Summary

Fixes #110.

ExpCurveFitAlgorithm and PowerCurveFitAlgorithm linearize in log space via LinearCurveFitAlgorithm, which fits log(y) = a*f(x) + b and returns sol.u = (a, b). Here b is the intercept in log space — i.e. log of the multiplicative scale factor — not the scale factor itself.

The docstrings and README, however, advertised the models as y = b*exp(a*x) and y = b*x^a with b returned directly. That is inconsistent with the actual (a, log(b)) output, as reported in the issue:

julia> x=[1,2,3]; y=exp.(0.2 .+ 0.3*x);
julia> solve(CurveFitProblem(x,y), ExpCurveFitAlgorithm()).u
(0.3000000000000001, 0.20000000000000048)   # u[2] = 0.2 = log-space intercept, not 2.0·…

The current behavior is intentional and already reflected elsewhere in the codebase:

• docs/src/tutorials/getting_started.md recovers the scale factor with exp(sol.u[2]).
• test/linfit_exp.jl and test/linfit_power.jl assert sol.u[2] ≈ log(2.0).

This PR fixes only the documentation to match that behavior (the non-breaking resolution requested in the issue's "Expected behavior"):

• Docstrings (ExpCurveFitAlgorithm, PowerCurveFitAlgorithm): restate the models as y = exp(a*x + b) and y = exp(a*log(x) + b), matching the linear fit log(y) = a*f(x) + b, and note that the scale factor of the equivalent y = B*exp(a*x) / y = B*x^a form is recovered as B = exp(b).
• README: same correction for the two entries.
• Tutorial prose: clarify the log-space convention (the code blocks were already correct).

Why docs and not code

The returned-parameter semantics are part of the v1 public API and are already pinned by the existing test assertions (sol.u[2] ≈ log(2.0)) and the tutorial's exp(sol.u[2]). Changing the returned values to the multiplicative form would be a breaking change and would also complicate the generic LinearCurveFitAlgorithm abstraction (which returns the raw log-space intercept for any yfun). The issue's stated expected behavior is the documentation fix.

Testing

The behavior is already covered by test/linfit_exp.jl and test/linfit_power.jl, both of which assert sol.u[2] ≈ log(2.0). Ran both locally — 7/7 pass each. Runic reports no changes needed on src/common_interface.jl.

This PR should be ignored until reviewed by @ChrisRackauckas.

🤖 Generated with Claude Code