Regressor¶
This module presents the regressor_obj class, which is related to machine learning regression. This class does not implement any supervised learning regression algorithm. Instead, it serves as an interface between a regressor object implemented elsewhere and the ESCAPE model and data objects. In other words, this class allows the use of any 3rd party or self-implemented regressor class.
One well-known 3rd party regressor is the Multilayer Perceptron Regressor from the scikit-learn package [1] or another good example is the torch based regressor from skorch package [2].
After training, machine learning regressors can predict a solution for a model based on a given experimental dataset. This solution can then be further optimized using any optimizer from ESCAPE.
Regressors can be quite useful for predicting solutions for a large number of experimental data, also known as batch processing. They provide initial solutions that are quite close to the actual ones for further optimization.
- class escape.core.regressor.regressor_obj¶
Wrapper class for regressor that provides an interface between ESCAPE and machine learning regressors.
This class allows integration of any 3rd party or custom regressor implementation with ESCAPE’s model and data objects. It handles training data generation, parameter management, and prediction.
- c_test¶
Test coordinates.
- c_train¶
Training coordinates.
- data_length¶
Length of training data.
- dumps()¶
- fit(X: ndarray[Any, dtype[_ScalarType_co]], y: ndarray[Any, dtype[_ScalarType_co]]) None¶
Fit the regressor to training data.
Must be implemented by subclasses.
- is_stopped¶
Check if training process is stopped.
- load_train_data(filename: str) None¶
Load training data from file.
- filenamestr
Path to load the training data from
- loads(data)¶
- modelstack¶
Modelstack used to generate training data.
- name¶
Name of the regressor object.
- noisify_coords(c: ndarray[Any, dtype[_ScalarType_co]], reset: bool = True) ndarray[Any, dtype[_ScalarType_co]]¶
Add noise to coordinates.
Must be implemented by subclasses.
- noisify_data(X: ndarray[Any, dtype[_ScalarType_co]], reset: bool = True) ndarray[Any, dtype[_ScalarType_co]]¶
Add noise to data.
Must be implemented by subclasses.
- normalize_data(X: ndarray[Any, dtype[_ScalarType_co]], c: ndarray[Any, dtype[_ScalarType_co]], reset: bool = True) ndarray[Any, dtype[_ScalarType_co]]¶
Normalize input data.
Must be implemented by subclasses.
- nsamples¶
Number of training samples.
- ntests¶
Number of test samples.
- num_train_params¶
Number of trainable parameters.
- parameters¶
List of model parameters.
- predict(X: ndarray[Any, dtype[_ScalarType_co]]) ndarray[Any, dtype[_ScalarType_co]]¶
Make predictions for input data.
Must be implemented by subclasses.
- save_train_data(filename: str) None¶
Save generated training data to file.
- filenamestr
Path to save the training data
- score(X: ndarray[Any, dtype[_ScalarType_co]], y: ndarray[Any, dtype[_ScalarType_co]]) float¶
Calculate score for predictions.
Must be implemented by subclasses.
- show(model_configs: Optional[Union[Any, List[Any], tuple[Any, ...]]] = None, config: Optional[Any] = None, **kwargs) Any¶
Show regressor object in a widget.
- Args:
model_configs: Model configurations to display config: Layout configuration **kwargs: Additional keyword arguments
- Returns:
RegressorObjectLayout instance
- test() datastack_obj¶
Create random test data using the current model configuration.
- datastack_obj
Test datastack containing generated test data
- train(reset_data: bool = True, fit: bool = True) Optional[float]¶
Train the regressor.
- reset_databool, optional
If True, reset previously generated training set, by default True
- fitbool, optional
If True, fit the model to the training data, by default True
- Optional[float]
Score of trained regressor on validation dataset if available, None otherwise
- x_test¶
Test input data.
- x_train¶
Training input data.
- y_test¶
Test target data.
- y_train¶
Training target data.