"""
This module provides a Multi-layer Perceptron (MLP) model wrapper, inheriting
from `aiqclib.common.base.scikit_learn_model_base.SklearnModelBase`.
It facilitates training, prediction, and evaluation of an MLP classifier using
Polars DataFrames, converting them to Pandas for compatibility with the
`sklearn` library.
"""
from typing import Dict, Any
from sklearn.neural_network import MLPClassifier as SklearnMLP
from aiqclib.common.base.config_base import ConfigBase
from aiqclib.common.base.scikit_learn_model_base import SklearnModelBase
[docs]
class MultilayerPerceptron(SklearnModelBase):
"""
A Multi-layer Perceptron (MLP) model wrapper class for training and testing.
Inherits from :class:`~aiqclib.common.base.scikit_learn_model_base.SklearnModelBase`
to reuse common Scikit-Learn API logic.
Features include:
- Automatic application of ``model_params`` from the YAML config, if defined;
otherwise, uses default hyperparameters.
- Uses :class:`sklearn.neural_network.MLPClassifier`.
.. note::
This class sets :attr:`expected_class_name` to ``"MultilayerPerceptron"``.
This is a feedforward neural network implementation.
"""
expected_class_name: str = "MultilayerPerceptron"
short_name: str = "MLP"
def __init__(self, config: ConfigBase) -> None:
"""
Initialize the MLP model with default or user-specified parameters.
:param config: A configuration object providing model parameters.
:type config: aiqclib.common.base.config_base.ConfigBase
"""
super().__init__(config=config)
self.model_params: Dict[str, Any] = {
"hidden_layer_sizes": (50,),
"activation": "relu",
"solver": "adam",
"alpha": 0.0001,
"batch_size": "auto",
"learning_rate": "constant",
"learning_rate_init": 0.001,
"max_iter": 100,
"shuffle": True,
"random_state": None,
"tol": 1e-3,
"early_stopping": True,
"n_iter_no_change": 5,
}
# Update model parameters with config step parameters
model_params = self.config.get_model_params(
self.expected_class_name, self.short_name
)
self.model_params.update(model_params)
self.allow_na = False
def _get_model_class(self) -> Any:
"""
Return the Scikit-Learn MLPClassifier class.
:return: The MLPClassifier class.
:rtype: Any
"""
return SklearnMLP