smash.default_optimize_options#

smash.default_optimize_options(model, mapping='uniform', optimizer=None)[source]#

Default optimization options of Model.

Parameters:
modelModel

Primary data structure of the hydrological model smash.

mappingstr, default ‘uniform’

Type of mapping. Should be one of

  • 'uniform'

  • 'distributed'

  • 'multi-linear'

  • 'multi-power'

  • 'ann'

Hint

See the Mapping section.

optimizerstr or None, default None

Name of optimizer. Should be one of

  • 'sbs' (only for 'uniform' mapping)

  • 'nelder-mead' (only for 'uniform' mapping)

  • 'powell' (only for 'uniform' mapping)

  • 'lbfgsb' (for all mappings except 'ann')

  • 'adam' (for all mappings)

  • 'adagrad' (for all mappings)

  • 'rmsprop' (for all mappings)

  • 'sgd' (for all mappings)

Note

If not given, a default optimizer will be set as follows:

  • 'sbs' for mapping = 'uniform'

  • 'lbfgsb' for mapping = 'distributed', 'multi-linear', 'multi-power'

  • 'adam' for mapping = 'ann'

Hint

See the Optimization Algorithms section.

Returns:
optimize_optionsdict[str, Any]

Dictionary containing optimization options for fine-tuning the optimization process. The specific keys returned depend on the chosen mapping and optimizer. This dictionary can be directly passed to the optimize_options argument of the optimize (or Model.optimize) method.

Examples

>>> from smash.factory import load_dataset
>>> setup, mesh = load_dataset("cance")
>>> model = smash.Model(setup, mesh)

Get the default optimization options for 'uniform' mapping

>>> opt_u = smash.default_optimize_options(model, mapping="uniform")
>>> opt_u
{
    'parameters': ['cp', 'ct', 'kexc', 'llr'],
    'bounds': {
        'cp': (1e-06, 1000.0),
        'ct': (1e-06, 1000.0),
        'kexc': (-50, 50),
        'llr': (1e-06, 1000.0)
    },
    'control_tfm': 'sbs',
    'termination_crit': {'maxiter': 50},
}

Directly pass this dictionary to the optimize_options argument of the optimize (or Model.optimize) method. It’s equivalent to set optimize_options to None (which is the default value)

>>> model_u = smash.optimize(model, mapping="uniform", optimize_options=opt_u)
</> Optimize
    At iterate     0    nfg =     1    J = 6.95010e-01    ddx = 0.64
    At iterate     1    nfg =    30    J = 9.84107e-02    ddx = 0.64
    At iterate     2    nfg =    59    J = 4.54087e-02    ddx = 0.32
    At iterate     3    nfg =    88    J = 3.81818e-02    ddx = 0.16
    At iterate     4    nfg =   117    J = 3.73617e-02    ddx = 0.08
    At iterate     5    nfg =   150    J = 3.70873e-02    ddx = 0.02
    At iterate     6    nfg =   183    J = 3.68004e-02    ddx = 0.02
    At iterate     7    nfg =   216    J = 3.67635e-02    ddx = 0.01
    At iterate     8    nfg =   240    J = 3.67277e-02    ddx = 0.01
    CONVERGENCE: DDX < 0.01

Customize the optimization options by removing 'kexc' from the optimized parameters

>>> opt_u["parameters"].remove("kexc")
>>> opt_u
{
    'parameters': ['cp', 'ct', 'llr'],
    'bounds': {
        'cp': (1e-06, 1000.0),
        'ct': (1e-06, 1000.0),
        'kexc': (-50, 50),
        'llr': (1e-06, 1000.0)
    },
    'control_tfm': 'sbs',
    'termination_crit': {'maxiter': 50},
}

Run the optimization method

>>> model_u = smash.optimize(model, mapping="uniform", optimize_options=opt_u)
ValueError: Unknown, non optimized, or unbounded parameter 'kexc' in bounds optimize_options.
Choices: ['cp', 'ct', 'llr']

An error is raised because we define bounds to a non optimized parameter kexc. Remove also kexc from bounds

>>> opt_u["bounds"].pop("kexc")
(-50, 50)

Note

The built-in dictionary method pop returns the value associated to the removed key

>>> opt_u
{
    'parameters': ['cp', 'ct', 'llr'],
    'bounds': {
        'cp': (1e-06, 1000.0),
        'ct': (1e-06, 1000.0),
        'llr': (1e-06, 1000.0)
    },
    'control_tfm': 'sbs',
    'termination_crit': {'maxiter': 50},
}

Run again the optimization to see the differences linked to a change in control vector

>>> model_u = smash.optimize(model, mapping="uniform", optimize_options=opt_u)
</> Optimize
    At iterate     0    nfg =     1    J = 6.95010e-01    ddx = 0.64
    At iterate     1    nfg =    17    J = 1.28863e-01    ddx = 0.64
    At iterate     2    nfg =    32    J = 6.94838e-02    ddx = 0.32
    At iterate     3    nfg =    49    J = 4.50720e-02    ddx = 0.16
    At iterate     4    nfg =    65    J = 4.40468e-02    ddx = 0.08
    At iterate     5    nfg =    84    J = 4.35278e-02    ddx = 0.04
    At iterate     6    nfg =   102    J = 4.26906e-02    ddx = 0.02
    At iterate     7    nfg =   122    J = 4.26645e-02    ddx = 0.01
    At iterate     8    nfg =   140    J = 4.26062e-02    ddx = 0.01
    CONVERGENCE: DDX < 0.01

Get the default optimization options for a different mapping

>>> opt_ann = smash.default_optimize_options(model, mapping="ann")
>>> opt_ann
{
    'parameters': ['cp', 'ct', 'kexc', 'llr'],
    'bounds': {
        'cp': (1e-06, 1000.0),
        'ct': (1e-06, 1000.0),
        'kexc': (-50, 50),
        'llr': (1e-06, 1000.0)
    },
    'net':
        +----------------------------------------------------------+
        | Layer Type            Input/Output Shape  Num Parameters |
        +----------------------------------------------------------+
        | Dense                 (2,)/(18,)          54             |
        | Activation (ReLU)     (18,)/(18,)         0              |
        | Dense                 (18,)/(9,)          171            |
        | Activation (ReLU)     (9,)/(9,)           0              |
        | Dense                 (9,)/(4,)           40             |
        | Activation (Sigmoid)  (4,)/(4,)           0              |
        | Scale (MinMaxScale)   (4,)/(4,)           0              |
        +----------------------------------------------------------+
        Total parameters: 265
        Trainable parameters: 265,
    'learning_rate': 0.003,
    'random_state': None,
    'termination_crit': {'maxiter': 200, 'early_stopping': 0}
}

Again, customize the optimization options and optimize the Model

>>> opt_ann["learning_rate"] = 0.006
>>> opt_ann["termination_crit"]["maxiter"] = 50
>>> opt_ann["termination_crit"]["early_stopping"] = 5
>>> opt_ann["random_state"] = 21
>>> model.optimize(mapping="ann", optimize_options=opt_ann)
</> Optimize
    At iterate     0    nfg =     1    J = 1.22206e+00    |proj g| = 2.09135e-04
    At iterate     1    nfg =     2    J = 1.21931e+00    |proj g| = 2.39937e-04
    ...
    At iterate    40    nfg =    41    J = 5.21514e-02    |proj g| = 1.31863e-02
    At iterate    41    nfg =    42    J = 5.12064e-02    |proj g| = 3.74748e-03
    At iterate    42    nfg =    43    J = 5.79208e-02    |proj g| = 5.08674e-03
    At iterate    43    nfg =    44    J = 6.38050e-02    |proj g| = 1.01001e-02
    At iterate    44    nfg =    45    J = 6.57343e-02    |proj g| = 1.33649e-02
    At iterate    45    nfg =    46    J = 6.45393e-02    |proj g| = 1.56155e-02
    At iterate    46    nfg =    47    J = 6.33092e-02    |proj g| = 1.72698e-02
    EARLY STOPPING: NO IMPROVEMENT for 5 CONSECUTIVE ITERATIONS
    Revert to iteration 41 with J = 5.12064e-02 due to early stopping

The training process was terminated after 46 iterations, where the loss did not decrease below the minimal value at iteration 41 for 5 consecutive iterations. The optimal parameters are thus recorded at iteration 41.

On this page

This Page