from typing import Any, Dict, List
from aeon.classification.ordinal_classification import OrdinalTDE
from skopt.space import Categorical, Integer, Real
from ml_grid.pipeline.data import pipe
# unknown ts
[docs]
class OrdinalTDE_class:
"""A wrapper for the aeon OrdinalTDE time-series classifier.
This class provides a consistent interface for the OrdinalTDE classifier,
including defining a hyperparameter search space.
Attributes:
algorithm_implementation: An instance of the aeon OrdinalTDE classifier.
method_name (str): The name of the classifier method.
parameter_space (Dict[str, List[Any]]): The hyperparameter search space
for the classifier.
"""
[docs]
algorithm_implementation: OrdinalTDE
[docs]
parameter_space: Dict[str, List[Any]]
def __init__(self, ml_grid_object: pipe):
"""Initializes the OrdinalTDE_class.
Args:
ml_grid_object (pipe): An instance of the main data pipeline object.
"""
random_state_val = ml_grid_object.global_params.random_state_val
n_jobs_model_val = ml_grid_object.global_params.n_jobs_model_val
time_limit_param = ml_grid_object.global_params.time_limit_param
if isinstance(time_limit_param, list):
time_limit_param = time_limit_param[0]
self.algorithm_implementation = OrdinalTDE()
self.method_name = "OrdinalTDE"
if getattr(ml_grid_object.global_params, "test_mode", False):
self.parameter_space = {
"n_parameter_samples": [10],
"max_ensemble_size": [5],
"time_limit_in_minutes": [0.05],
"n_jobs": [1],
}
return
if ml_grid_object.global_params.bayessearch:
self.parameter_space = {
"n_parameter_samples": Categorical([100, 250, 500]),
"max_ensemble_size": Categorical([30, 50, 100]),
"max_win_len_prop": Real(0.8, 1.0),
"min_window": Integer(5, 15),
"randomly_selected_params": Categorical([30, 50, 70]),
"bigrams": Categorical([True, False, None]),
"dim_threshold": Real(0.75, 0.95),
"max_dims": Integer(10, 30),
"time_limit_in_minutes": time_limit_param,
"contract_max_n_parameter_samples": Categorical([1000, 2000]),
"typed_dict": Categorical([True, False]),
"n_jobs": [n_jobs_model_val],
"random_state": [random_state_val],
}
else:
self.parameter_space = {
"n_parameter_samples": [
100,
250,
500,
], # Number of parameter combinations to consider for the final ensemble
"max_ensemble_size": [
30,
50,
100,
], # Maximum number of estimators in the ensemble
"max_win_len_prop": [
0.8,
1.0,
], # Maximum window length as a proportion of series length
"min_window": [5, 10, 15], # Minimum window length
"randomly_selected_params": [
30,
50,
70,
], # Number of parameters randomly selected before Gaussian process parameter selection
"bigrams": [True, False, None], # Whether to use bigrams
"dim_threshold": [
0.75,
0.85,
0.95,
], # Dimension accuracy threshold for multivariate data
"max_dims": [
10,
20,
30,
], # Max number of dimensions per classifier for multivariate data
"time_limit_in_minutes": time_limit_param, # Time contract to limit build time in minutes
"contract_max_n_parameter_samples": [
1000,
2000,
], # Max number of parameter combinations when time_limit_in_minutes is set
"typed_dict": [
True,
False,
], # Whether to use numba typed Dict to store word counts
#'save_train_predictions': [True, False], # Save ensemble member train predictions in fit for LOOCV
"n_jobs": [
n_jobs_model_val
], # Number of jobs to run in parallel for fit and predict
"random_state": [random_state_val], # Seed for random number generation
}