Saving and Loading Fit Results

Fit results can be persisted to disk using save_fit and reloaded with load_fit. Results are stored as JLD2 files. Because the DataModel contains runtime-generated closures (from the @Model macro) that cannot be serialized, these are stripped on save and must be reconstructed on load. All numerical content — parameters, objectives, chains, empirical Bayes modes — is preserved exactly.

Basic Usage

using NoLimits

res = fit_model(dm, NoLimits.Laplace())

# Save to disk
save_fit("my_fit.jld2", res)

# Load in the same session — supply dm so RE/plotting accessors work
res2 = load_fit("my_fit.jld2"; dm=dm)

get_params(res2; scale=:untransformed)
get_random_effects(res2)

If dm is not provided, the result still loads successfully but accessors that require the data model — get_random_effects, get_loglikelihood, and all plotting functions — will raise an informative error.

Reloading in a New Session

Pass include_data=true to embed the original DataFrame and DataModelConfig in the file. When loading, only the model is then required — the DataModel is reconstructed automatically.

# Save with data embedded
save_fit("my_fit.jld2", res; include_data=true)

# --- new Julia session ---
include("model.jl")          # defines build_model(...)
model = build_model(...)

res2 = load_fit("my_fit.jld2"; model=model)
dm2  = get_data_model(res2)  # reconstructed from saved DataFrame

get_params(res2; scale=:untransformed)
get_random_effects(res2)

If both dm and model are provided to load_fit, dm takes precedence over the saved DataFrame.

Multistart Results

save_fit and load_fit work identically for MultistartFitResult. All individual run results are preserved.

res_ms = fit_model(dm, ms, NoLimits.Laplace())
save_fit("multistart.jld2", res_ms; include_data=true)

# --- new Julia session ---
model   = build_model(...)
res_ms2 = load_fit("multistart.jld2"; model=model)

best = get_multistart_best(res_ms2)
get_params(best; scale=:untransformed)
plot_multistart_waterfall(res_ms2)

Note: get_multistart_errors returns Vector{String} after loading rather than Vector{Exception}, since exception objects are not serializable.

What Is Preserved

API

save_fit(path, res; include_data=false) -> path::String
load_fit(path; model=nothing, dm=nothing) -> FitResult or MultistartFitResult