smash.default_optimize_options#
- smash.default_optimize_options(model, mapping='uniform', optimizer=None)[source]#
Default optimize options of Model.
- Parameters:
- model
Model Primary data structure of the hydrological model
smash.- mappingstr, default ‘uniform’
Type of mapping. Should be one of
'uniform''distributed''multi-linear''multi-polynomial''ann'
Hint
See the Mapping section
- optimizerstr or None, default None
Name of optimizer. Should be one of
'sbs'('uniform'mapping only)'lbfgsb'('uniform','distributed','multi-linear'or'multi-polynomial'mapping only)'sgd'('ann'mapping only)'adam'('ann'mapping only)'adagrad'('ann'mapping only)'rmsprop'('ann'mapping only)
Note
If not given, a default optimizer will be set depending on the optimization mapping:
mapping =
'uniform'; optimizer ='sbs'mapping =
'distributed','multi-linear', or'multi-polynomial'; optimizer ='lbfgsb'mapping =
'ann'; optimizer ='adam'
Hint
See the Optimization Algorithm section
- model
- 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 pass to the optimize_options argument of the optimize method
optimize(orModel.optimize).
See also
smash.optimizeModel assimilation using numerical optimization algorithms.
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 method
optimize(orModel.optimize). 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 = 0.695010 ddx = 0.64 At iterate 1 nfg = 30 J = 0.098411 ddx = 0.64 At iterate 2 nfg = 59 J = 0.045409 ddx = 0.32 At iterate 3 nfg = 88 J = 0.038182 ddx = 0.16 At iterate 4 nfg = 117 J = 0.037362 ddx = 0.08 At iterate 5 nfg = 150 J = 0.037087 ddx = 0.02 At iterate 6 nfg = 183 J = 0.036800 ddx = 0.02 At iterate 7 nfg = 216 J = 0.036763 ddx = 0.01 CONVERGENCE: DDX < 0.01
Customize the optimize options dictionary 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 optimize method
>>> model_u = smash.optimize(model, mapping="uniform", optimize_options=opt_u) ValueError: Unknown or non optimized parameter 'kexc' in bounds optimize_options. Choices: ['cp', 'ct', 'llr']
An error is raised because we define
boundsto a non optimized parameterkexc. Remove alsokexcfrom 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 optimize method 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 = 0.695010 ddx = 0.64 At iterate 1 nfg = 17 J = 0.128863 ddx = 0.64 At iterate 2 nfg = 32 J = 0.069483 ddx = 0.32 At iterate 3 nfg = 49 J = 0.045072 ddx = 0.16 At iterate 4 nfg = 65 J = 0.044047 ddx = 0.08 At iterate 5 nfg = 84 J = 0.043528 ddx = 0.04 At iterate 6 nfg = 102 J = 0.042690 ddx = 0.02 At iterate 7 nfg = 122 J = 0.042665 ddx = 0.01 At iterate 8 nfg = 140 J = 0.042606 ddx = 0.00 CONVERGENCE: DDX < 0.0
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': 0.001, 'random_state': None, 'termination_crit': {'epochs': 200, 'early_stopping': 0} }
Again, customize the optimize options and optimize the Model
>>> opt_ann["learning_rate"] = 0.004 >>> opt_ann["termination_crit"]["epochs"] = 40 >>> opt_ann["termination_crit"]["early_stopping"] = 10 >>> opt_ann["random_state"] = 11 >>> model.optimize(mapping="ann", optimize_options=opt_ann) </> Optimize At epoch 1 J = 1.208756 |proj g| = 0.001343 At epoch 2 J = 1.204409 |proj g| = 0.001567 ... At epoch 25 J = 0.099455 |proj g| = 0.009175 At epoch 26 J = 0.078379 |proj g| = 0.013312 At epoch 27 J = 0.080410 |proj g| = 0.022210 ... At epoch 35 J = 0.176823 |proj g| = 0.015846 At epoch 36 J = 0.172950 |proj g| = 0.015864 Training: 90%|████████████████▌ | 36/40 [00:21<00:02, 1.65it/s]
The training process was terminated after 36 epochs, where the loss did not decrease below the minimal value at epoch 26 for 10 consecutive epochs. The optimal parameters are thus recorded at epoch 26