ExprT -> ExprT
|
checks the value for NaNs and infinities, outputs their location and stops the computation |
concat dim es
Signature: dim:int -> es:seq<ExprT> -> ExprT
|
concatenates the sequence of tensors in the specified dimension |
constructElementwise op a b
Signature: op:BinaryOpT -> a:ExprT -> b:ExprT -> ExprT
|
emits an elementwise binary operation with broadcasting of the inputs if necessary |
contains subExpr expr
Signature: subExpr:ExprT -> expr:ExprT -> bool
|
returns true if subExpr is contained in expr |
countOps expr
Signature: expr:ExprT -> int
|
counts operators, including repeating subexpressions |
countUniqueOps expr
Signature: expr:ExprT -> int
|
counts operators, not counting repeating subexpressions |
diag a
Signature: a:ExprT -> ExprT
|
Extracts the diagonal of a matrix. If the expression has more than two dimensions, the diagonals are extracted along the last two dimensions. |
diagAxis ax1 ax2 a
Signature: ax1:int -> ax2:int -> a:ExprT -> ExprT
|
Extracts the diagonal along the given axes. |
diagMat a
Signature: a:ExprT -> ExprT
|
Creates a matrix with the given vector on its diagonal. All other elements are zeros. If the input has more than one dimension, the operation is performed batch-wise on the last dimension. |
diagMatAxis ax1 ax2 a
Signature: ax1:int -> ax2:int -> a:ExprT -> ExprT
|
Creates a diagonal matrix by duplicating the given dimension. |
disableBroadcast dim a
Signature: dim:int -> a:ExprT -> ExprT
|
disables broadcasting in the given dimension |
discard es
Signature: es:ExprT list -> ExprT
|
computes specified expressions, but discards the result |
dot a b
Signature: a:ExprT -> b:ExprT -> ExprT
|
Dot product. Behavior depends on the dimensionality of the arguments. Cases: (1, 1) -> vector-vector dot product resulting in a scalar (2, 1) -> matrix-vector dot product resulting in a vector (2, 2) -> matrix-matrix dot product resulting in a matrix (n, n) with n>2 -> batched matrix-matrix dot product resulting in a matrix (n+1, n) with n>2 -> batched matrix-vector dot product resulting in a vector. |
dump name a
Signature: name:string -> a:ExprT -> ExprT
|
dumps the result into the active dump session HDF5 file |
elements trgtShp elemExpr args
Signature: trgtShp:ShapeSpecT -> elemExpr:ElemExprT -> args:ExprT list -> ExprT
|
calculates a tensor elementwise using the given element expression and result shape |
enableBroadcast dim a
Signature: dim:int -> a:ExprT -> ExprT
|
enables broadcasting in the given dimension, it must be of size one |
extractVar expr
Signature: expr:ExprT -> VarSpecT
|
extract VarSpec from variable expression |
extractVars expr
Signature: expr:ExprT -> Set<VarSpecT>
|
extract all variables from an expression |
failOnNotEvalableSymSize expr
Signature: expr:ExprT -> unit
|
fails if the expression contains a shape that cannot be evaluated to a numeric shape |
failshape op sa sb
Signature: op:'?177979 -> sa:'?177980 -> sb:'?177981 -> '?177982
Type parameters: '?177979, '?177980, '?177981, '?177982 |
Produces an error message about incompatible shapes. |
filled shp value
Signature: shp:ShapeSpecT -> value:'?178132 -> ExprT
Type parameters: '?178132 |
tensor of given shape filled with specified value |
gather indices a
Signature: indices:ExprT option list -> a:ExprT -> ExprT
|
select elements according to the specified index arrays |
identity size
Signature: size:SizeSpecT -> ExprT
Type parameters: 'T |
identity matrix of given size |
identityOfSameType expr size
Signature: expr:ExprT -> size:SizeSpecT -> ExprT
|
identity matrix of given size and same type as given expression |
ifThenElse cond ifTrue ifFalse
Signature: cond:ExprT -> ifTrue:ExprT -> ifFalse:ExprT -> ExprT
|
elementwise uses elements from ifTrue if cond is true, otherwise elements from ifFalse |
insertBroadcastAxis dim a
Signature: dim:int -> a:ExprT -> ExprT
|
inserts a broadcast axis at the given dimension |
interpolate interpolator e
Signature: interpolator:InterpolatorT -> e:ExprT list -> ExprT
|
Element-wise n-dimensional interpolation using the specified interpolator. The interpolator is created using the Interpolator.create function. |
interpolate1D interpolator a
Signature: interpolator:InterpolatorT -> a:ExprT -> ExprT
|
Element-wise one-dimensional interpolation using the specified interpolator. The interpolator is created using the Interpolator.create function. |
interpolate2D interpolator a b
Signature: interpolator:InterpolatorT -> a:ExprT -> b:ExprT -> ExprT
|
Element-wise two-dimensional interpolation using the specified interpolator. The interpolator is created using the Interpolator.create function. |
interpolate3D interpolator a b c
Signature: interpolator:InterpolatorT -> a:ExprT -> b:ExprT -> c:ExprT -> ExprT
|
Element-wise three-dimensional interpolation using the specified interpolator. The interpolator is created using the Interpolator.create function. |
invert a
Signature: a:ExprT -> ExprT
|
Computes the inverse of a matrix. If the input has more than two dimensions, the inverses along the last two dimensions are returned. The inverse of a singular matrix is undefinied. No error is raised in that case. |
isZero expr
Signature: expr:ExprT -> bool
|
True if expression is zero. False does not indicate that expression is non-zero. |
loop spec channel args
Signature: spec:LoopSpecT -> channel:ChannelT -> args:ExprT list -> ExprT
|
A loop provides iterative evaluation of one or multiple expresisons. |
loopNoLift spec channel args
Signature: spec:LoopSpecT -> channel:ChannelT -> args:ExprT list -> ExprT
|
A loop provides iterative evaluation of one or multiple expresisons.
All variables occurs in the loop channel expressions must be defined as loop variables.
The function |
makeVar vs
Signature: vs:VarSpecT -> ExprT
|
make variable expression from VarSpec |
mapOperands (...)
Signature: unaryMapping:(UnaryOpT -> ExprT -> ExprT) -> binaryMapping:(BinaryOpT -> ExprT -> ExprT -> ExprT * ExprT) -> naryMapping:(NaryOpT -> ExprT list -> ExprT list) -> expr:ExprT -> ExprT
|
Traverses the op tree and for each op calls a function on its arguments and replaces them by the function's return value(s). |
maxAxis ax a
Signature: ax:int -> a:ExprT -> ExprT
|
maximum over given dimension |
maxElemwise a b
Signature: a:ExprT -> b:ExprT -> ExprT
|
elementwise maximum |
maxKeepingAxis ax a
Signature: ax:int -> a:ExprT -> ExprT
|
maximum over given dimension, while keeping the axis with one (broadcastable) element |
mean a
Signature: a:ExprT -> ExprT
|
mean over all elements |
meanAxis ax a
Signature: ax:int -> a:ExprT -> ExprT
|
mean over given dimension |
meanKeepingAxis ax a
Signature: ax:int -> a:ExprT -> ExprT
|
mean over given dimension, while keeping the axis with one (broadcastable) element |
minAxis ax a
Signature: ax:int -> a:ExprT -> ExprT
|
maximum over given dimension |
minElemwise a b
Signature: a:ExprT -> b:ExprT -> ExprT
|
elementwise minimum |
minKeepingAxis ax a
Signature: ax:int -> a:ExprT -> ExprT
|
maximum over given dimension, while keeping the axis with one (broadcastable) element |
nDims expr
Signature: expr:ExprT -> int
|
number of dimensions of given expression |
nElems expr
Signature: expr:ExprT -> SizeSpecT
|
number of elements of given expression |
oneOfSameType expr
Signature: expr:ExprT -> ExprT
|
scalar 1 of the same type as given expression |
padLeft a
Signature: a:ExprT -> ExprT
|
adds one broadcastable dimension to the left |
padRight a
Signature: a:ExprT -> ExprT
|
adds one broadcastable dimension to the right |
permuteAxes perm a
Signature: perm:int list -> a:ExprT -> ExprT
|
Permutes the axes as specified. Each entry in the specified permutation specifies the new position of the corresponding axis, i.e. to which position the axis should move. |
print msg a
Signature: msg:string -> a:ExprT -> ExprT
|
print the result with the given message when evaluated |
product a
Signature: a:ExprT -> ExprT
|
product of all elements |
productAxis ax a
Signature: ax:int -> a:ExprT -> ExprT
|
product over given dimension |
productKeepingAxis ax a
Signature: ax:int -> a:ExprT -> ExprT
|
product over given dimension, while keeping the axis with one (broadcastable) element |
replicate dim reps a
Signature: dim:int -> reps:SizeSpecT -> a:ExprT -> ExprT
|
Replicates the tensor the given number of repetitions along the given axis. |
replicateTo dim size a
Signature: dim:int -> size:SizeSpecT -> a:ExprT -> ExprT
|
Replicates the tensor along the given axis, so that after replication it has
the specified |
reshape ss a
Signature: ss:ShapeSpecT -> a:ExprT -> ExprT
|
reshape (assuming C-continguous order) tensor; element count does not change |
reshapeIfNecessary ss expr
Signature: ss:ShapeSpecT -> expr:ExprT -> ExprT
|
Wraps the given op in a Reshape op if its shape does not match ss. |
reverseAxis dim a
Signature: dim:int -> a:ExprT -> ExprT
|
reverses the tensor in the given dimension |
scalar f
Signature: f:'?178014 -> ExprT
Type parameters: '?178014 |
scalar constant of given value |
scalarOfSameType expr f
Signature: expr:ExprT -> f:'a -> ExprT
Type parameters: 'a |
scalar of given value converted to same type as given expression |
scatter indices trgtShp a
Signature: indices:ExprT option list -> trgtShp:ShapeSpecT -> a:ExprT -> ExprT
|
select elements according to the specified index arrays |
setSubtensor a b
Signature: a:ExprT -> b:ExprT -> ExprT
|
expression a with the specified subtensor replaced with b |
shapeOf expr
Signature: expr:ExprT -> ShapeSpecT
|
Returns the shape of the given expression. |
signt a
Signature: a:ExprT -> ExprT
|
sign keeping type |
sizeValue size
Signature: size:SizeSpecT -> ExprT
Type parameters: 'T |
scalar with value of given size converted to the given type |
sizeValueOfSameType expr size
Signature: expr:ExprT -> size:SizeSpecT -> ExprT
|
scalar with value of given size converted to the same type as given expression |
sqrtt a
Signature: a:ExprT -> ExprT
|
square root |
storeToVar ve a
Signature: ve:ExprT -> a:ExprT -> ExprT
|
store to variable |
subst replacements expr
Signature: replacements:Map<ExprT,ExprT> -> expr:ExprT -> ExprT
|
Replaces all occurences of the map key with its value in the specified expression. Does not replace subexpressions within loop channel value expressions. |
substSymSizes symSizes expr
Signature: symSizes:Map<SizeSymbolT,SizeSpecT> -> expr:ExprT -> ExprT
|
substitues the given symbol sizes into the expression |
sum a
Signature: a:ExprT -> ExprT
|
summaiton of all elements |
sumAxis ax a
Signature: ax:int -> a:ExprT -> ExprT
|
summation over given dimension |
sumKeepingAxis ax a
Signature: ax:int -> a:ExprT -> ExprT
|
summation over given dimension, while keeping the axis with one (broadcastable) element |
swapDim ax1 ax2 a
Signature: ax1:int -> ax2:int -> a:ExprT -> ExprT
|
swaps two dimensions of a tensor |
tensorProduct a b
Signature: a:ExprT -> b:ExprT -> ExprT
|
tensor product |
trace a
Signature: a:ExprT -> ExprT
|
Computes the trace of a matrix. If the input has more than two dimensions, the traces along the last two dimensions are returned. |
traceAxis ax1 ax2 a
Signature: ax1:int -> ax2:int -> a:ExprT -> ExprT
|
Computes the traces along the given axes. |
transpose a
Signature: a:ExprT -> ExprT
|
Transpose matrix. If the input has more than two dimensions, the last two axes are transposed. |
twoOfSameType expr
Signature: expr:ExprT -> ExprT
|
scalar 2 of the same type as given expression |
typename expr
Signature: expr:ExprT -> TypeNameT
|
Returns the type of the given expression. |
var name ss
Signature: name:string -> ss:ShapeSpecT -> ExprT
Type parameters: 'T |
variable of given name and shape |
varOfType name typ ss
Signature: name:string -> typ:Type -> ss:ShapeSpecT -> ExprT
|
variable of given name, type and shape |
zeroOfSameType expr
Signature: expr:ExprT -> ExprT
|
scalar 0 of the same type as given expression |
zeros shp
Signature: shp:ShapeSpecT -> ExprT
Type parameters: 'T |
zero tensor of given shape |
zerosLike expr
Signature: expr:ExprT -> ExprT
|
zero tensor with same shape and type as given tensor |
zerosOfSameType expr shp
Signature: expr:ExprT -> shp:ShapeSpecT -> ExprT
|
zero tensor of given shape and same type as given expression |
zerosOfType typ shp
Signature: typ:Type -> shp:ShapeSpecT -> ExprT
|
zero tensor of given type and shape |
Active patterns
| Active pattern | Description |
( |BinaryElemwiseOp|_| ) bop
Signature: bop:BinaryOpT -> unit option
|
matches all binary ops that work elementwise CompiledName: |
( |UnaryElemwiseOp|_| ) uop
Signature: uop:UnaryOpT -> unit option
|
matches all unary ops that work elementwise CompiledName: |
( |ZeroExpr|_| ) expr
Signature: expr:ExprT -> unit option
|
Matches expressions with value zero. CompiledName: |
