Source code for evalml.tuners.grid_search_tuner
"""Grid Search Optimizer, which generates all of the possible points to search for using a grid."""
import itertools
from skopt.space import Integer, Real
from evalml.tuners.tuner import Tuner
from evalml.tuners.tuner_exceptions import NoParamsException
[docs]class GridSearchTuner(Tuner):
"""Grid Search Optimizer, which generates all of the possible points to search for using a grid.
Args:
pipeline_hyperparameter_ranges (dict): a set of hyperparameter ranges corresponding to a pipeline's parameters
n_points (int): The number of points to sample from along each dimension
defined in the ``space`` argument. Defaults to 10.
random_seed (int): Seed for random number generator. Unused in this class, defaults to 0.
Examples:
>>> tuner = GridSearchTuner({'My Component': {'param a': [0.0, 10.0], 'param b': ['a', 'b', 'c']}}, n_points=5)
>>> proposal = tuner.propose()
...
>>> assert proposal.keys() == {'My Component'}
>>> assert proposal['My Component'] == {'param a': 0.0, 'param b': 'a'}
Determines points using a grid search approach.
>>> for each in range(5):
... print(tuner.propose())
{'My Component': {'param a': 0.0, 'param b': 'b'}}
{'My Component': {'param a': 0.0, 'param b': 'c'}}
{'My Component': {'param a': 10.0, 'param b': 'a'}}
{'My Component': {'param a': 10.0, 'param b': 'b'}}
{'My Component': {'param a': 10.0, 'param b': 'c'}}
"""
def __init__(self, pipeline_hyperparameter_ranges, n_points=10, random_seed=0):
super().__init__(pipeline_hyperparameter_ranges, random_seed=random_seed)
raw_dimensions = list()
import pandas as pd
pd.DataFrame().iterrows()
for dimension in self._search_space_ranges:
# Categorical dimension
if isinstance(dimension, list):
range_values = dimension
elif isinstance(dimension, (Real, Integer, tuple)):
if (
isinstance(dimension, (tuple))
and isinstance(dimension[0], (int, float))
and isinstance(dimension[1], (int, float))
):
if dimension[1] > dimension[0]:
low = dimension[0]
high = dimension[1]
else:
error_text = "Upper bound must be greater than lower bound. Parameter lower bound is {0} and upper bound is {1}"
error_text = error_text.format(dimension[0], dimension[1])
raise ValueError(error_text)
else:
low = dimension.low
high = dimension.high
delta = (high - low) / (n_points - 1)
if isinstance(dimension, Integer):
range_values = [int((x * delta) + low) for x in range(n_points)]
else:
range_values = [(x * delta) + low for x in range(n_points)]
raw_dimensions.append(range_values)
self._grid_points = itertools.product(*raw_dimensions)
self.curr_params = None
[docs] def add(self, pipeline_parameters, score):
"""Not applicable to grid search tuner as generated parameters are not dependent on scores of previous parameters.
Args:
pipeline_parameters (dict): a dict of the parameters used to evaluate a pipeline
score (float): the score obtained by evaluating the pipeline with the provided parameters
"""
pass
[docs] def propose(self):
"""Returns parameters from _grid_points iterations.
If all possible combinations of parameters have been scored, then ``NoParamsException`` is raised.
Returns:
dict: proposed pipeline parameters
"""
if not self.curr_params:
self.is_search_space_exhausted()
params = self.curr_params
self.curr_params = None
return self._convert_to_pipeline_parameters(params)
[docs] def is_search_space_exhausted(self):
"""Checks if it is possible to generate a set of valid parameters. Stores generated parameters in ``self.curr_params`` to be returned by ``propose()``.
Returns:
bool: If no more valid parameters exists in the search space, return False.
Raises:
NoParamsException: If a search space is exhausted, then this exception is thrown.
"""
try:
self.curr_params = next(self._grid_points)
return False
except StopIteration:
raise NoParamsException(
"Grid search has exhausted all possible parameters.",
)
return True