neuraxle.metaopt.data.vanilla

Module-level documentation for neuraxle.metaopt.data.vanilla. Here is an inheritance diagram, including dependencies to other base modules of Neuraxle:

Neuraxle’s Base Hyperparameter Repository Classes

Data objects and related repositories used by AutoML.

Classes are splitted like this for the AutoML: - Projects - Clients - Rounds (runs) - Trials - TrialSplits - MetricResults

Functions

as_named_odict(obj, …)
from_json(_json)
object_decoder(obj[, odictify])
object_pairs_decoder(obj)
to_json(obj)

Classes

AutoMLContext(root, flow, execution_phase, …)
BaseDataclass(*args, **kwds)
BaseHyperparameterOptimizer
BaseTrialDataclassMixin(hyperparams, status, …)	Mixin class for TrialMetadata and TrialSplitMetadata that also must inherit from BaseMetadata.
ClientDataclass(*args, **kwds)
DataclassHasListMixin()
DataclassHasOrderedDictMixin()
HyperparameterSamplerStub(…)
HyperparamsRepository()	Hyperparams repository that saves hyperparams, and scores for every AutoML trial.
InMemoryHyperparamsRepository(*kargs, **kwargs)	In memory hyperparams repository that can print information about trials.
MetadataJSONEncoder(*[, skipkeys, …])
MetricResultsDataclass(metric_name, …)	MetricResult object used by AutoML algorithm classes.
ProjectDataclass(*args, **kwds)
RootDataclass(*args, **kwds)
RoundDataclass(*args, **kwds)
ScopedLocation(project_name, client_name, …)	A location in the metadata tree.
TrialDataclass(hyperparams, status, …)	This class is a data structure most often used under AutoML to store information about a trial.
TrialSplitDataclass(hyperparams, status, …)	TrialSplit object used by AutoML algorithm classes.
VanillaHyperparamsRepository(cache_folder)	Hyperparams repository that saves data AutoML-related info.

class neuraxle.metaopt.data.vanilla.ScopedLocation(project_name: str = None, client_name: str = None, round_number: int = None, trial_number: int = None, split_number: int = None, metric_name: str = None)

Bases: neuraxle.base.BaseService

A location in the metadata tree.

project_name = None

client_name = None

round_number = None

trial_number = None

split_number = None

metric_name = None

copy() → neuraxle.metaopt.data.vanilla.ScopedLocation

Returns a copy of the ScopedLocation.

with_dc(dc: neuraxle.metaopt.data.vanilla.BaseDataclass) → neuraxle.metaopt.data.vanilla.ScopedLocation

Returns a new ScopedLocation with the provided BaseDataclass (dc) type’s id added.

with_id(_id: Union[str, int]) → neuraxle.metaopt.data.vanilla.ScopedLocation

Returns a longer ScopedLocation with the provided id added at the end.

at_dc(dc: neuraxle.metaopt.data.vanilla.BaseDataclass) → neuraxle.metaopt.data.vanilla.ScopedLocation

Returns a trimmed ScopedLocation with the provided BaseDataclass (dc) type’s id as the ScopedLocation’s deepest attribute.

static default_full() → neuraxle.metaopt.data.vanilla.ScopedLocation

Returns a ScopedLocation with all attributes set to the default non-null value instead of None.

static default(round_number: Optional[int] = None, trial_number: Optional[int] = None, split_number: Optional[int] = None, metric_name: Optional[str] = None) → neuraxle.metaopt.data.vanilla.ScopedLocation

Returns the default ScopedLocation. That is:

peek() → Union[str, int]

Pop without removing the last element: return the last non-None element.

pop() → Union[str, int]

Returns the last not None scoped location attribute and remove it from self.

popped() → neuraxle.metaopt.data.vanilla.ScopedLocation

Returns a new ScopedLocation with the last not None scoped location attribute removed.

as_list(stringify: bool = False) → List[Union[int, str]]

Returns a list of the scoped location attributes. Item that has a value of None are not included in the list.

Parameters

stringify (bool) – If True, the scoped location attributes are converted to strings.

Returns

list of not none scoped location attributes

new_dataclass_from_id()

Creates a new BaseDataclass of the right type with just the provided ID filled.

last_dc_type() → Type[neuraxle.metaopt.data.vanilla.BaseDataclass]

__init__(project_name: str = None, client_name: str = None, round_number: int = None, trial_number: int = None, split_number: int = None, metric_name: str = None) → None

_abc_impl = <_abc_data object>

class neuraxle.metaopt.data.vanilla.BaseDataclass(*args, **kwds)

Bases: typing.Generic, abc.ABC

_validate()

static from_dict(_dict: Dict[str, Any]) → neuraxle.metaopt.data.vanilla.BaseDataclass

to_dict() → Dict[str, Any]

_id_attr_name

_sublocation_attr_name

get_id() → Union[str, int]

set_id(_id: Union[str, int]) → neuraxle.metaopt.data.vanilla.BaseDataclass

get_sublocation() → Union[List[SubDataclassT], OrderedDict[str, SubDataclassT]]

set_sublocation(sublocation: Union[List[SubDataclassT], OrderedDict[str, SubDataclassT]]) → BaseDataclass

set_sublocation_keys(keys: List[Union[int, str]]) → neuraxle.metaopt.data.vanilla.BaseDataclass

Use this to set a shallow sublocation only from their keys.

_is_key_in_subattr(key_subattr)

store(dc: SubDataclassT) → Union[str, int]

Add a subdataclass to the sublocation, at its proper ID.

shallow() → neuraxle.metaopt.data.vanilla.BaseDataclass

Replaces the sublocation items with None when the sublocation is a BaseDataclass type.

empty() → neuraxle.metaopt.data.vanilla.BaseDataclass

Do empty the sublocation when the sublocation is a BaseDataclass type.

has_sublocation_dataclasses() → bool

get_sublocation_values() → List[SubDataclassT]

get_sublocation_keys() → List[Union[int, str]]

classmethod subdataclass_type() → Type[SubDataclassT]

__init__() → None

_abc_impl = <_abc_data object>

class neuraxle.metaopt.data.vanilla.DataclassHasOrderedDictMixin

Bases: object

_validate()

get_sublocation() → OrderedDict[ScopedLocationAttrStr, SubDataclassT]

set_sublocation(sublocation: OrderedDict[ScopedLocationAttrStr, SubDataclassT]) → None

set_sublocation_keys(keys: List[Union[int, str]]) → neuraxle.metaopt.data.vanilla.BaseDataclass

get_sublocation_values() → List[SubDataclassT]

get_sublocation_keys() → List[str]

store(dc: SubDataclassT) → str

shallow() → neuraxle.metaopt.data.vanilla.BaseDataclass

empty() → neuraxle.metaopt.data.vanilla.BaseDataclass

__init__() → None

class neuraxle.metaopt.data.vanilla.DataclassHasListMixin

Bases: object

_validate()

get_sublocation() → List[SubDataclassT]

set_sublocation(sublocation: List[SubDataclassT]) → None

set_sublocation_keys(keys: List[Union[int, str]]) → neuraxle.metaopt.data.vanilla.BaseDataclass

get_sublocation_values() → List[SubDataclassT]

get_sublocation_keys() → List[int]

_is_key_in_subattr(key_subattr)

store(dc: SubDataclassT) → int

shallow() → neuraxle.metaopt.data.vanilla.BaseDataclass

empty() → neuraxle.metaopt.data.vanilla.BaseDataclass

get_next_i() → int

__init__() → None

class neuraxle.metaopt.data.vanilla.BaseTrialDataclassMixin(hyperparams: neuraxle.hyperparams.space.HyperparameterSamples = <factory>, status: neuraxle.base.TrialStatus = <TrialStatus.PLANNED: 'PLANNED'>, created_time: datetime.datetime = <factory>, start_time: datetime.datetime = <factory>, end_time: datetime.datetime = None)

Bases: object

Mixin class for TrialMetadata and TrialSplitMetadata that also must inherit from BaseMetadata.

status = 'PLANNED'

end_time = None

_validate()

_validate_date_attr(attr_name) → None

start() → neuraxle.metaopt.data.vanilla.BaseTrialDataclassMixin

end(status: neuraxle.base.TrialStatus) → neuraxle.metaopt.data.vanilla.BaseTrialDataclassMixin

__init__(hyperparams: neuraxle.hyperparams.space.HyperparameterSamples = <factory>, status: neuraxle.base.TrialStatus = <TrialStatus.PLANNED: 'PLANNED'>, created_time: datetime.datetime = <factory>, start_time: datetime.datetime = <factory>, end_time: datetime.datetime = None) → None

class neuraxle.metaopt.data.vanilla.RootDataclass(*args, **kwds)

Bases: neuraxle.metaopt.data.vanilla.DataclassHasOrderedDictMixin, neuraxle.metaopt.data.vanilla.BaseDataclass

_id_attr_name

_sublocation_attr_name

get_id() → Union[str, int]

get_sublocation() → OrderedDict[str, SubDataclassT]

__init__(projects: OrderedDict[str, ProjectDataclass] = <factory>) → None

_abc_impl = <_abc_data object>

class neuraxle.metaopt.data.vanilla.ProjectDataclass(*args, **kwds)

Bases: neuraxle.metaopt.data.vanilla.DataclassHasOrderedDictMixin, neuraxle.metaopt.data.vanilla.BaseDataclass

project_name = 'default_project'

__init__(project_name: str = 'default_project', clients: OrderedDict[str, ClientDataclass] = <factory>) → None

_abc_impl = <_abc_data object>

class neuraxle.metaopt.data.vanilla.ClientDataclass(*args, **kwds)

Bases: neuraxle.metaopt.data.vanilla.DataclassHasListMixin, neuraxle.metaopt.data.vanilla.BaseDataclass

client_name = 'default_client'

__init__(client_name: str = 'default_client', rounds: List[RoundDataclass] = <factory>) → None

_abc_impl = <_abc_data object>

class neuraxle.metaopt.data.vanilla.RoundDataclass(*args, **kwds)

Bases: neuraxle.metaopt.data.vanilla.DataclassHasListMixin, neuraxle.metaopt.data.vanilla.BaseDataclass

round_number = 0

main_metric_name = None

__init__(round_number: int = 0, trials: List[TrialDataclass] = <factory>, main_metric_name: str = None) → None

_abc_impl = <_abc_data object>

class neuraxle.metaopt.data.vanilla.TrialDataclass(hyperparams: neuraxle.hyperparams.space.HyperparameterSamples = <factory>, status: neuraxle.base.TrialStatus = <TrialStatus.PLANNED: 'PLANNED'>, created_time: datetime.datetime = <factory>, start_time: datetime.datetime = <factory>, end_time: datetime.datetime = None, trial_number: int = 0, validation_splits: List[TrialSplitDataclass] = <factory>, retrained_split: TrialSplitDataclass = None)

Bases: neuraxle.metaopt.data.vanilla.DataclassHasListMixin, neuraxle.metaopt.data.vanilla.BaseTrialDataclassMixin, neuraxle.metaopt.data.vanilla.BaseDataclass

This class is a data structure most often used under AutoML to store information about a trial. This information is itself managed by the HyperparameterRepository class and the Trial class within the AutoML.

trial_number = 0

retrained_split = None

store(dc: SubDataclassT) → int

Add a subdataclass to the sublocation, at its proper ID.

set_sublocation_keys(keys: List[Union[int, str]]) → neuraxle.metaopt.data.vanilla.BaseDataclass

Use this to set a shallow sublocation only from their keys.

shallow() → neuraxle.metaopt.data.vanilla.BaseDataclass

Replaces the sublocation items with None when the sublocation is a BaseDataclass type.

__init__(hyperparams: neuraxle.hyperparams.space.HyperparameterSamples = <factory>, status: neuraxle.base.TrialStatus = <TrialStatus.PLANNED: 'PLANNED'>, created_time: datetime.datetime = <factory>, start_time: datetime.datetime = <factory>, end_time: datetime.datetime = None, trial_number: int = 0, validation_splits: List[TrialSplitDataclass] = <factory>, retrained_split: TrialSplitDataclass = None) → None

_abc_impl = <_abc_data object>

class neuraxle.metaopt.data.vanilla.TrialSplitDataclass(hyperparams: neuraxle.hyperparams.space.HyperparameterSamples = <factory>, status: neuraxle.base.TrialStatus = <TrialStatus.PLANNED: 'PLANNED'>, created_time: datetime.datetime = <factory>, start_time: datetime.datetime = <factory>, end_time: datetime.datetime = None, split_number: int = 0, metric_results: OrderedDict[str, MetricResultsDataclass] = <factory>)

Bases: neuraxle.metaopt.data.vanilla.DataclassHasOrderedDictMixin, neuraxle.metaopt.data.vanilla.BaseTrialDataclassMixin, neuraxle.metaopt.data.vanilla.BaseDataclass

TrialSplit object used by AutoML algorithm classes.

split_number = 0

is_retrain_split() → bool

__init__(hyperparams: neuraxle.hyperparams.space.HyperparameterSamples = <factory>, status: neuraxle.base.TrialStatus = <TrialStatus.PLANNED: 'PLANNED'>, created_time: datetime.datetime = <factory>, start_time: datetime.datetime = <factory>, end_time: datetime.datetime = None, split_number: int = 0, metric_results: OrderedDict[str, MetricResultsDataclass] = <factory>) → None

_abc_impl = <_abc_data object>

class neuraxle.metaopt.data.vanilla.MetricResultsDataclass(metric_name: str = 'main', validation_values: List[float] = <factory>, train_values: List[float] = <factory>, higher_score_is_better: bool = True)

Bases: neuraxle.metaopt.data.vanilla.DataclassHasListMixin, neuraxle.metaopt.data.vanilla.BaseDataclass

MetricResult object used by AutoML algorithm classes.

metric_name = 'main'

higher_score_is_better = True

shallow() → neuraxle.metaopt.data.vanilla.BaseDataclass

Replaces the sublocation items with None when the sublocation is a BaseDataclass type.

empty() → neuraxle.metaopt.data.vanilla.BaseDataclass

Do empty the sublocation when the sublocation is a BaseDataclass type.

__init__(metric_name: str = 'main', validation_values: List[float] = <factory>, train_values: List[float] = <factory>, higher_score_is_better: bool = True) → None

_abc_impl = <_abc_data object>

neuraxle.metaopt.data.vanilla.as_named_odict(obj: Union[neuraxle.metaopt.data.vanilla.BaseDataclass, List[neuraxle.metaopt.data.vanilla.BaseDataclass]]) → OrderedDict[ScopedLocationAttrStr, BaseDataclass]

neuraxle.metaopt.data.vanilla.object_pairs_decoder(obj)

neuraxle.metaopt.data.vanilla.object_decoder(obj, odictify=False)

class neuraxle.metaopt.data.vanilla.MetadataJSONEncoder(*, skipkeys=False, ensure_ascii=True, check_circular=True, allow_nan=True, sort_keys=False, indent=None, separators=None, default=None)

Bases: json.encoder.JSONEncoder

default(o)

Implement this method in a subclass such that it returns a serializable object for o, or calls the base implementation (to raise a TypeError).

For example, to support arbitrary iterators, you could implement default like this:

def default(self, o):
    try:
        iterable = iter(o)
    except TypeError:
        pass
    else:
        return list(iterable)
    # Let the base class default method raise the TypeError
    return JSONEncoder.default(self, o)

neuraxle.metaopt.data.vanilla.to_json(obj: neuraxle.metaopt.data.vanilla.BaseDataclass) → str

neuraxle.metaopt.data.vanilla.from_json(_json: str) → neuraxle.metaopt.data.vanilla.BaseDataclass

class neuraxle.metaopt.data.vanilla.HyperparamsRepository

Bases: neuraxle.metaopt.observable._ObservableRepo, neuraxle.base.BaseService

Hyperparams repository that saves hyperparams, and scores for every AutoML trial. Cache folder can be changed to do different round numbers.

For more information, read this article by Martin Fowler on DDD Aggregates.

See also

AutoML, Trainer,

__init__()

Initialize self. See help(type(self)) for accurate signature.

load(scope: neuraxle.metaopt.data.vanilla.ScopedLocation, deep=False) → SubDataclassT

Get metadata from scope.

The fetched metadata will be the one that is the last item that is not a None in the provided scope.

Parameters

scope (ScopedLocation) – scope to get metadata from.

Returns

metadata from scope.

save(_dataclass: SubDataclassT, scope: neuraxle.metaopt.data.vanilla.ScopedLocation, deep=False) → neuraxle.metaopt.data.vanilla.HyperparamsRepository

Save metadata to scope.

Parameters

• metadata – metadata to save.

• scope (ScopedLocation) – scope to save metadata to.

• deep (bool) – if True, save metadata’s sublocations recursively.

add_logging_handler(logger: neuraxle.logging.logging.NeuraxleLogger, scope: neuraxle.metaopt.data.vanilla.ScopedLocation) → neuraxle.metaopt.data.vanilla.HyperparamsRepository

Add logging handler to repository’s provided scope. Handler must be set manually for each scope below this scope.

Parameters

• logger (NeuraxleLogger) – logger to add handler to.

• scope (ScopedLocation) – scope to add handler to.

Returns

self.

get_log_from_logging_handler(logger: neuraxle.logging.logging.NeuraxleLogger, scope: neuraxle.metaopt.data.vanilla.ScopedLocation) → str

Get log from repository’s provided scope and handler that was set with add_logging_handler().

Return type

str

Parameters

scope (ScopedLocation) – scope to get log from.

Returns

log from scope.

_abc_impl = <_abc_data object>

class neuraxle.metaopt.data.vanilla._InMemoryRepositoryLoggerHandlerMixin

Bases: object

Mixin to add a in-memory logging handler to a repository.

add_logging_handler(logger: neuraxle.logging.logging.NeuraxleLogger, scope: neuraxle.metaopt.data.vanilla.ScopedLocation) → neuraxle.metaopt.data.vanilla.HyperparamsRepository

get_log_from_logging_handler(logger: neuraxle.logging.logging.NeuraxleLogger, scope: neuraxle.metaopt.data.vanilla.ScopedLocation) → str

class neuraxle.metaopt.data.vanilla.VanillaHyperparamsRepository(cache_folder: str)

Bases: neuraxle.metaopt.data.vanilla._InMemoryRepositoryLoggerHandlerMixin, neuraxle.metaopt.data.vanilla.HyperparamsRepository

Hyperparams repository that saves data AutoML-related info.

__init__(cache_folder: str)

Parameters

• cache_folder (str) – folder to store trials.

• hyperparams_repo_class – class to use to save hyperparams.

• hyperparams_repo_kwargs – kwargs to pass to hyperparams_repo_class.

static from_root(root: neuraxle.metaopt.data.vanilla.RootDataclass, cache_folder: str) → neuraxle.metaopt.data.vanilla.VanillaHyperparamsRepository

load(scope: neuraxle.metaopt.data.vanilla.ScopedLocation, deep=False) → SubDataclassT

Get metadata from scope.

The fetched metadata will be the one that is the last item that is not a None in the provided scope.

Parameters

scope (ScopedLocation) – scope to get metadata from.

Returns

metadata from scope.

save(_dataclass: SubDataclassT, scope: neuraxle.metaopt.data.vanilla.ScopedLocation, deep=False) → neuraxle.metaopt.data.vanilla.VanillaHyperparamsRepository

Save metadata to scope.

Parameters

• metadata – metadata to save.

• scope (ScopedLocation) – scope to save metadata to.

• deep (bool) – if True, save metadata’s sublocations recursively.

_abc_impl = <_abc_data object>

class neuraxle.metaopt.data.vanilla.InMemoryHyperparamsRepository(*kargs, **kwargs)

Bases: neuraxle.logging.warnings.RaiseDeprecatedClass, neuraxle.metaopt.data.vanilla._InMemoryRepositoryLoggerHandlerMixin, neuraxle.metaopt.data.vanilla.HyperparamsRepository

In memory hyperparams repository that can print information about trials. Useful for debugging.

__init__(*kargs, **kwargs)

Initialize self. See help(type(self)) for accurate signature.

_abc_impl = <_abc_data object>

class neuraxle.metaopt.data.vanilla.BaseHyperparameterOptimizer

Bases: abc.ABC

find_next_best_hyperparams(round_scope) → neuraxle.hyperparams.space.HyperparameterSamples

Find the next best hyperparams using previous trials, that is the whole neuraxle.metaopt.data.aggregate.Round.

Return type

HyperparameterSamples

Parameters

round – a neuraxle.metaopt.data.aggregate.Round

Returns

next hyperparameter samples to train on

_abc_impl = <_abc_data object>

class neuraxle.metaopt.data.vanilla.HyperparameterSamplerStub(preconfigured_hp_samples: neuraxle.hyperparams.space.HyperparameterSamples)

Bases: neuraxle.metaopt.data.vanilla.BaseHyperparameterOptimizer

__init__(preconfigured_hp_samples: neuraxle.hyperparams.space.HyperparameterSamples)

Initialize self. See help(type(self)) for accurate signature.

find_next_best_hyperparams(round_scope) → neuraxle.hyperparams.space.HyperparameterSamples

Find the next best hyperparams using previous trials, that is the whole neuraxle.metaopt.data.aggregate.Round.

Return type

HyperparameterSamples

Parameters

round – a neuraxle.metaopt.data.aggregate.Round

Returns

next hyperparameter samples to train on

_abc_impl = <_abc_data object>

class neuraxle.metaopt.data.vanilla.AutoMLContext(root: str = None, flow: neuraxle.base.Flow = None, execution_phase: neuraxle.base.ExecutionPhase = <ExecutionPhase.UNSPECIFIED: None>, execution_mode: neuraxle.base.ExecutionMode = <ExecutionMode.FIT_OR_FIT_TRANSFORM_OR_TRANSFORM: 'fit_or_fit_transform_or_transform'>, stripped_saver: neuraxle.base.BaseSaver = None, parents: List[BaseStep] = None, services: Dict[Union[str, Type[BaseServiceT]], BaseServiceT] = None)

Bases: neuraxle.base.ExecutionContext

logger

logger_at_scoped_loc

add_scoped_logger_file_handler() → neuraxle.logging.logging.NeuraxleLogger

Add a file handler to the logger at the current scoped location to capture logs at this scope and below this scope.

free_scoped_logger_file_handler()

Remove file handlers from logger to free file lock (especially on Windows).

read_scoped_log() → str

Read the scoped logger file.

copy()

_get_copy_kwargs()

static from_context(context: neuraxle.base.ExecutionContext = None, repo: neuraxle.metaopt.data.vanilla.HyperparamsRepository = None, loc: neuraxle.metaopt.data.vanilla.ScopedLocation = None) → neuraxle.metaopt.data.vanilla.AutoMLContext

Create a new AutoMLContext from an ExecutionContext.

Parameters

context (ExecutionContext) – ExecutionContext

lock

loc

repo

push_attr(subdataclass: SubDataclassT) → neuraxle.metaopt.data.vanilla.AutoMLContext

Push a new attribute into the ScopedLocation.

Parameters

• name – attribute name

• value – attribute value

Returns

an AutoMLContext copy with the new loc attribute.

pop_attr() → neuraxle.metaopt.data.vanilla.AutoMLContext

Pop an attribute from the ScopedLocation.

Returns

an AutoMLContext copy with the new popped loc attribute.

with_loc(loc: neuraxle.metaopt.data.vanilla.ScopedLocation) → neuraxle.metaopt.data.vanilla.AutoMLContext

Replace the ScopedLocation by the one provided.

Parameters

loc (ScopedLocation) – ScopedLocation

Returns

an AutoMLContext copy with the new loc attribute.

load_dc(deep=True) → neuraxle.metaopt.data.vanilla.BaseDataclass

Load the current dc from the repo.

load_agg(deep=True) → neuraxle.metaopt.data.aggregates.BaseAggregate

Load the current agg from the repo.

_abc_impl = <_abc_data object>