The `nls2` package, released on July 14, 2024, version 0.3-4, introduces brute force and multiple starting values to the `nls` function for nonlinear regression. Authored by G. Grothendieck and the R Core Team, the package enhances the `nls` function by providing additional algorithms such as "brute-force," "grid-search," "random-search," "lhs" (Latin Hypercube Sampling), and "plinear-brute," "plinear-random," and "plinear-lhs" for generating starting values. Key features include: - **Brute Force and Grid Search**: These algorithms generate a grid or random set of starting values and perform an `nls` optimization starting from each point. - **Random Search**: uniform sampling of points within a defined rectangle. - **PLinear Algorithms**: Similar to brute force and random search but for plinear-style formulas. - **CPoptim**: Uses the convex partition algorithm from the `CPoptim` package for optimization. The package supports weighted regression, subset data, and the `all` argument to return a list of `nls` objects or the best fit. Examples demonstrate how to use these features, including handling singular Jacobians and using the `CPoptim` algorithm for complex models. The package is licensed under GPL-2 and is available on CRAN. It does not require compilation and is maintained by G. Grothendieck.The `nls2` package, released on July 14, 2024, version 0.3-4, introduces brute force and multiple starting values to the `nls` function for nonlinear regression. Authored by G. Grothendieck and the R Core Team, the package enhances the `nls` function by providing additional algorithms such as "brute-force," "grid-search," "random-search," "lhs" (Latin Hypercube Sampling), and "plinear-brute," "plinear-random," and "plinear-lhs" for generating starting values. Key features include: - **Brute Force and Grid Search**: These algorithms generate a grid or random set of starting values and perform an `nls` optimization starting from each point. - **Random Search**: uniform sampling of points within a defined rectangle. - **PLinear Algorithms**: Similar to brute force and random search but for plinear-style formulas. - **CPoptim**: Uses the convex partition algorithm from the `CPoptim` package for optimization. The package supports weighted regression, subset data, and the `all` argument to return a list of `nls` objects or the best fit. Examples demonstrate how to use these features, including handling singular Jacobians and using the `CPoptim` algorithm for complex models. The package is licensed under GPL-2 and is available on CRAN. It does not require compilation and is maintained by G. Grothendieck.