Fits a user-defined nonlinear model
to each bottle in a rumen_gp dataset.
Value
A custom_fit object containing:
Model name
Formula
Starting values
Parameter bounds
Parameter estimates
Model diagnostics
Predicted values
Residuals
Details
Overview
This function allows researchers to fit
custom nonlinear kinetic equations using
minpack.lm::nlsLM().
Custom models integrate directly with:
making them fully compatible with the rumenGP modeling framework.
Formula Requirements
The model formula must:
Use
Gas_mLas the response variableUse
Time_has the time variableInclude all parameters listed in
start
Example:
Gas_mL ~
A *
(
Time_h /
(
Time_h + K
)
)Starting Values
Starting values are supplied through
start.
Example:
start = list(
A = 150,
K = 10
)Good starting values often improve convergence and reduce fitting failures.
Examples
files <- example_data()
raw_data <- read_ankom(
files$ankom
)
metadata <- read_metadata(
files$metadata
)
gp <- process_ankom(
raw_data,
metadata,
headspace_ml = 210,
temperature_c = 39
)
gp_clean <- exclude_heads(
gp,
heads = c(
"1",
"2",
"4",
"20"
),
reason = c(
"Abnormal gas profile",
"Abnormal gas profile",
"Abnormal gas profile",
"Abnormal gas profile"
)
)
# Hyperbolic model
custom_fit <- fit_custom(
data = gp_clean,
formula =
Gas_mL ~
A *
(
Time_h /
(
Time_h + K
)
),
start = list(
A = 150,
K = 10
),
lower = c(
A = 0,
K = 0
),
model_name = "Hyperbolic"
)
#> Minor negative pressure values detected. Minimum PSI = -0.255 . These may reflect normal sensor variation.
#> rumenGP data validation passed.
#> Observations: 3038
#> Heads: 21
#> Treatments: 5
summary(custom_fit)
#>
#> Custom model summary
#> --------------------
#> Model name: Hyperbolic
#>
#> Formula:
#> Gas_mL ~ A * (Time_h/(Time_h + K))
#>
#> Total bottles: 21
#> Successful fits: 21
#> Failed fits: 0
#> Mean R-squared: 0.9951
#> Mean RMSE: 1.2719
#> Mean AIC: 468.4219
#> Mean BIC: 477.345
#>
# Find a bottle with a successful fit
successful_head <-
custom_fit$diagnostics |>
dplyr::filter(
Converged
) |>
dplyr::slice(1) |>
dplyr::pull(
Head
)
plot_fit(
custom_fit,
head = successful_head
)
plot_residuals(
custom_fit,
head = successful_head
)
#> `geom_smooth()` using formula = 'y ~ x'
# Compare against built-in models
compare_models(
Groot = fit_groot(gp_clean),
BurrXII = fit_burr_xii(gp_clean),
Hyperbolic = custom_fit
)
#> Minor negative pressure values detected. Minimum PSI = -0.255 . These may reflect normal sensor variation.
#> rumenGP data validation passed.
#> Observations: 3038
#> Heads: 21
#> Treatments: 5
#> Minor negative pressure values detected. Minimum PSI = -0.255 . These may reflect normal sensor variation.
#> rumenGP data validation passed.
#> Observations: 3038
#> Heads: 21
#> Treatments: 5
#> Model Bottles Successful_Fits Failed_Fits Mean_R2 Mean_RMSE Mean_RSS
#> 1 Groot 21 21 0 0.9967168 1.0031864 152.2996
#> 2 BurrXII 21 21 0 0.9977206 0.8545103 116.2486
#> 3 Hyperbolic 21 21 0 0.9950576 1.2718832 264.6107
#> Mean_AIC Mean_BIC Lambda_Boundary
#> 1 409.1374 421.0085 0
#> 2 360.5435 375.4154 0
#> 3 468.4219 477.3450 0