LMFit

Warnings

• breaking Since version 1.0.3, the `guess()` method of `lmfit.Model` (and built-in models) now explicitly requires the `x` (independent variable) argument, even if it worked without it in older versions. Scripts not providing `x` will raise an error.
  Fix: Ensure `x` (independent variable data) is always passed to `Model.guess(data, x=x_data)`.
• gotcha Fit procedures will stop if `NaN` values are encountered in the objective function or model output. If `NaN`s are present in your input data and are meant to represent missing values, they must be explicitly handled.
  Fix: Use the `nan_policy='omit'` argument when creating a `Model` or calling `Model.fit()` / `lmfit.minimize()` if `NaN`s in data should be ignored. Alternatively, preprocess data to remove or impute `NaN`s.
• gotcha Parameters can sometimes get 'stuck' at their initial values or fail to converge if small changes to their values do not significantly affect the residual (e.g., discrete steps in the model, or initial guesses are extremely far from the true minimum).
  Fix: Provide reasonable initial guesses for parameters. For models with discrete-like transitions, consider using smoother functions (e.g., error functions for steps) or a 'brute' force method for initial scanning. Review the `fit_report` to check for parameters with zero standard error or very small changes.
• deprecated Older Python versions are no longer supported. For instance, `lmfit` version 1.2.0 dropped support for Python 3.6, and version 1.3.3 dropped support for Python 3.8. The current minimum required Python version is 3.9.
  Fix: Upgrade your Python environment to 3.9 or newer to use current `lmfit` versions.
• gotcha When using `lmfit.minimize` directly (rather than `Model.fit`), the `Parameters` object passed as an argument can be modified in-place by the minimization routine. Reusing the same `Parameters` object for multiple fits without explicit copying can lead to unexpected results.
  Fix: Always pass a `copy.deepcopy()` of the `Parameters` object to `minimize()` if you intend to reuse the original `Parameters` for subsequent fits or if a `Minimizer` instance is reused across multiple fits without re-initialization.

Install

• pip install lmfit Install latest version

Imports

• Model

  from lmfit import Model

• Parameters

  from lmfit import Parameters

• minimize

  from lmfit import minimize

Quickstart

This quickstart demonstrates fitting an exponential decay model to noisy data using `lmfit.Model`. It covers defining the model function, creating `Model` and `Parameters` objects, setting initial guesses and constraints, performing the fit, and reporting the results.

import numpy as np
from lmfit import Model

# 1. Generate some data
x = np.linspace(0, 10, 100)
y_true = 3.0 * np.exp(-0.5 * x) + 2.0
np.random.seed(0)
y_data = y_true + np.random.normal(0, 0.2, x.shape)

# 2. Define your model function
def exponential_decay(x, amplitude, decay, offset):
    return amplitude * np.exp(-decay * x) + offset

# 3. Create a Model instance from your function
exp_model = Model(exponential_decay)

# 4. Create initial parameters with guess() or manually
# guess() method often requires x and y data for good initial estimates
params = exp_model.make_params(amplitude=5., decay=0.1, offset=1.)

# Or, for more refined control:
# params = exp_model.guess(y_data, x=x)

# Optionally set bounds or fix parameters
params['amplitude'].set(min=0.0)
params['decay'].set(min=0.0)

# 5. Fit the model to the data
result = exp_model.fit(y_data, params, x=x)

# 6. Print the fitting report
print(result.fit_report())

# You can also access best-fit parameters, statistics, etc.
# print(f"Best-fit amplitude: {result.params['amplitude'].value:.3f}")
# print(f"Reduced Chi-square: {result.redchi:.3f}")