Skip to content

Latest commit

 

History

History
62 lines (52 loc) · 1.39 KB

README.md

File metadata and controls

62 lines (52 loc) · 1.39 KB

Build Status

ForwardDiffOverMeasurements

Motivation

starting with:

f(x,y) = x*y*exp(x+y) + 2x-2y + x/y

This is fine:

using Measurements
f(x,y) = x*y*exp(x+y) + 2x-2y + x/y
x1 = 1.0 ±0.1
y1 = 2.0 ±0.001
f(x1,y1) #fine

This is also fine:

using ForwardDiff
x1 = 1.0
y1 = 2.0
dfx(x,y) = ForwardDiff.derivative(_x -> f(_x,y),x)
dfx(x1,y1) #also fine

But,this combination is not fine:

using ForwardDiff,Measurements
x1 = 1.0 ±0.1
y1 = 2.0 ±0.001
dfx(x,y) = ForwardDiff.derivative(_x -> f(_x,y),x)
dfx(x1,y1) #NOT FINE

Solution:

using ForwardDiff,Measurements,ForwardDiffOverMeasurements
x1 = 1.0 ±0.1
y1 = 2.0 ±0.001
dfx(x,y) = ForwardDiff.derivative(_x -> f(_x,y),x)
dfx(x1,y1) #fine again

Usage

on the REPL:

>julia ]
(@v1.7) pkg> add ForwardDiffOverMeasurements
>julia using ForwardDiffOverMeasurements

The package loads promote rules and some basic operations that favor a ForwardDiff.Dual over a Measurements.Measurement, so this holds:

f(x::Dual,y::Measurement)::Dual

The package does not export anything nor does define new functions.

At the moment, +, -, *, and / are defined. if any Base function is missing, please open an issue.