| 1 |
+++ |
| 2 |
title = "Custom objective function" |
| 3 |
weight = 20 |
| 4 |
+++ |
| 5 |
|
| 6 |
## Custom objective function |
| 7 |
|
| 8 |
The BornAgain fitting API allows users to define a custom objective function to for the minimization engine. |
| 9 |
|
| 10 |
In this example we are going to construct a vector of residuals calculated between the |
| 11 |
experimental and simulated intensity values after applying an additional $sqrt$ function to the amplitudes. |
| 12 |
|
| 13 |
$$ |
| 14 |
residuals = [r\_{0}, r_{1}, ... , r\_{n-1}], ~~~ r\_{i} = \sqrt{e\_{i}} - \sqrt{s\_{i}} |
| 15 |
$$ |
| 16 |
|
| 17 |
The length of vector `n` corresponds to the total number of *non-masked* detector channels. |
| 18 |
|
| 19 |
This is done by defining a custom residual function `my_residuals`. |
| 20 |
It calls `fit_objective.evaluate(P)` to run the simulation and |
| 21 |
store the intensity arrays, then retrieves them via |
| 22 |
`fit_objective.flatSimData()` and `fit_objective.flatExpData()` |
| 23 |
and applies `sqrt` before differencing. |
| 24 |
The custom function is then passed directly to `lmfit.minimize`. |
| 25 |
|
| 26 |
{{< show-ex file="fit/scatter2d/custom_objective_function.py" >}} |