Deep.Net


Optimizer

Namespace: SymTensor

Nested types and modules

TypeDescription
BroadcastInfoT

Broadcast information

Functions and values

Function or valueDescription
axesBroadcasted expr
Signature: expr:ExprT -> BroadcastInfoT list

Returns a list containing one element each axis of the expression. The element is true if the axis is broadcasted.

binaryOpToElemOp op
Signature: op:BinaryOpT -> BinaryOpT option
broadcastInsignificantElementsAxes (...)
Signature: elements:ExprT -> ExprT

replaces result dimensions that compute identical values for all elements by broadcasting

combineIntoElementsRec exprInfo expr
Signature: exprInfo:ExprInfoT -> expr:ExprT -> ExprT

combines elemwise and elements operations into one elements operation

elementsAxesSignificancy elements
Signature: elements:ExprT -> seq<bool>

Returns a list containing one element for each axis in the result of the elements expression. The element is true if the result can have different values over the corresponding axis. I.e. an axis is significant if values change depending on it.

leafOpToElemOp op
Signature: op:LeafOpT -> LeafOpT option
optimize exprs
Signature: exprs:ExprT list -> ExprT list

Optimizes a group of expressions.

optimizeElements elements
Signature: elements:ExprT -> ExprT

optimizes elements expression in an expression

optimizeElemExpr elemExpr
Signature: elemExpr:ElemExprT -> ElemExprT

optimizes an element expression

pullSumOutOfElements elements
Signature: elements:ExprT -> ExprT

pulls summations out of elements expressions

unaryOpToElemOp op
Signature: op:UnaryOpT -> UnaryOpT option
Fork me on GitHub