"""EvalML's core AutoML object."""
import copy
import logging
import pickle
import sys
import time
import traceback
import warnings
from collections import defaultdict
from concurrent.futures import ProcessPoolExecutor, ThreadPoolExecutor
import cloudpickle
import numpy as np
import pandas as pd
from dask import distributed as dd
from plotly import io as pio
from sklearn.model_selection import BaseCrossValidator
from skopt.space import Categorical
from evalml.automl.automl_algorithm import DefaultAlgorithm, IterativeAlgorithm
from evalml.automl.callbacks import log_error_callback
from evalml.automl.engine import SequentialEngine
from evalml.automl.engine.cf_engine import CFClient, CFEngine
from evalml.automl.engine.dask_engine import DaskEngine
from evalml.automl.pipeline_search_plots import PipelineSearchPlots, SearchIterationPlot
from evalml.automl.progress import Progress
from evalml.automl.utils import (
AutoMLConfig,
check_all_pipeline_names_unique,
get_best_sampler_for_data,
get_default_primary_search_objective,
make_data_splitter,
)
from evalml.data_checks import (
ClassImbalanceDataCheck,
DataCheckMessageType,
DefaultDataChecks,
)
from evalml.exceptions import (
AutoMLSearchException,
PipelineNotFoundError,
PipelineScoreError,
)
from evalml.model_family import ModelFamily
from evalml.objectives import (
get_default_recommendation_objectives,
get_non_core_objectives,
get_objective,
get_optimization_objectives,
normalize_objectives,
organize_objectives,
recommendation_score,
)
from evalml.pipelines import (
BinaryClassificationPipeline,
ComponentGraph,
MulticlassClassificationPipeline,
RegressionPipeline,
)
from evalml.pipelines.components import ARIMARegressor
from evalml.pipelines.utils import make_timeseries_baseline_pipeline
from evalml.preprocessing import split_data
from evalml.problem_types import (
ProblemTypes,
handle_problem_types,
is_binary,
is_classification,
is_multiseries,
is_time_series,
)
from evalml.tuners import SKOptTuner
from evalml.utils import convert_to_seconds, infer_feature_types
from evalml.utils.gen_utils import contains_all_ts_parameters
from evalml.utils.logger import (
get_logger,
log_batch_times,
log_subtitle,
log_title,
)
[docs]def build_engine_from_str(engine_str):
"""Function that converts a convenience string for an parallel engine type and returns an instance of that engine.
Args:
engine_str (str): String representing the requested engine.
Returns:
(EngineBase): Instance of the requested engine.
Raises:
ValueError: If engine_str is not a valid engine.
"""
valid_engines = [
"sequential",
"cf_threaded",
"cf_process",
"dask_threaded",
"dask_process",
]
if engine_str not in valid_engines:
raise ValueError(
f"'{engine_str}' is not a valid engine, please choose from {valid_engines}",
)
elif engine_str == "sequential":
return SequentialEngine()
elif engine_str == "cf_threaded":
return CFEngine(CFClient(ThreadPoolExecutor()))
elif engine_str == "cf_process":
return CFEngine(CFClient(ProcessPoolExecutor()))
elif engine_str == "dask_threaded":
return DaskEngine(cluster=dd.LocalCluster(processes=False))
elif engine_str == "dask_process":
return DaskEngine(cluster=dd.LocalCluster(processes=True))
[docs]def search(
X_train=None,
y_train=None,
problem_type=None,
objective="auto",
mode="fast",
max_time=None,
patience=None,
tolerance=None,
problem_configuration=None,
n_splits=3,
verbose=False,
timing=False,
):
"""Given data and configuration, run an automl search.
This method will run EvalML's default suite of data checks. If the data checks produce errors, the data check results will be returned before running the automl search. In that case we recommend you alter your data to address these errors and try again.
This method is provided for convenience. If you'd like more control over when each of these steps is run, consider making calls directly to the various pieces like the data checks and AutoMLSearch, instead of using this method.
Args:
X_train (pd.DataFrame): The input training data of shape [n_samples, n_features]. Required.
y_train (pd.Series): The target training data of length [n_samples]. Required for supervised learning tasks.
problem_type (str or ProblemTypes): Type of supervised learning problem. See evalml.problem_types.ProblemType.all_problem_types for a full list.
objective (str, ObjectiveBase): The objective to optimize for. Used to propose and rank pipelines, but not for optimizing each pipeline during fit-time.
When set to 'auto', chooses:
- LogLossBinary for binary classification problems,
- LogLossMulticlass for multiclass classification problems, and
- R2 for regression problems.
mode (str): mode for DefaultAlgorithm. There are two modes: fast and long, where fast is a subset of long. Please look at DefaultAlgorithm for more details.
max_time (int, str): Maximum time to search for pipelines.
This will not start a new pipeline search after the duration
has elapsed. If it is an integer, then the time will be in seconds.
For strings, time can be specified as seconds, minutes, or hours.
patience (int): Number of iterations without improvement to stop search early. Must be positive.
If None, early stopping is disabled. Defaults to None.
tolerance (float): Minimum percentage difference to qualify as score improvement for early stopping.
Only applicable if patience is not None. Defaults to None.
problem_configuration (dict): Additional parameters needed to configure the search. For example,
in time series problems, values should be passed in for the time_index, gap, forecast_horizon, and max_delay variables.
n_splits (int): Number of splits to use with the default data splitter.
verbose (boolean): Whether or not to display semi-real-time updates to stdout while search is running. Defaults to False.
timing (boolean): Whether or not to write pipeline search times to the logger. Defaults to False.
Returns:
(AutoMLSearch, dict): The automl search object containing pipelines and rankings, and the results from running the data checks. If the data check results contain errors, automl search will not be run and an automl search object will not be returned.
Raises:
ValueError: If search configuration is not valid.
"""
X_train = infer_feature_types(X_train)
y_train = infer_feature_types(y_train)
problem_type = handle_problem_types(problem_type)
if is_time_series(problem_type):
is_valid, msg = contains_all_ts_parameters(problem_configuration)
if not is_valid:
raise ValueError(msg)
if objective == "auto":
objective = get_default_primary_search_objective(problem_type)
objective = get_objective(objective, return_instance=False)
if mode != "fast" and mode != "long":
raise ValueError("Mode must be either 'fast' or 'long'")
max_batches = None
if mode == "fast":
max_batches = 3 # corresponds to end of 'fast' mode
elif mode == "long" and max_time:
max_batches = 999 # defers to stopping criterion
elif mode == "long" and max_time is None:
max_batches = 6 # corresponds to end of 'long' exploration phase
data_splitter = make_data_splitter(
X=X_train,
y=y_train,
problem_type=problem_type,
problem_configuration=problem_configuration,
n_splits=n_splits,
)
automl_config = {
"X_train": X_train,
"y_train": y_train,
"problem_type": problem_type,
"objective": objective,
"max_batches": max_batches,
"max_time": max_time,
"patience": patience,
"tolerance": tolerance,
"verbose": verbose,
"problem_configuration": problem_configuration,
"data_splitter": data_splitter,
"timing": timing,
}
data_checks = DefaultDataChecks(
problem_type=problem_type,
objective=objective,
n_splits=n_splits,
problem_configuration=problem_configuration,
)
data_check_results = data_checks.validate(X_train, y=y_train)
for data_check_result in data_check_results:
if data_check_result["level"] == DataCheckMessageType.ERROR.value:
return None, data_check_results
automl = AutoMLSearch(automl_algorithm="default", ensembling=True, **automl_config)
automl.search()
return automl, data_check_results
[docs]def search_iterative(
X_train=None,
y_train=None,
problem_type=None,
objective="auto",
problem_configuration=None,
n_splits=3,
timing=False,
**kwargs,
):
"""Given data and configuration, run an automl search.
This method will run EvalML's default suite of data checks. If the data checks produce errors, the data check results will be returned before running the automl search. In that case we recommend you alter your data to address these errors and try again.
This method is provided for convenience. If you'd like more control over when each of these steps is run, consider making calls directly to the various pieces like the data checks and AutoMLSearch, instead of using this method.
Args:
X_train (pd.DataFrame): The input training data of shape [n_samples, n_features]. Required.
y_train (pd.Series): The target training data of length [n_samples]. Required for supervised learning tasks.
problem_type (str or ProblemTypes): Type of supervised learning problem. See evalml.problem_types.ProblemType.all_problem_types for a full list.
objective (str, ObjectiveBase): The objective to optimize for. Used to propose and rank pipelines, but not for optimizing each pipeline during fit-time.
When set to 'auto', chooses:
- LogLossBinary for binary classification problems,
- LogLossMulticlass for multiclass classification problems, and
- R2 for regression problems.
problem_configuration (dict): Additional parameters needed to configure the search. For example,
in time series problems, values should be passed in for the time_index, gap, forecast_horizon, and max_delay variables.
n_splits (int): Number of splits to use with the default data splitter.
timing(boolean): Whether or not to write pipeline search times to the logger. Defaults to False.
**kwargs: Other keyword arguments which are provided will be passed to AutoMLSearch.
Returns:
(AutoMLSearch, dict): the automl search object containing pipelines and rankings, and the results from running the data checks. If the data check results contain errors, automl search will not be run and an automl search object will not be returned.
Raises:
ValueError: If the search configuration is invalid.
"""
X_train = infer_feature_types(X_train)
y_train = infer_feature_types(y_train)
problem_type = handle_problem_types(problem_type)
if is_time_series(problem_type):
is_valid, msg = contains_all_ts_parameters(problem_configuration)
if not is_valid:
raise ValueError(msg)
if objective == "auto":
objective = get_default_primary_search_objective(problem_type)
objective = get_objective(objective, return_instance=False)
data_splitter = make_data_splitter(
X=X_train,
y=y_train,
problem_type=problem_type,
problem_configuration=problem_configuration,
n_splits=n_splits,
)
automl_config = kwargs
automl_config.update(
{
"X_train": X_train,
"y_train": y_train,
"problem_type": problem_type,
"objective": objective,
"max_batches": 1,
"problem_configuration": problem_configuration,
"data_splitter": data_splitter,
"timing": timing,
},
)
data_checks = DefaultDataChecks(
problem_type=problem_type,
objective=objective,
n_splits=n_splits,
problem_configuration=problem_configuration,
)
data_check_results = data_checks.validate(X_train, y=y_train)
for data_check_result in data_check_results:
if data_check_result["level"] == DataCheckMessageType.ERROR.value:
return None, data_check_results
automl = AutoMLSearch(**automl_config)
automl.search()
return automl, data_check_results
[docs]class AutoMLSearch:
"""Automated Pipeline search.
Args:
X_train (pd.DataFrame): The input training data of shape [n_samples, n_features]. Required.
y_train (pd.Series): The target training data of length [n_samples]. Required for supervised learning tasks.
X_holdout (pd.DataFrame): The input holdout data of shape [n_samples, n_features].
y_holdout (pd.Series): The target holdout data of length [n_samples].
problem_type (str or ProblemTypes): Type of supervised learning problem. See evalml.problem_types.ProblemType.all_problem_types for a full list.
objective (str, ObjectiveBase): The objective to optimize for. Used to propose and rank pipelines, but not for optimizing each pipeline during fit-time.
When set to 'auto', chooses:
- LogLossBinary for binary classification problems,
- LogLossMulticlass for multiclass classification problems, and
- R2 for regression problems.
max_iterations (int): Maximum number of iterations to search. If max_iterations and
max_time is not set, then max_iterations will default to max_iterations of 5.
max_time (int, str): Maximum time to search for pipelines.
This will not start a new pipeline search after the duration
has elapsed. If it is an integer, then the time will be in seconds.
For strings, time can be specified as seconds, minutes, or hours.
patience (int): Number of iterations without improvement to stop search early. Must be positive.
If None, early stopping is disabled. Defaults to None.
tolerance (float): Minimum percentage difference to qualify as score improvement for early stopping.
Only applicable if patience is not None. Defaults to None.
allowed_component_graphs (dict): A dictionary of lists or ComponentGraphs indicating the component graphs allowed in the search.
The format should follow { "Name_0": [list_of_components], "Name_1": ComponentGraph(...) }
The default of None indicates all pipeline component graphs for this problem type are allowed. Setting this field will cause
allowed_model_families to be ignored.
e.g. allowed_component_graphs = { "My_Graph": ["Imputer", "One Hot Encoder", "Random Forest Classifier"] }
allowed_model_families (list(str, ModelFamily)): The model families to search. The default of None searches over all
model families. Run evalml.pipelines.components.utils.allowed_model_families("binary") to see options. Change `binary`
to `multiclass` or `regression` depending on the problem type. Note that if allowed_pipelines is provided,
this parameter will be ignored. For default algorithm, this only applies to estimators in the non-naive batches.
features (list)[FeatureBase]: List of features to run DFS on AutoML pipelines. Defaults to None.
Features will only be computed if the columns used by the feature exist in the search input
and if the feature itself is not in search input. If features is an empty list, the DFS Transformer will not be included in pipelines.
run_feature_selection (bool): If True, will run a separate feature selection pipeline and only use selected features in subsequent batches.
If False, will use all of the features for every pipeline. Only used for default algorithm, setting is no-op for iterative algorithm.
data_splitter (sklearn.model_selection.BaseCrossValidator): Data splitting method to use. Defaults to StratifiedKFold.
tuner_class: The tuner class to use. Defaults to SKOptTuner.
optimize_thresholds (bool): Whether or not to optimize the binary pipeline threshold. Defaults to True.
start_iteration_callback (callable): Function called before each pipeline training iteration.
Callback function takes three positional parameters: The pipeline instance and the AutoMLSearch object.
add_result_callback (callable): Function called after each pipeline training iteration.
Callback function takes three positional parameters: A dictionary containing the training results for the new pipeline, an untrained_pipeline containing the parameters used during training, and the AutoMLSearch object.
error_callback (callable): Function called when `search()` errors and raises an Exception.
Callback function takes three positional parameters: the Exception raised, the traceback, and the AutoMLSearch object.
Must also accepts kwargs, so AutoMLSearch is able to pass along other appropriate parameters by default.
Defaults to None, which will call `log_error_callback`.
additional_objectives (list): Custom set of objectives to score on.
Will override default objectives for problem type if not empty.
alternate_thresholding_objective (str): The objective to use for thresholding binary classification pipelines if the main objective provided isn't tuneable.
Defaults to F1.
random_seed (int): Seed for the random number generator. Defaults to 0.
n_jobs (int or None): Non-negative integer describing level of parallelism used for pipelines.
None and 1 are equivalent. If set to -1, all CPUs are used. For n_jobs below -1, (n_cpus + 1 + n_jobs) are used.
ensembling (boolean): If True, runs ensembling in a separate batch after every allowed pipeline class has been iterated over.
If the number of unique pipelines to search over per batch is one, ensembling will not run. Defaults to False.
max_batches (int): The maximum number of batches of pipelines to search. Parameters max_time, and
max_iterations have precedence over stopping the search.
problem_configuration (dict, None): Additional parameters needed to configure the search. For example,
in time series problems, values should be passed in for the time_index, gap, forecast_horizon, and max_delay variables.
For multiseries time series problems, the values passed in should also include the name of a series_id column.
train_best_pipeline (boolean): Whether or not to train the best pipeline before returning it. Defaults to True.
search_parameters (dict): A dict of the hyperparameter ranges or pipeline parameters used to iterate over during search.
Keys should consist of the component names and values should specify a singular value/list for pipeline parameters, or skopt.Space for hyperparameter ranges.
In the example below, the Imputer parameters would be passed to the hyperparameter ranges, and the Label Encoder parameters would be used as the component parameter.
e.g. search_parameters = { 'Imputer' : { 'numeric_impute_strategy': Categorical(['most_frequent', 'median']) },
'Label Encoder': {'positive_label': True} }
sampler_method (str): The data sampling component to use in the pipelines if the problem type is classification and the target balance is smaller than the sampler_balanced_ratio.
Either 'auto', which will use our preferred sampler for the data, 'Undersampler', 'Oversampler', or None. Defaults to 'auto'.
sampler_balanced_ratio (float): The minority:majority class ratio that we consider balanced, so a 1:4 ratio would be equal to 0.25. If the class balance is larger than this provided value,
then we will not add a sampler since the data is then considered balanced. Overrides the `sampler_ratio` of the samplers. Defaults to 0.25.
allow_long_running_models (bool): Whether or not to allow longer-running models for large multiclass problems. If False and no pipelines, component graphs, or model families are provided,
AutoMLSearch will not use Elastic Net or XGBoost when there are more than 75 multiclass targets and will not use CatBoost when there are more than 150 multiclass targets. Defaults to False.
_ensembling_split_size (float): The amount of the training data we'll set aside for training ensemble metalearners. Only used when ensembling is True.
Must be between 0 and 1, exclusive. Defaults to 0.2
_pipelines_per_batch (int): The number of pipelines to train for every batch after the first one.
The first batch will train a baseline pipline + one of each pipeline family allowed in the search.
automl_algorithm (str): The automl algorithm to use. Currently the two choices are 'iterative' and 'default'. Defaults to `default`.
engine (EngineBase or str): The engine instance used to evaluate pipelines. Dask or concurrent.futures engines can also
be chosen by providing a string from the list ["sequential", "cf_threaded", "cf_process", "dask_threaded", "dask_process"].
If a parallel engine is selected this way, the maximum amount of parallelism, as determined by the engine, will be used. Defaults to "sequential".
verbose (boolean): Whether or not to display semi-real-time updates to stdout while search is running. Defaults to False.
timing (boolean): Whether or not to write pipeline search times to the logger. Defaults to False.
exclude_featurizers (list[str]): A list of featurizer components to exclude from the pipelines built by search.
Valid options are "DatetimeFeaturizer", "EmailFeaturizer", "URLFeaturizer", "NaturalLanguageFeaturizer", "TimeSeriesFeaturizer"
excluded_model_families (list(str, ModelFamily)): A list of model families to exclude from the estimators used when building pipelines. For default algorithm, this only excludes estimators in the non-naive batches.
holdout_set_size (float): The size of the holdout set that AutoML search will take for datasets larger than 500 rows. If set to 0, holdout set will not be taken regardless of number of rows. Must be between 0 and 1, exclusive. Defaults to 0.1.
use_recommendation (bool): Whether or not to use a recommendation score to rank pipelines instead of optimization objective. Defaults to False.
include_recommendation (list[str]): A list of objectives to include beyond the defaults in the recommendation score. Defaults to None.
exclude_recommendation (list[str]): A list of objectives to exclude from the defaults in the recommendation score. Defaults to None.
"""
_MAX_NAME_LEN = 40
# Minimum number of rows dataset must have before a holdout set is used to rank pipelines.
_HOLDOUT_SET_MIN_ROWS = 500
def __init__(
self,
X_train=None,
y_train=None,
X_holdout=None,
y_holdout=None,
problem_type=None,
objective="auto",
max_iterations=None,
max_time=None,
patience=None,
tolerance=None,
data_splitter=None,
allowed_component_graphs=None,
allowed_model_families=None,
excluded_model_families=None,
features=None,
run_feature_selection=True,
start_iteration_callback=None,
add_result_callback=None,
error_callback=None,
additional_objectives=None,
alternate_thresholding_objective="F1",
random_seed=0,
n_jobs=-1,
tuner_class=None,
optimize_thresholds=True,
ensembling=False,
max_batches=None,
problem_configuration=None,
train_best_pipeline=True,
search_parameters=None,
sampler_method="auto",
sampler_balanced_ratio=0.25,
allow_long_running_models=False,
_pipelines_per_batch=5,
automl_algorithm="default",
engine="sequential",
verbose=False,
timing=False,
exclude_featurizers=None,
holdout_set_size=0,
use_recommendation=False,
include_recommendation=None,
exclude_recommendation=None,
):
self.verbose = verbose
if verbose:
self.logger = get_logger(f"{__name__}.verbose")
else:
self.logger = logging.getLogger(__name__)
self.timing = timing
if X_train is None:
raise ValueError(
"Must specify training data as a 2d array using the X_train argument",
)
if y_train is None:
raise ValueError(
"Must specify training data target values as a 1d vector using the y_train argument",
)
if X_holdout is not None and y_holdout is not None:
self.passed_holdout_set = True
elif X_holdout is None and y_holdout is None:
self.passed_holdout_set = False
elif X_holdout is None and y_holdout is not None:
raise ValueError(
"Must specify holdout data as a 2d array using the X_holdout argument",
)
elif X_holdout is not None and y_holdout is None:
raise ValueError(
"Must specify training data target values as a 1d vector using the y_holdout argument",
)
try:
self.problem_type = handle_problem_types(problem_type)
except ValueError:
raise ValueError(
"choose one of (binary, multiclass, regression) as problem_type",
)
if is_time_series(self.problem_type):
warnings.warn(
"Time series support in evalml is still in beta, which means we are still actively building "
"its core features. Please be mindful of that when running search().",
)
self.errors = {}
self._SLEEP_TIME = 0.1
self.tuner_class = tuner_class or SKOptTuner
self.start_iteration_callback = start_iteration_callback
self.add_result_callback = add_result_callback
self.error_callback = error_callback or log_error_callback
self.data_splitter = data_splitter
self.optimize_thresholds = optimize_thresholds
self.ensembling = ensembling
if not isinstance(max_time, (int, float, str, type(None))):
raise TypeError(
f"Parameter max_time must be a float, int, string or None. Received {type(max_time)} with value {str(max_time)}..",
)
if isinstance(max_time, (int, float)) and max_time < 0:
raise ValueError(
f"Parameter max_time must be None or non-negative. Received {max_time}.",
)
if max_batches is not None and max_batches < 0:
raise ValueError(
f"Parameter max_batches must be None or non-negative. Received {max_batches}.",
)
if max_iterations is not None and max_iterations < 0:
raise ValueError(
f"Parameter max_iterations must be None or non-negative. Received {max_iterations}.",
)
self.max_time = (
convert_to_seconds(max_time) if isinstance(max_time, str) else max_time
)
self.max_iterations = max_iterations
self.max_batches = max_batches
self._pipelines_per_batch = _pipelines_per_batch
self.holdout_set_size = holdout_set_size
if patience and (not isinstance(patience, int) or patience < 0):
raise ValueError(
"patience value must be a positive integer. Received {} instead".format(
patience,
),
)
if tolerance and (tolerance > 1.0 or tolerance < 0.0):
raise ValueError(
"tolerance value must be a float between 0.0 and 1.0 inclusive. Received {} instead".format(
tolerance,
),
)
self.patience = patience
self.tolerance = tolerance or 0.0
self._results = {
"pipeline_results": {},
"search_order": [],
}
self._pipelines_searched = dict()
self.random_seed = random_seed
self.n_jobs = n_jobs
if allowed_component_graphs is not None:
if not isinstance(allowed_component_graphs, dict):
raise ValueError(
"Parameter allowed_component_graphs must be either None or a dictionary!",
)
for graph_name, graph in allowed_component_graphs.items():
if not isinstance(graph, (list, dict, ComponentGraph)):
raise ValueError(
"Every component graph passed must be of type list, dictionary, or ComponentGraph!",
)
self.allowed_component_graphs = allowed_component_graphs
self.allowed_model_families = allowed_model_families
self.allow_long_running_models = allow_long_running_models
self._start = 0.0
self._baseline_cv_scores = {}
self.show_batch_output = False
self.problem_configuration = self._validate_problem_configuration(
problem_configuration,
)
self._train_best_pipeline = train_best_pipeline
self._best_pipeline = None
self._searched = False
if self.holdout_set_size < 0 or holdout_set_size >= 1:
raise ValueError(
"Holdout set size must be greater than 0 and less than 1. Set holdout set size to 0 to disable holdout set evaluation.",
)
if self.passed_holdout_set is False and self.holdout_set_size > 0:
if len(X_train) >= self._HOLDOUT_SET_MIN_ROWS:
# Create holdout set from X_train and y_train data because X_train above or at row threshold
X_train, X_holdout, y_train, y_holdout = split_data(
X_train,
y_train,
problem_type=self.problem_type,
problem_configuration=self.problem_configuration,
test_size=self.holdout_set_size,
random_seed=self.random_seed,
)
self.logger.info(
f"Created a holdout dataset with {len(X_holdout)} rows. Training dataset has {len(X_train)} rows.",
)
else:
self.logger.info(
f"Dataset size is too small to create holdout set. Minimum dataset size is {self._HOLDOUT_SET_MIN_ROWS} rows, X_train has {len(X_train)} rows. Holdout set evaluation is disabled.",
)
# For multiseries problems, we need to make sure that the data is primarily ordered by the time_index rather than the series_id
if is_multiseries(self.problem_type):
time_index = self.problem_configuration.get("time_index")
series_id = self.problem_configuration.get("series_id")
X_train = X_train.sort_values([time_index, series_id])
y_train = y_train[X_train.index].reset_index(drop=True)
X_train = X_train.reset_index(drop=True)
# Set holdout data in AutoML search if provided as parameter
self.X_train = infer_feature_types(X_train)
self.y_train = infer_feature_types(y_train)
self.X_holdout = (
infer_feature_types(X_holdout) if X_holdout is not None else None
)
self.y_holdout = (
infer_feature_types(y_holdout) if y_holdout is not None else None
)
if self.X_holdout is None and self.y_holdout is None:
# Holdout set enabled but not enough rows
self.logger.info(
"AutoMLSearch will use mean CV score to rank pipelines.",
)
else:
self.logger.info(
"AutoMLSearch will use the holdout set to score and rank pipelines.",
)
if self.data_splitter is not None and not issubclass(
self.data_splitter.__class__,
BaseCrossValidator,
):
raise ValueError("Not a valid data splitter")
default_data_splitter = make_data_splitter(
self.X_train,
self.y_train,
self.problem_type,
self.problem_configuration,
n_splits=3,
shuffle=True,
random_seed=self.random_seed,
)
self.data_splitter = self.data_splitter or default_data_splitter
self.search_parameters = search_parameters or {}
# Fitting takes a long time if the data is too wide or long.
if is_time_series(problem_type) and (
self.X_train.shape[1] >= ARIMARegressor.max_cols
or self.X_train.shape[0] >= ARIMARegressor.max_rows
):
user_arima_hyperparams = ARIMARegressor.name in self.search_parameters
if user_arima_hyperparams and not self.search_parameters[
ARIMARegressor.name
].get("use_covariates"):
self.search_parameters[ARIMARegressor.name].update(
{"use_covariates": Categorical([False])},
)
elif not user_arima_hyperparams:
self.search_parameters[ARIMARegressor.name] = {
"use_covariates": Categorical([False]),
}
def _is_imbalanced(X, y, problem_type):
if problem_type != ProblemTypes.MULTICLASS:
return False
imbalance_data_check = ClassImbalanceDataCheck()
results = imbalance_data_check.validate(X, y)
return bool(len(results))
recommendation_objectives = {}
if use_recommendation:
imbalanced = _is_imbalanced(self.X_train, self.y_train, self.problem_type)
default_objectives = get_default_recommendation_objectives(problem_type)
if include_recommendation is not None:
if not isinstance(include_recommendation, list):
raise ValueError(
"Objectives to include from the recommendation score should be a list",
)
for include_objective in include_recommendation:
include_objective = get_objective(include_objective)
if include_objective.name in default_objectives:
self.logger.warning(
f"Objective to include {include_objective} is already one of the default objectives being evaluated. No behavior will be changed.",
)
if exclude_recommendation is not None and not isinstance(
exclude_recommendation,
list,
):
raise ValueError(
"Objectives to exclude from the recommendation score should be a list",
)
recommendation_objectives = organize_objectives(
self.problem_type,
include_recommendation,
exclude_recommendation,
imbalanced,
)
self.use_recommendation = use_recommendation
self.recommendation_objectives = recommendation_objectives
if objective == "auto":
objective = get_default_primary_search_objective(self.problem_type.value)
objective = get_objective(objective, return_instance=False)
self.objective = self._validate_objective(objective)
self.alternate_thresholding_objective = None
if (
is_binary(self.problem_type)
and self.optimize_thresholds
and self.objective.score_needs_proba
):
self.alternate_thresholding_objective = get_objective(
alternate_thresholding_objective,
return_instance=True,
)
if (
self.alternate_thresholding_objective is not None
and self.alternate_thresholding_objective.score_needs_proba
):
raise ValueError(
"Alternate thresholding objective must be a tuneable objective and cannot need probabilities!",
)
if not objective.is_defined_for_problem_type(self.problem_type):
raise ValueError(
"Given objective {} is not compatible with a {} problem.".format(
self.objective.name,
self.problem_type.value,
),
)
if additional_objectives is None:
additional_objectives = get_optimization_objectives(self.problem_type)
# if our main objective is part of default set of objectives for problem_type, remove it
existing_main_objective = next(
(
obj
for obj in additional_objectives
if obj.name == self.objective.name
),
None,
)
if existing_main_objective is not None:
additional_objectives.remove(existing_main_objective)
else:
additional_objectives = [get_objective(o) for o in additional_objectives]
additional_objectives = [
self._validate_objective(obj) for obj in additional_objectives
]
additional_objective_names = [
objective.name for objective in additional_objectives
]
for recommendation_obj in self.recommendation_objectives:
if (
recommendation_obj not in additional_objective_names
and recommendation_obj != existing_main_objective.name
):
additional_objectives.append(get_objective(recommendation_obj))
self.additional_objectives = additional_objectives
self.objective_name_to_class = {
o.name: o for o in [self.objective] + self.additional_objectives
}
self._validate_problem_type()
self.search_iteration_plot = None
self._interrupted = False
internal_search_parameters = copy.copy(self.search_parameters)
if self.problem_configuration:
internal_search_parameters.update({"pipeline": self.problem_configuration})
self.features = features
if self.features is not None:
internal_search_parameters.update(
{"DFS Transformer": {"features": self.features}},
)
self.sampler_method = sampler_method
self.sampler_balanced_ratio = sampler_balanced_ratio
self._sampler_name = None
featurizer_names = [
"DatetimeFeaturizer",
"EmailFeaturizer",
"URLFeaturizer",
"NaturalLanguageFeaturizer",
"TimeSeriesFeaturizer",
]
if exclude_featurizers and (set(exclude_featurizers) - set(featurizer_names)):
raise ValueError(
f"Invalid value provided for exclude_featurizers. Must be one of: {', '.join(featurizer_names)}",
)
if exclude_featurizers and is_time_series(problem_type):
if (
"DatetimeFeaturizer" in exclude_featurizers
and "TimeSeriesFeaturizer" not in exclude_featurizers
):
raise ValueError(
"For time series problems, if DatetimeFeaturizer is excluded, must also exclude TimeSeriesFeaturizer",
)
elif (
"TimeSeriesFeaturizer" in exclude_featurizers
and "DatetimeFeaturizer" not in exclude_featurizers
):
raise ValueError(
"For time series problems, if TimeSeriesFeaturizer is excluded, must also exclude DatetimeFeaturizer",
)
self.exclude_featurizers = exclude_featurizers or []
self.run_feature_selection = run_feature_selection
if excluded_model_families:
if not isinstance(excluded_model_families, list):
raise ValueError(
"`excluded_model_families` must be passed in the form of a list.",
)
if not all(
isinstance(x, ModelFamily) or isinstance(x, str)
for x in excluded_model_families
):
raise ValueError(
"All values in `excluded_model_families` must be of type `ModelFamily` or `str`.",
)
self.excluded_model_families = excluded_model_families
if is_classification(self.problem_type):
self._sampler_name = self.sampler_method
if self.sampler_method == "auto":
self._sampler_name = get_best_sampler_for_data(
self.X_train,
self.y_train,
self.sampler_method,
self.sampler_balanced_ratio,
)
if (
self._sampler_name not in internal_search_parameters
and self._sampler_name is not None
):
internal_search_parameters[self._sampler_name] = {
"sampling_ratio": self.sampler_balanced_ratio,
}
elif self._sampler_name is not None:
internal_search_parameters[self._sampler_name].update(
{"sampling_ratio": self.sampler_balanced_ratio},
)
if isinstance(engine, str):
self._engine = build_engine_from_str(engine)
elif isinstance(engine, (DaskEngine, CFEngine, SequentialEngine)):
self._engine = engine
else:
raise TypeError(
"Invalid type provided for 'engine'. Requires string, DaskEngine instance, or CFEngine instance.",
)
self.automl_config = AutoMLConfig(
self.data_splitter,
self.problem_type,
self.objective,
self.additional_objectives,
self.alternate_thresholding_objective,
self.optimize_thresholds,
self.error_callback,
self.random_seed,
self.X_train.ww.schema,
self.y_train.ww.schema,
self.errors,
)
text_in_ensembling = (
len(self.X_train.ww.select("natural_language", return_schema=True).columns)
> 0
)
if automl_algorithm == "iterative":
self.automl_algorithm = IterativeAlgorithm(
X=self.X_train,
y=self.y_train,
problem_type=self.problem_type,
sampler_name=self._sampler_name,
allowed_component_graphs=self.allowed_component_graphs,
allowed_model_families=self.allowed_model_families,
excluded_model_families=self.excluded_model_families,
max_iterations=self.max_iterations,
max_batches=self.max_batches,
tuner_class=self.tuner_class,
random_seed=self.random_seed,
n_jobs=self.n_jobs,
number_features=self.X_train.shape[1],
pipelines_per_batch=self._pipelines_per_batch,
ensembling=self.ensembling,
text_in_ensembling=text_in_ensembling,
search_parameters=internal_search_parameters,
allow_long_running_models=allow_long_running_models,
features=features,
verbose=self.verbose,
exclude_featurizers=self.exclude_featurizers,
)
elif automl_algorithm == "default":
self.automl_algorithm = DefaultAlgorithm(
X=self.X_train,
y=self.y_train,
problem_type=self.problem_type,
sampler_name=self._sampler_name,
allowed_model_families=self.allowed_model_families,
excluded_model_families=self.excluded_model_families,
tuner_class=self.tuner_class,
random_seed=self.random_seed,
search_parameters=internal_search_parameters,
text_in_ensembling=text_in_ensembling,
allow_long_running_models=allow_long_running_models,
features=features,
run_feature_selection=run_feature_selection,
ensembling=self.ensembling,
verbose=self.verbose,
n_jobs=self.n_jobs,
exclude_featurizers=self.exclude_featurizers,
)
else:
raise ValueError("Please specify a valid automl algorithm.")
self.allowed_pipelines = self.automl_algorithm.allowed_pipelines
self.allowed_model_families = [p.model_family for p in self.allowed_pipelines]
if automl_algorithm == "iterative":
self.max_iterations = self.automl_algorithm.max_iterations
if not self.max_iterations and not self.max_time and not self.max_batches:
self.max_batches = self.automl_algorithm.default_max_batches
self.logger.info(
f"Using default limit of max_batches={self.max_batches}.\n",
)
self.progress = Progress(
max_time=self.max_time,
max_batches=self.max_batches,
max_iterations=self.max_iterations,
patience=self.patience,
tolerance=self.tolerance,
automl_algorithm=self.automl_algorithm,
objective=self.objective,
verbose=verbose,
)
[docs] def close_engine(self):
"""Function to explicitly close the engine, client, parallel resources."""
self._engine.close()
def _get_batch_number(self):
batch_number = 0
if self.automl_algorithm is not None and self.automl_algorithm.batch_number > 0:
batch_number = self.automl_algorithm.batch_number
return batch_number
def _pre_evaluation_callback(self, pipeline):
if self.start_iteration_callback:
self.start_iteration_callback(pipeline, self)
def _validate_objective(self, objective):
non_core_objectives = get_non_core_objectives()
if isinstance(objective, type):
if objective in non_core_objectives:
raise ValueError(
f"{objective.name.lower()} is not allowed in AutoML! "
"Use evalml.objectives.utils.get_optimization_objectives() "
"to get all objectives allowed for automl optimization.",
)
return objective()
return objective
def __str__(self):
"""Returns string representation of the AutoMLSearch object."""
def _print_list(obj_list):
lines = sorted(["\t{}".format(o.name) for o in obj_list])
return "\n".join(lines)
def _get_funct_name(function):
if callable(function):
return function.__name__
else:
return None
search_desc = (
f"{handle_problem_types(self.problem_type).name} Search\n\n"
f"Parameters: \n{'='*20}\n"
f"Objective: {get_objective(self.objective).name}\n"
f"Max Time: {self.max_time}\n"
f"Max Iterations: {self.max_iterations}\n"
f"Max Batches: {self.max_batches}\n"
f"Allowed Pipelines: \n{_print_list(self.allowed_pipelines or [])}\n"
f"Patience: {self.patience}\n"
f"Tolerance: {self.tolerance}\n"
f"Data Splitting: {self.data_splitter}\n"
f"Tuner: {self.tuner_class.__name__}\n"
f"Start Iteration Callback: {_get_funct_name(self.start_iteration_callback)}\n"
f"Add Result Callback: {_get_funct_name(self.add_result_callback)}\n"
f"Additional Objectives: {_print_list(self.additional_objectives or [])}\n"
f"Random Seed: {self.random_seed}\n"
f"n_jobs: {self.n_jobs}\n"
f"Optimize Thresholds: {self.optimize_thresholds}\n"
)
rankings_desc = ""
if not self.rankings.empty:
rankings_str = self.rankings.drop(
["parameters"],
axis="columns",
).to_string()
rankings_desc = f"\nSearch Results: \n{'='*20}\n{rankings_str}"
return search_desc + rankings_desc
def _validate_problem_configuration(self, problem_configuration=None):
if is_time_series(self.problem_type):
is_valid, msg = contains_all_ts_parameters(problem_configuration)
if not is_valid:
raise ValueError(msg)
if (
is_multiseries(self.problem_type)
and "series_id" not in problem_configuration
):
raise ValueError(
"Must provide 'series_id' column in problem_configuration for multiseries time series problems.",
)
return problem_configuration or {}
def _handle_keyboard_interrupt(self):
"""Presents a prompt to the user asking if they want to stop the search.
Returns:
bool: If True, search should terminate early.
"""
leading_char = "\n"
start_of_loop = time.time()
while True:
choice = (
input(leading_char + "Do you really want to exit search (y/n)? ")
.strip()
.lower()
)
if choice == "y":
self.logger.info("Exiting AutoMLSearch.")
return True
elif choice == "n":
# So that the time in this loop does not count towards the time budget (if set)
time_in_loop = time.time() - start_of_loop
self.progress.start_time += time_in_loop
return False
else:
leading_char = ""
[docs] def search(self, interactive_plot=True):
"""Find the best pipeline for the data set.
Args:
interactive_plot (boolean, True): Shows an iteration vs. score plot in Jupyter notebook.
Disabled by default in non-Jupyter enviroments.
Raises:
AutoMLSearchException: If all pipelines in the current AutoML batch produced a score of np.nan on the primary objective.
Returns:
Dict[int, Dict[str, Timestamp]]: Dictionary keyed by batch number that maps to the timings for pipelines run in that batch,
as well as the total time for each batch. Pipelines within a batch are labeled by pipeline name.
"""
batch_times = {}
if self._searched:
self.logger.error(
"AutoMLSearch.search() has already been run and will not run again on the same instance. Re-initialize AutoMLSearch to search again.",
)
return
# don't show iteration plot outside of a jupyter notebook
if interactive_plot:
try:
get_ipython
except NameError:
interactive_plot = False
log_title(self.logger, "Beginning pipeline search")
self.logger.info("Optimizing for %s. " % self.objective.name)
self.logger.info(
"{} score is better.\n".format(
"Greater" if self.objective.greater_is_better else "Lower",
),
)
self.logger.info(
f"Using {self._engine.__class__.__name__} to train and score pipelines.",
)
if self.max_batches is not None:
self.logger.info(
f"Searching up to {self.max_batches} batches for a total of {self.max_iterations} pipelines. ",
)
elif self.max_iterations is not None:
self.logger.info("Searching up to %s pipelines. " % self.max_iterations)
if self.max_time is not None:
self.logger.info(
"Will stop searching for new pipelines after %d seconds.\n"
% self.max_time,
)
self.logger.info(
"Allowed model families: %s\n"
% ", ".join([model.value for model in self.allowed_model_families]),
)
self.search_iteration_plot = None
if self.plot and self.verbose:
self.search_iteration_plot = self.plot.search_iteration_plot(
interactive_plot=interactive_plot,
)
self.progress.start_timing()
try:
self._add_baseline_pipelines()
except KeyboardInterrupt:
if self._handle_keyboard_interrupt():
self._interrupted = True
current_batch_pipelines = []
current_batch_pipeline_scores = []
new_pipeline_ids = []
loop_interrupted = False
while self.progress.should_continue(
results=self._results,
interrupted=self._interrupted,
):
pipeline_times = {}
start_batch_time = time.time()
computations = []
try:
if not loop_interrupted:
current_batch_pipelines = self.automl_algorithm.next_batch()
except StopIteration:
self.logger.info("AutoML Algorithm out of recommendations, ending")
break
try:
if self.progress.should_continue(
results=self._results,
interrupted=self._interrupted,
mid_batch=True,
):
new_pipeline_ids = []
log_title(
self.logger,
f"Evaluating Batch Number {self._get_batch_number()}",
)
for pipeline in current_batch_pipelines:
self._pre_evaluation_callback(pipeline)
computation = self._engine.submit_evaluation_job(
self.automl_config,
pipeline,
self.X_train,
self.y_train,
self.X_holdout,
self.y_holdout,
)
computations.append((computation, False))
current_computation_index = 0
computations_left_to_process = len(computations)
while (
self.progress.should_continue(
results=self._results,
interrupted=self._interrupted,
mid_batch=True,
)
and computations_left_to_process > 0
):
computation, has_been_processed = computations[
current_computation_index
]
if computation.done() and not has_been_processed:
start_pipeline_time = time.time()
evaluation = computation.get_result()
data, cached_data, pipeline, job_log = (
evaluation.get("scores"),
evaluation.get("cached_data"),
evaluation.get("pipeline"),
evaluation.get("logger"),
)
pipeline_id = self._post_evaluation_callback(
pipeline,
data,
cached_data,
job_log,
)
pipeline_times[pipeline.name] = (
time.time() - start_pipeline_time
)
new_pipeline_ids.append(pipeline_id)
computations[current_computation_index] = (computation, True)
computations_left_to_process -= 1
current_computation_index = (current_computation_index + 1) % max(
len(computations),
1,
)
time.sleep(self._sleep_time)
loop_interrupted = False
except KeyboardInterrupt:
loop_interrupted = True
if self._handle_keyboard_interrupt():
self._interrupted = True
for computation, has_been_processed in computations:
if not has_been_processed:
computation.cancel()
full_rankings = self.full_rankings
current_batch_idx = full_rankings["id"].isin(new_pipeline_ids)
current_batch_pipeline_scores = full_rankings[current_batch_idx][
"ranking_score"
]
if (
len(current_batch_pipeline_scores)
and current_batch_pipeline_scores.isna().all()
):
error_msgs = set(
[str(pl_fold["Exception"]) for pl_fold in self.errors.values()],
)
raise AutoMLSearchException(
f"All pipelines in the current AutoML batch produced a score of np.nan on the primary objective {self.objective}. Exception(s) raised: {error_msgs}. Check the 'errors' attribute of the AutoMLSearch object for a full breakdown of errors and tracebacks.",
)
if len(pipeline_times) > 0:
pipeline_times["Total time of batch"] = time.time() - start_batch_time
batch_times[self._get_batch_number()] = pipeline_times
self.search_duration = time.time() - self.progress.start_time
desc = f"\nSearch finished after {self.search_duration:.2f} seconds"
desc = desc.ljust(self._MAX_NAME_LEN)
self.logger.info(desc)
if self.timing is True:
log_batch_times(self.logger, batch_times)
self._find_best_pipeline()
if self._best_pipeline is not None:
best_pipeline = self.rankings.iloc[0]
best_pipeline_name = best_pipeline["pipeline_name"]
self.logger.info(f"Best pipeline: {best_pipeline_name}")
self.logger.info(
f"Best pipeline {self.objective.name}: {best_pipeline['ranking_score']:3f}",
)
self._searched = True
if self.search_iteration_plot is not None:
if self.verbose and not interactive_plot:
self.search_iteration_plot = self.plot.search_iteration_plot(
interactive_plot=interactive_plot,
)
if pio.renderers.default != "browser":
self.search_iteration_plot.show()
return batch_times
def _find_best_pipeline(self):
"""Finds the best pipeline in the rankings If self._best_pipeline already exists, check to make sure it is different from the current best pipeline before training and thresholding."""
if len(self.rankings) == 0:
return
best_pipeline = self.rankings.iloc[0]
if not (
self._best_pipeline
and self._best_pipeline == self.get_pipeline(best_pipeline["id"])
):
best_pipeline = self.get_pipeline(best_pipeline["id"])
if self._train_best_pipeline:
X_train = self.X_train
y_train = self.y_train
best_pipeline = self._engine.submit_training_job(
self.automl_config,
best_pipeline,
X_train,
y_train,
).get_result()[0]
self._best_pipeline = best_pipeline
def _num_pipelines(self):
"""Return the number of pipeline evaluations which have been made.
Returns:
int: The number of pipeline evaluations made in the search.
"""
return len(self._results["pipeline_results"])
def _validate_problem_type(self):
for obj in self.additional_objectives:
if not obj.is_defined_for_problem_type(self.problem_type):
raise ValueError(
"Additional objective {} is not compatible with a {} problem.".format(
obj.name,
self.problem_type.value,
),
)
def _get_baseline_pipeline(self):
"""Creates a baseline pipeline instance."""
classification_component_graph = {
"Label Encoder": ["Label Encoder", "X", "y"],
"Baseline Classifier": [
"Baseline Classifier",
"Label Encoder.x",
"Label Encoder.y",
],
}
if self.problem_type == ProblemTypes.BINARY:
baseline = BinaryClassificationPipeline(
component_graph=classification_component_graph,
custom_name="Mode Baseline Binary Classification Pipeline",
parameters={"Baseline Classifier": {"strategy": "mode"}},
)
elif self.problem_type == ProblemTypes.MULTICLASS:
baseline = MulticlassClassificationPipeline(
component_graph=classification_component_graph,
custom_name="Mode Baseline Multiclass Classification Pipeline",
parameters={"Baseline Classifier": {"strategy": "mode"}},
)
elif self.problem_type == ProblemTypes.REGRESSION:
baseline = RegressionPipeline(
component_graph=["Baseline Regressor"],
custom_name="Mean Baseline Regression Pipeline",
parameters={"Baseline Regressor": {"strategy": "mean"}},
)
else:
gap = self.problem_configuration["gap"]
forecast_horizon = self.problem_configuration["forecast_horizon"]
time_index = self.problem_configuration["time_index"]
series_id = self.problem_configuration.get("series_id", None)
exclude_timeseries_featurizer = (
"TimeSeriesFeaturizer" in self.exclude_featurizers
)
baseline = make_timeseries_baseline_pipeline(
self.problem_type,
gap,
forecast_horizon,
time_index,
exclude_timeseries_featurizer,
series_id,
)
return baseline
def _add_baseline_pipelines(self):
"""Fits a baseline pipeline to the data.
This is the first pipeline fit during search.
"""
baseline = self._get_baseline_pipeline()
self._pre_evaluation_callback(baseline)
self.logger.info(f"Evaluating Baseline Pipeline: {baseline.name}")
computation = self._engine.submit_evaluation_job(
self.automl_config,
baseline,
self.X_train,
self.y_train,
self.X_holdout,
self.y_holdout,
)
evaluation = computation.get_result()
data, cached_data, pipeline, job_log = (
evaluation.get("scores"),
evaluation.get("cached_data"),
evaluation.get("pipeline"),
evaluation.get("logger"),
)
self._post_evaluation_callback(pipeline, data, cached_data, job_log)
@staticmethod
def _get_mean_cv_scores_for_all_objectives(cv_data, objective_name_to_class):
scores = defaultdict(int)
n_folds = len(cv_data)
for fold_data in cv_data:
for field, value in fold_data["all_objective_scores"].items():
# The 'all_objective_scores' field contains scores for all objectives
# but also fields like "# Training" and "# Testing", so we want to exclude them since
# they are not scores
if field in objective_name_to_class:
scores[field] += value
return {
objective: float(score) / n_folds for objective, score in scores.items()
}
def _post_evaluation_callback(
self,
pipeline,
evaluation_results,
cached_data,
job_log,
):
job_log.write_to_logger(self.logger)
training_time = evaluation_results["training_time"]
cv_data = evaluation_results["cv_data"]
mean_cv_all_objectives = self._get_mean_cv_scores_for_all_objectives(
cv_data,
self.objective_name_to_class,
)
cv_scores = evaluation_results["cv_scores"]
is_baseline = pipeline.model_family == ModelFamily.BASELINE
mean_cv_score = np.nan if len(cv_scores) == 1 else cv_scores.mean()
holdout_score = evaluation_results["holdout_score"]
if holdout_score is None:
ranking_additional_objectives = mean_cv_all_objectives
if len(cv_scores) == 1:
ranking_score = cv_scores[0]
else:
ranking_score = mean_cv_score
else:
holdout_scores = evaluation_results["holdout_scores"]
ranking_additional_objectives = dict(holdout_scores)
ranking_score = holdout_score
cv_sd = cv_scores.std()
percent_better_than_baseline = {}
if is_baseline:
self._baseline_cv_scores = mean_cv_all_objectives
for obj_name in mean_cv_all_objectives:
objective_class = self.objective_name_to_class[obj_name]
# In the event add_to_rankings is called before search _baseline_cv_scores will be empty so we will return
# nan for the base score.
percent_better = objective_class.calculate_percent_difference(
mean_cv_all_objectives[obj_name],
self._baseline_cv_scores.get(obj_name, np.nan),
)
percent_better_than_baseline[obj_name] = percent_better
high_variance_cv = self._check_for_high_variance(pipeline, cv_scores)
pipeline_id = len(self._results["pipeline_results"])
self._results["pipeline_results"][pipeline_id] = {
"id": pipeline_id,
"pipeline_name": pipeline.name,
"pipeline_class": pipeline.__class__,
"pipeline_summary": pipeline.summary,
"parameters": pipeline.parameters,
"mean_cv_score": mean_cv_score,
"standard_deviation_cv_score": cv_sd,
"high_variance_cv": high_variance_cv,
"training_time": training_time,
"cv_data": cv_data,
"percent_better_than_baseline_all_objectives": percent_better_than_baseline,
"percent_better_than_baseline": percent_better_than_baseline[
self.objective.name
],
"ranking_score": ranking_score,
"ranking_additional_objectives": ranking_additional_objectives,
"holdout_score": holdout_score,
}
self._pipelines_searched.update({pipeline_id: pipeline.clone()})
if pipeline.model_family == ModelFamily.ENSEMBLE:
input_pipeline_ids = [
self.automl_algorithm._best_pipeline_info[model_family]["id"]
for model_family in self.automl_algorithm._best_pipeline_info
]
self._results["pipeline_results"][pipeline_id][
"input_pipeline_ids"
] = input_pipeline_ids
self._results["search_order"].append(pipeline_id)
if not is_baseline:
score_to_minimize = (
-ranking_score if self.objective.greater_is_better else ranking_score
)
try:
self.automl_algorithm.add_result(
score_to_minimize,
pipeline,
self._results["pipeline_results"][pipeline_id],
cached_data,
)
except PipelineNotFoundError:
pass
# True when running in a jupyter notebook, else the plot is an instance of plotly.Figure
if isinstance(self.search_iteration_plot, SearchIterationPlot):
self.search_iteration_plot.update(self.results, self.objective)
if self.add_result_callback:
self.add_result_callback(
self._results["pipeline_results"][pipeline_id],
pipeline,
self,
)
return pipeline_id
def _check_for_high_variance(self, pipeline, cv_scores, threshold=0.5):
"""Checks cross-validation scores and logs a warning if variance is higher than specified threshhold."""
pipeline_name = pipeline.name
high_variance_cv = False
allowed_range = (
self.objective.expected_range[1] - self.objective.expected_range[0]
)
if allowed_range == float("inf"):
return high_variance_cv
cv_range = max(cv_scores) - min(cv_scores)
if cv_range >= threshold * allowed_range:
self.logger.warning(
f"\tHigh coefficient of variation (cv >= {threshold}) within cross validation scores.\n\t{pipeline_name} may not perform as estimated on unseen data.",
)
high_variance_cv = True
return high_variance_cv
[docs] def get_pipeline(self, pipeline_id):
"""Given the ID of a pipeline training result, returns an untrained instance of the specified pipeline initialized with the parameters used to train that pipeline during automl search.
Args:
pipeline_id (int): Pipeline to retrieve.
Returns:
PipelineBase: Untrained pipeline instance associated with the provided ID.
Raises:
PipelineNotFoundError: if pipeline_id is not a valid ID.
"""
pipeline_results = self.results["pipeline_results"].get(pipeline_id)
if pipeline_results is None:
raise PipelineNotFoundError("Pipeline not found in automl results")
pipeline = self._pipelines_searched.get(pipeline_id)
parameters = pipeline_results.get("parameters")
if pipeline is None or parameters is None:
raise PipelineNotFoundError(
"Pipeline class or parameters not found in automl results",
)
new_pipeline = pipeline.new(parameters, random_seed=self.random_seed)
if is_binary(self.problem_type):
new_pipeline.threshold = None
return new_pipeline
[docs] def describe_pipeline(self, pipeline_id, return_dict=False):
"""Describe a pipeline.
Args:
pipeline_id (int): pipeline to describe
return_dict (bool): If True, return dictionary of information
about pipeline. Defaults to False.
Returns:
Description of specified pipeline. Includes information such as
type of pipeline components, problem, training time, cross validation, etc.
Raises:
PipelineNotFoundError: If pipeline_id is not a valid ID.
"""
logger = get_logger(f"{__name__}.describe_pipeline")
if pipeline_id not in self._results["pipeline_results"]:
raise PipelineNotFoundError("Pipeline not found")
pipeline = self.get_pipeline(pipeline_id)
pipeline_results = self._results["pipeline_results"][pipeline_id]
pipeline.describe()
if pipeline.model_family == ModelFamily.ENSEMBLE:
logger.info(
"Input for ensembler are pipelines with IDs: "
+ str(pipeline_results["input_pipeline_ids"]),
)
log_subtitle(logger, "Training")
logger.info("Training for {} problems.".format(pipeline.problem_type))
if (
self.optimize_thresholds
and self.objective.is_defined_for_problem_type(ProblemTypes.BINARY)
and self.objective.can_optimize_threshold
):
logger.info(
"Objective to optimize binary classification pipeline thresholds for: {}".format(
self.objective,
),
)
logger.info(
"Total training time (including CV): %.1f seconds"
% pipeline_results["training_time"],
)
log_subtitle(logger, "Cross Validation", underline="-")
all_objective_scores = [
fold["all_objective_scores"] for fold in pipeline_results["cv_data"]
]
all_objective_scores = pd.DataFrame(all_objective_scores)
for c in all_objective_scores:
if c in ["# Training", "# Validation"]:
all_objective_scores[c] = all_objective_scores[c].map(
lambda x: "{:2,.0f}".format(x) if not pd.isna(x) else np.nan,
)
continue
mean = all_objective_scores[c].mean(axis=0)
std = all_objective_scores[c].std(axis=0)
all_objective_scores.loc["mean", c] = mean
all_objective_scores.loc["std", c] = std
all_objective_scores.loc["coef of var", c] = (
std / mean if abs(mean) > 0 else np.inf
)
all_objective_scores = all_objective_scores.fillna("-")
with pd.option_context(
"display.float_format",
"{:.3f}".format,
"expand_frame_repr",
False,
):
logger.info(all_objective_scores)
if return_dict:
return pipeline_results
[docs] def add_to_rankings(self, pipeline):
"""Fits and evaluates a given pipeline then adds the results to the automl rankings with the requirement that automl search has been run.
Args:
pipeline (PipelineBase): pipeline to train and evaluate.
"""
pipeline_rows = self.full_rankings[
self.full_rankings["pipeline_name"] == pipeline.name
]
for parameter in pipeline_rows["parameters"]:
if pipeline.parameters == parameter:
return
computation = self._engine.submit_evaluation_job(
self.automl_config,
pipeline,
self.X_train,
self.y_train,
self.X_holdout,
self.y_holdout,
)
evaluation = computation.get_result()
data, cached_data, pipeline, job_log = (
evaluation.get("scores"),
evaluation.get("cached_data"),
evaluation.get("pipeline"),
evaluation.get("logger"),
)
self._post_evaluation_callback(pipeline, data, cached_data, job_log)
self._find_best_pipeline()
@property
def results(self):
"""Class that allows access to a copy of the results from `automl_search`.
Returns:
dict: Dictionary containing `pipeline_results`, a dict with results from each pipeline,
and `search_order`, a list describing the order the pipelines were searched.
"""
return copy.deepcopy(self._results)
@property
def rankings(self):
"""Returns a pandas.DataFrame with scoring results from the highest-scoring set of parameters used with each pipeline."""
return self.full_rankings.drop_duplicates(subset="pipeline_name", keep="first")
@property
def full_rankings(self):
"""Returns a pandas.DataFrame with scoring results from all pipelines searched."""
ascending = True
if self.objective.greater_is_better:
ascending = False
pipeline_results_cols = [
"id",
"pipeline_name",
"ranking_score",
]
if self.X_holdout is not None:
pipeline_results_cols += ["holdout_score"]
pipeline_results_cols += [
"mean_cv_score",
"standard_deviation_cv_score",
"percent_better_than_baseline",
"high_variance_cv",
"parameters",
]
if not self._results["pipeline_results"]:
full_rankings_cols = (
pipeline_results_cols[0:2]
+ ["search_order"]
+ pipeline_results_cols[2:]
) # place search_order after pipeline_name
return pd.DataFrame(columns=full_rankings_cols)
rankings_df = pd.DataFrame(self._results["pipeline_results"].values())
rankings_df = rankings_df[pipeline_results_cols]
rankings_df.insert(
2,
"search_order",
pd.Series(self._results["search_order"]),
) # place search_order after pipeline_name
ranking_column = "ranking_score"
if self.use_recommendation:
recommendation_scores = self.get_recommendation_scores()
ranking_column = "recommendation_score"
ascending = False
rankings_df.insert(
3,
"recommendation_score",
pd.Series(recommendation_scores.values()),
)
rankings_df.sort_values(ranking_column, ascending=ascending, inplace=True)
rankings_df.reset_index(drop=True, inplace=True)
return rankings_df
[docs] def get_recommendation_score_breakdown(self, pipeline_id):
"""Reports the scores for the objectives used in the given pipeline's recommendation score calculation.
Note that these scores are reported in their raw form, not scaled to be between 0 and 1.
Args:
pipeline_id (int): The id of the pipeline to get the recommendation score breakdown for.
Returns:
dict: A dictionary of the scores for each objective used in the recommendation score calculation.
"""
if len(self.recommendation_objectives) == 0:
self.recommendation_objectives = get_default_recommendation_objectives(
self.problem_type,
)
all_objectives = self._results["pipeline_results"][pipeline_id][
"ranking_additional_objectives"
]
rec_objectives = {
objective: all_objectives[objective]
for objective in self.recommendation_objectives
}
return rec_objectives
[docs] def get_recommendation_scores(
self,
priority=None,
custom_weights=None,
use_pipeline_names=False,
):
"""Calculates recommendation scores for all pipelines in the search results.
Args:
priority (str): An optional name of a priority objective that should be given heavier weight (of 0.5)
than the other objectives contributing to the score. Defaults to None, where all objectives are
weighted equally.
custom_weights (dict[str,float]): A dictionary mapping objective names to corresponding weights between 0 and 1.
Should not be used at the same time as prioritized_objective. Defaults to None.
use_pipeline_names (bool): Whether or not to return the pipeline names instead of ids as the keys
to the recommendation score dictionary. Defaults to False.
Returns:
A dictionary mapping pipeline IDs to recommendation scores
"""
def _get_scores_and_max_min(objectives_to_evaluate):
all_scores = {}
max_scores = {objective: 0 for objective in objectives_to_evaluate}
min_scores = {objective: 0 for objective in objectives_to_evaluate}
for pl_id, pl_results in self._results["pipeline_results"].items():
ranking_results = pl_results["ranking_additional_objectives"]
for objective in objectives_to_evaluate:
objective_obj = get_objective(objective)
if (
objective_obj.is_bounded_like_percentage
or objective_obj.name == "R2"
):
continue
max_scores[objective] = max(
max_scores[objective],
ranking_results[objective],
)
min_scores[objective] = min(
min_scores[objective],
ranking_results[objective],
)
all_scores[pl_id] = {
objective: ranking_results[objective]
for objective in objectives_to_evaluate
}
return all_scores, max_scores, min_scores
if len(self.recommendation_objectives) == 0:
self.recommendation_objectives = get_default_recommendation_objectives(
self.problem_type,
)
all_scores, max_scores, min_scores = _get_scores_and_max_min(
self.recommendation_objectives,
)
recommendation_scores = {}
for pipeline_id, pipeline_scores in all_scores.items():
rescaled_scores = normalize_objectives(
pipeline_scores,
max_scores,
min_scores,
)
score = recommendation_score(
rescaled_scores,
priority,
custom_weights,
)
recommendation_scores[pipeline_id] = score
if use_pipeline_names:
recommendation_scores = {
self.get_pipeline(pipeline_id).estimator.name: score
for pipeline_id, score in recommendation_scores.items()
}
return recommendation_scores
@property
def best_pipeline(self):
"""Returns a trained instance of the best pipeline and parameters found during automl search. If `train_best_pipeline` is set to False, returns an untrained pipeline instance.
Returns:
PipelineBase: A trained instance of the best pipeline and parameters found during automl search. If `train_best_pipeline` is set to False, returns an untrained pipeline instance.
Raises:
PipelineNotFoundError: If this is called before .search() is called.
"""
if not self._best_pipeline:
raise PipelineNotFoundError(
"automl search must be run before selecting `best_pipeline`.",
)
return self._best_pipeline
[docs] def save(
self,
file_path,
pickle_type="cloudpickle",
pickle_protocol=cloudpickle.DEFAULT_PROTOCOL,
):
"""Saves AutoML object at file path.
Args:
file_path (str): Location to save file.
pickle_type ({"pickle", "cloudpickle"}): The pickling library to use.
pickle_protocol (int): The pickle data stream format.
Raises:
ValueError: If pickle_type is not "pickle" or "cloudpickle".
"""
if pickle_type == "cloudpickle":
pkl_lib = cloudpickle
elif pickle_type == "pickle":
pkl_lib = pickle
else:
raise ValueError(
f"`pickle_type` must be either 'pickle' or 'cloudpickle'. Received {pickle_type}",
)
with open(file_path, "wb") as f:
pkl_lib.dump(self, f, protocol=pickle_protocol)
[docs] @staticmethod
def load(
file_path,
pickle_type="cloudpickle",
):
"""Loads AutoML object at file path.
Args:
file_path (str): Location to find file to load
pickle_type ({"pickle", "cloudpickle"}): The pickling library to use. Currently not used since the standard pickle library can handle cloudpickles.
Returns:
AutoSearchBase object
"""
with open(file_path, "rb") as f:
return pickle.load(f)
[docs] def train_pipelines(self, pipelines):
"""Train a list of pipelines on the training data.
This can be helpful for training pipelines once the search is complete.
Args:
pipelines (list[PipelineBase]): List of pipelines to train.
Returns:
Dict[str, PipelineBase]: Dictionary keyed by pipeline name that maps to the fitted pipeline.
Note that the any pipelines that error out during training will not be included in the dictionary
but the exception and stacktrace will be displayed in the log.
"""
check_all_pipeline_names_unique(pipelines)
fitted_pipelines = {}
computations = []
X_train = self.X_train
y_train = self.y_train
for pipeline in pipelines:
computations.append(
self._engine.submit_training_job(
self.automl_config,
pipeline,
X_train,
y_train,
),
)
while computations:
computation = computations.pop(0)
if computation.done():
try:
fitted_pipeline = computation.get_result()[0]
fitted_pipelines[fitted_pipeline.name] = fitted_pipeline
except Exception as e:
self.logger.error(f"Train error for {pipeline.name}: {str(e)}")
tb = traceback.format_tb(sys.exc_info()[2])
self.logger.error("Traceback:")
self.logger.error("\n".join(tb))
else:
computations.append(computation)
return fitted_pipelines
[docs] def score_pipelines(self, pipelines, X_holdout, y_holdout, objectives):
"""Score a list of pipelines on the given holdout data.
Args:
pipelines (list[PipelineBase]): List of pipelines to train.
X_holdout (pd.DataFrame): Holdout features.
y_holdout (pd.Series): Holdout targets for scoring.
objectives (list[str], list[ObjectiveBase]): Objectives used for scoring.
Returns:
dict[str, Dict[str, float]]: Dictionary keyed by pipeline name that maps to a dictionary of scores.
Note that the any pipelines that error out during scoring will not be included in the dictionary
but the exception and stacktrace will be displayed in the log.
"""
X_holdout, y_holdout = infer_feature_types(X_holdout), infer_feature_types(
y_holdout,
)
check_all_pipeline_names_unique(pipelines)
scores = {}
objectives = [get_objective(o, return_instance=True) for o in objectives]
computations = []
for pipeline in pipelines:
X_train, y_train = None, None
if is_time_series(self.problem_type):
X_train, y_train = self.X_train, self.y_train
computations.append(
self._engine.submit_scoring_job(
self.automl_config,
pipeline,
X_holdout,
y_holdout,
objectives,
X_train=X_train,
y_train=y_train,
),
)
while computations:
computation = computations.pop(0)
if computation.done():
pipeline_name = computation.meta_data["pipeline_name"]
try:
scores[pipeline_name] = computation.get_result()
except Exception as e:
self.logger.error(f"Score error for {pipeline_name}: {str(e)}")
if isinstance(e, PipelineScoreError):
nan_scores = {objective: np.nan for objective in e.exceptions}
scores[pipeline_name] = {**nan_scores, **e.scored_successfully}
else:
# Traceback already included in the PipelineScoreError so we only
# need to include it for all other errors
tb = traceback.format_tb(sys.exc_info()[2])
self.logger.error("Traceback:")
self.logger.error("\n".join(tb))
scores[pipeline_name] = {
objective.name: np.nan for objective in objectives
}
else:
computations.append(computation)
return scores
@property
def plot(self):
"""Return an instance of the plot with the latest scores."""
return PipelineSearchPlots(self.results, self.objective)
@property
def _sleep_time(self):
return self._SLEEP_TIME