Deep.Net


DerivCheck

Namespace: SymTensor

Functions and values

Function or valueDescription
checkExpr (...)
Signature: device:IDevice -> maxDeviation:^T -> epsilon:^T -> varEnv:VarEnvT -> expr:ExprT -> unit
Type parameters: ^T

Checks that symbolic and numeric derivatives of the given expression are close enough. The derivatives are evaluated at the location specified by the given VarEnv.

checkExprTree (...)
Signature: device:IDevice -> maxDeviation:^T -> epsilon:^T -> varEnv:VarEnvT -> expr:ExprT -> unit
Type parameters: ^T

Recursively checks that symbolic and numeric derivatives of all ops in the given expression are close enough. The derivatives are evaluated at the location specified by the given VarEnv.

numDeriv f x
Signature: f:(Tensor<^T> -> Tensor<^T>) -> x:Tensor<^T> -> Tensor<^T>
Type parameters: ^T

evaluates the Jacobian of f at x numerically

numDerivEpsilon epsilon f x
Signature: epsilon:^T -> f:(Tensor<^T> -> Tensor<^T>) -> x:Tensor<^T> -> Tensor<^T>
Type parameters: ^T

evaluates the Jacobian of f at x numerically with specified finite difference step

Fork me on GitHub