pastas.model.Model.solve#

Method to solve the time series model.

Parameters:

tmin (pandas.Timestamp or str, optional) – A string or pandas.Timestamp with the start date for the simulation period (E.g. ‘1980-01-01 00:00:00’). Strings are converted to pandas.Timestamp internally. If none is provided, the tmin from the oseries is used.
tmax (pandas.Timestamp or str, optional) – A string or pandas.Timestamp with the end date for the simulation period (E.g. ‘2020-01-01 00:00:00’). Strings are converted to pandas.Timestamp internally. If none is provided, the tmax from the oseries is used.
freq (str, optional) – String with the frequency the stressmodels are simulated. Must be one of the following (D, h, m, s, ms, us, ns) or a multiple of that e.g. “7D”.
warmup (float, optional) – Warmup period (in Days) for which the simulation is calculated, but not used for the calibration period.
solver (Class pastas.solver.Solver, optional) – Instance of a pastas Solver class used to solve the model. Options are: ps.LeastSquares() (default) or ps.LmfitSolve(). An instance is needed as of Pastas 0.23, not a class!
report (bool | Literal["full"], optional) – Print a report to the screen after optimization finished. Set to True (default) to print a standard report, “full” to print a report including the correlation matrix and standard errors of the parameters, or False to suppress the report. This can also be manually triggered after optimization by calling print(ml.fit_report( )) on the Pastas model instance.
initial (bool, optional) – Reset initial parameters from the individual stress models. Default is True. If False, the optimal values from an earlier optimization are used.
weights (pandas.Series, optional) – A pandas Series used to scale the residual or noise (in the case of a NoiseModel) during optimization. The weights must share the same DateTimeIndex as the observations (ml.observations()) to ensure proper alignment. These weights are applied such that the minimized objective function in least-squares solvers is sum((weights * residuals)**2). This means that a residual with double the weight has four times as much influence. If None, equal weights are used. This can be used to put extra/less weight on certain periods (e.g., droughts) or measurements (i.e. outliers), and make more complex calibration schemes (see, for example, []). Note that the weights are only used during optimization and not when computing the goodness-of-fit metrics.
fit_constant (bool, optional) – Argument that determines if the constant is fitted as a parameter. If it is set to False, the constant is set equal to the mean of the residuals.
freq_obs (str, optional) – String with the frequency of the observations that the model will be calibrated on. Must be one of the following (D, h, m, s, ms, us, ns) or a multiple of that e.g. “7D”. Should generally be larger than the frequency of the original observations and the model frequency (freq). If freq_obs is not set, the frequency of the model (freq) will be used.
initialize (bool, optional) – If True, the model is initialized via the Model.initialize() method (setting certain model settings) before solving. If False, the model is not initialized before solving. Note that the latter is an advanced option since some model settings can be missing. Default is False and deprecated since version 2.0.0.
reset_settings (bool = False,) – If True, the model settings are reset to their default values before solving. This calls the Model.set_settings() method with default values. Default is False.
**kwargs (dict, optional) – All keyword arguments will be passed onto minimization method from the solver. It depends on the solver used which arguments can be used.

Notes

The solver instance including some results are stored as ml.solver. From here one can access the covariance (ml.solver.pcov) and correlation matrix ( ml.solver.pcor).
Each solver returns a number of results after optimization. These solver specific results are stored in ml.solver.result and can be accessed from there.