tfaip.scenario

ScenarioBase

Definition of ScenarioBase

tfaip.scenario.scenariobase.list_scenarios_in_module(module) → List[Tuple[str, Type[tfaip.scenario.scenariobase.ScenarioBase]]]

class tfaip.scenario.scenariobase.KerasModelData(extended_loss_names: List[str], extended_metric_names: List[str], tensorboard_output_names: List[str])

Bases: object

Utility structure to hold shared information about the model

extended_loss_names: List[str]

extended_metric_names: List[str]

tensorboard_output_names: List[str]

__init__(extended_loss_names: List[str], extended_metric_names: List[str], tensorboard_output_names: List[str]) → None

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

classmethod from_dict(kvs: Optional[Union[dict, list, str, int, float, bool]], *, infer_missing=False) → A

classmethod from_json(s: Union[str, bytes, bytearray], *, parse_float=None, parse_int=None, parse_constant=None, infer_missing=False, **kw) → A

classmethod schema(*, infer_missing: bool = False, only=None, exclude=(), many: bool = False, context=None, load_only=(), dump_only=(), partial: bool = False, unknown=None) → dataclasses_json.mm.SchemaF[A]

to_dict(encode_json=False) → Dict[str, Optional[Union[dict, list, str, int, float, bool]]]

to_json(*, skipkeys: bool = False, ensure_ascii: bool = True, check_circular: bool = True, allow_nan: bool = True, indent: Optional[Union[int, str]] = None, separators: Optional[Tuple[str, str]] = None, default: Optional[Callable] = None, sort_keys: bool = False, **kw) → str

class tfaip.scenario.scenariobase.ScenarioBaseMeta(*args, **kwargs)

Bases: tfaip.util.generic_meta.CollectGenericTypes

class tfaip.scenario.scenariobase.ScenarioBase(params: TScenarioParams)

Bases: Generic[tfaip.scenario.scenariobase.TData, tfaip.scenario.scenariobase.TModel, tfaip.scenario.scenariobase.TScenarioParams, tfaip.scenario.scenariobase.TTrainerPipelineParams], abc.ABC

The scenario base handles the setup of the scenario including training and model exporting.

Several abstract methods must be overridden that define how to create the Model, and Data for this scenario. If your scenario requires different default params in contrast to the values defined in the respective dataclasses, override default_params(). Furthermore it is possible to change the Trainer or LAV class which is usually not required!

A typical Scenario only requires to implement get_meta() and get_param_cls() To set scenario specific (hidden) parameters use the subclasses __init__ or override create_model() or create_data(). Note that create_model() will be called after create_data()!

The metaclass of the Scenario (ScenarioBaseMeta) will track the types of the Generics which is then use to create instances of the actual types automatically (see, e.g., model_cls(), data_cls(), …).

data

After setup() was called, the DataBase of the scenario

Type

DataBase

model

After setup() was called, the ModelBase of the scenario

Type

ModelBase

classmethod default_params() → TScenarioParams

Override to change the default parameters for this scenario.

Returns: ScenarioParams with adapted defaults of the respective Scenario

classmethod default_trainer_params() → TrainerParams[TScenarioParams, TTrainerPipelineParams]

Override to change the default trainer params for this scenario.

Use this to override hyperparameters such as the learning rate.

Returns: TrainerParams with adapted defaults

classmethod params_from_dict(d: dict) → TScenarioParams

Instantiate ScenarioParams from a dict. Not that since all dataclasses are a pai_dataclass the actual ScenarioParams will be read from the __cls__ arg within the dict.

See also

ScenarioBaseParams.from_dict

Parameters

d – trainer or scenario params dict

Returns: ScenarioParams of the respective Scenario

classmethod params_from_path(path: str) → TScenarioParams

Read params from a file path and construct the stored ScenarioParams

See also

params_from_dict, read_params_dict_from_path

Parameters

path – path to the file or to the directory containin a trainer_params.json or scenario_params.json

Returns

Parsed ScenarioParams

classmethod read_params_dict_from_path(path: str) → dict

Read a params dict from a path (or dictionary)

See also

params_from_path

Parameters

path – path to the file or to the directory containin a trainer_params.json or scenario_params.json

Returns

dict of ScenarioParams

classmethod from_path(path: str) → Tuple[Type[tfaip.scenario.scenariobase.ScenarioBase], TScenarioParams]

Instantiate a scenario from a path.

Parameters

path – Path to the scenario_params.json or dictionary containing them

Returns

Tuple of the actual Scenario class and the parsed Scenario params

classmethod from_dict(d: dict) → Tuple[Type[tfaip.scenario.scenariobase.ScenarioBase], TScenarioParams]

Instantiate a scenario from a dict.

Parameters

d – Dict of the scenario_params.json or dictionary containing them

Returns

Tuple of the actual Scenario class and the parsed Scenario params

classmethod data_cls() → Type[TData]

classmethod model_cls() → Type[TModel]

classmethod trainer_cls() → Type[Trainer[TrainerParams]]

classmethod lav_cls() → Type[LAV]

classmethod multi_lav_cls() → Type[tfaip.lav.multilav.MultiLAV]

classmethod predictor_cls() → Type[Predictor]

classmethod evaluator_cls() → Type[tfaip.evaluator.evaluator.EvaluatorBase]

classmethod multi_predictor_cls() → Type[MultiModelPredictor]

classmethod create_trainer(trainer_params: TrainerParams, restore=False) → Trainer

classmethod create_lav(lav_params: LAVParams, scenario_params: TScenarioParams) → LAV

classmethod create_multi_lav(lav_params: LAVParams, scenario_params: TScenarioParams, predictor_params: Optional[PredictorParams] = None)

classmethod create_predictor(model: str, params: PredictorParams) → Predictor

classmethod create_multi_predictor(paths: List[str], params: PredictorParams) → MultiModelPredictor

classmethod create_evaluator(params: tfaip.evaluator.params.EvaluatorParams) → tfaip.evaluator.evaluator.EvaluatorBase

classmethod params_cls() → Type[TScenarioParams]

classmethod trainer_pipeline_params_cls() → Type[TTrainerPipelineParams]

classmethod predict_generator_params_cls() → Type[tfaip.data.databaseparams.DataGeneratorParams]

static net_config_cls() → Type[tfaip.scenario.scenariobaseparams.NetConfigParamsBase]

__init__(params: TScenarioParams)

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

property params

property keras_train_model

property keras_predict_model

setup()

create_data() → TData

create_model() → TModel

print_params() → NoReturn

Print the ScenarioParams to logger.info

best_logging_settings() → Tuple[str, str]

Returns: See ModelBase.best_logging_settings()

export(path: str, trainer_params: Optional[TrainerParams] = None, export_resources: bool = True) → NoReturn

Export the prediction model to a given path

Parameters

• path – Directory where to export the model to

• trainer_params – if given export the full trainer params instead of the scenario params

• export_resources – Set to true (default) to also export the resources

setup_training(optimizer, skip_model_load_test=False, run_eagerly=False, no_train_scope: Optional[str] = None) → NoReturn

Set training by constructing the training and prediction keras Models.

Parameters

• optimizer – The optimizer to use for training

• skip_model_load_test – Skip the test checking if the prediction model can be stored and loaded

• run_eagerly – Run the model in eager mode

• no_train_scope – Regex to match layers to exclude from training

fit(epochs, callbacks, steps_per_epoch, initial_epoch=0, **kwargs) → NoReturn

Train the scenario.

This function will create the training and optionally validation pipelines, and call fit of the keras model.

Parameters

• epochs – The number of epochs to train

• callbacks – Callbacks of training

• steps_per_epoch – Number of steps per epoch

• initial_epoch – Initial epoch (use if resuming the training)

• **kwargs – Additional args passed to keras.Model.fit

net_config() → tfaip.scenario.scenariobaseparams.NetConfigParamsBase

Create the NetConfigParams for this scenario.

Usually a custom Scenario should override net_config_cls() and _fill_net_config() instead of overwriting this function.

Returns

NetConfigParamsBase

tfaip.scenario.scenariobase.import_scenario(module_name: str) → Type[tfaip.scenario.scenariobase.ScenarioBase]

module_name can either be: - module and class name separated by a ‘:’, e.g.: tfaip.scenarios.tutorial.min.scenario:TutorialScenario - module containing one class inheriting ScenarioBase, e.g.: tfaip.scenarios.tutorial.min.scenario - module containing containing a scenario module that contains one class inheriting ScenarioBase, e.g.: tfaip.scenarios.tutorial.min

ScenarioBaseParams

Definition of the ScenarioBaseParams

class tfaip.scenario.scenariobaseparams.ScenarioBaseParamsMeta(*args, **kwargs)

Bases: tfaip.util.generic_meta.ReplaceDefaultDataClassFieldsMeta

class tfaip.scenario.scenariobaseparams.ScenarioBaseParams(debug_graph_construction: bool = False, debug_graph_n_examples: int = 1, print_eval_limit: int = 10, tensorboard_logger_history_size: int = 5, export_serve: bool = True, model: TModelParams = <factory>, data: TDataParams = <factory>, evaluator: tfaip.evaluator.params.EvaluatorParams = <factory>, export_net_config: bool = True, net_config_filename: str = 'net_config.json', default_serve_dir: str = 'serve', additional_serve_dir: str = 'additional', trainer_params_filename: str = 'trainer_params.json', scenario_params_filename: str = 'scenario_params.json', scenario_base_path: Optional[str] = None, scenario_id: Optional[str] = None, id: Optional[str] = None, tfaip_commit_hash: str = <factory>, tfaip_version: str = '1.1.1')

Bases: Generic[tfaip.scenario.scenariobaseparams.TDataParams, tfaip.scenario.scenariobaseparams.TModelParams], abc.ABC

Define the global params of a scenario contains model_params and data_params of the model

NOTE: add @pai_dataclass and @dataclass annotations to inherited class

debug_graph_construction: bool = False

debug_graph_n_examples: int = 1

print_eval_limit: int = 10

tensorboard_logger_history_size: int = 5

export_serve: bool = True

model: TModelParams

data: TDataParams

evaluator: tfaip.evaluator.params.EvaluatorParams

export_net_config: bool = True

net_config_filename: str = 'net_config.json'

default_serve_dir: str = 'serve'

additional_serve_dir: str = 'additional'

trainer_params_filename: str = 'trainer_params.json'

scenario_params_filename: str = 'scenario_params.json'

scenario_base_path: Optional[str] = None

scenario_id: Optional[str] = None

id: Optional[str] = None

tfaip_commit_hash: str

tfaip_version: str = '1.1.1'

cls() → Type[ScenarioBase]

create() → ScenarioBase

__init__(debug_graph_construction: bool = False, debug_graph_n_examples: int = 1, print_eval_limit: int = 10, tensorboard_logger_history_size: int = 5, export_serve: bool = True, model: TModelParams = <factory>, data: TDataParams = <factory>, evaluator: tfaip.evaluator.params.EvaluatorParams = <factory>, export_net_config: bool = True, net_config_filename: str = 'net_config.json', default_serve_dir: str = 'serve', additional_serve_dir: str = 'additional', trainer_params_filename: str = 'trainer_params.json', scenario_params_filename: str = 'scenario_params.json', scenario_base_path: Optional[str] = None, scenario_id: Optional[str] = None, id: Optional[str] = None, tfaip_commit_hash: str = <factory>, tfaip_version: str = '1.1.1') → None

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

classmethod from_dict(kvs: Optional[Union[dict, list, str, int, float, bool]], *, infer_missing=False) → A

classmethod from_json(s: Union[str, bytes, bytearray], *, parse_float=None, parse_int=None, parse_constant=None, infer_missing=False, **kw) → A

classmethod schema(*, infer_missing: bool = False, only=None, exclude=(), many: bool = False, context=None, load_only=(), dump_only=(), partial: bool = False, unknown=None) → dataclasses_json.mm.SchemaF[A]

to_dict(encode_json=False) → Dict[str, Optional[Union[dict, list, str, int, float, bool]]]

to_json(*, skipkeys: bool = False, ensure_ascii: bool = True, check_circular: bool = True, allow_nan: bool = True, indent: Optional[Union[int, str]] = None, separators: Optional[Tuple[str, str]] = None, default: Optional[Callable] = None, sort_keys: bool = False, **kw) → str

class tfaip.scenario.scenariobaseparams.NetConfigNodeSpec(shape: List[str], dtype: str, node_serve: str)

Bases: object

shape: List[str]

dtype: str

node_serve: str

__init__(shape: List[str], dtype: str, node_serve: str) → None

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

class tfaip.scenario.scenariobaseparams.NetConfigParamsBase(id_model: str, id_serve: str, in_nodes: Dict[str, tfaip.scenario.scenariobaseparams.NetConfigNodeSpec], out_nodes: Dict[str, tfaip.scenario.scenariobaseparams.NetConfigNodeSpec], tf_version: str, tfaip_commit_hash: str = <factory>, tfaip_version: str = '1.1.1')

Bases: object

id_model: str

id_serve: str

in_nodes: Dict[str, tfaip.scenario.scenariobaseparams.NetConfigNodeSpec]

out_nodes: Dict[str, tfaip.scenario.scenariobaseparams.NetConfigNodeSpec]

tf_version: str

tfaip_commit_hash: str

tfaip_version: str = '1.1.1'

__init__(id_model: str, id_serve: str, in_nodes: Dict[str, tfaip.scenario.scenariobaseparams.NetConfigNodeSpec], out_nodes: Dict[str, tfaip.scenario.scenariobaseparams.NetConfigNodeSpec], tf_version: str, tfaip_commit_hash: str = <factory>, tfaip_version: str = '1.1.1') → None

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

classmethod from_dict(kvs: Optional[Union[dict, list, str, int, float, bool]], *, infer_missing=False) → A

classmethod from_json(s: Union[str, bytes, bytearray], *, parse_float=None, parse_int=None, parse_constant=None, infer_missing=False, **kw) → A

classmethod schema(*, infer_missing: bool = False, only=None, exclude=(), many: bool = False, context=None, load_only=(), dump_only=(), partial: bool = False, unknown=None) → dataclasses_json.mm.SchemaF[A]

to_dict(encode_json=False) → Dict[str, Optional[Union[dict, list, str, int, float, bool]]]

to_json(*, skipkeys: bool = False, ensure_ascii: bool = True, check_circular: bool = True, allow_nan: bool = True, indent: Optional[Union[int, str]] = None, separators: Optional[Tuple[str, str]] = None, default: Optional[Callable] = None, sort_keys: bool = False, **kw) → str

.listfile

ListFileScenario

Definition of a ListFileScenario

class tfaip.scenario.listfile.listfilescenario.ListFileScenario(params: TScenarioParams)

Bases: tfaip.scenario.scenariobase.ScenarioBase[tfaip.scenario.scenariobase.TData, tfaip.scenario.scenariobase.TModel, tfaip.scenario.scenariobase.TScenarioParams, tfaip.scenario.listfile.params.ListFileTrainerPipelineParams]

Base-Class for a Scenario working with list files. A list file is a simple text file where each line is the path to a sample. The ListFileScenario uses ListFileTrainerPipelineParams which create a DataGenerator that will yield each line as a Sample’s input and target. This must then be processed by DataProcessors.

classmethod lav_cls()

Parameters

Definition of the parameters for a ListFileScenario

class tfaip.scenario.listfile.params.ListsFileGeneratorParams(lists: Optional[List[str]] = <factory>, list_ratios: Optional[List[float]] = None)

Bases: tfaip.data.databaseparams.DataGeneratorParams

Parameters for the ListsFileDataGenerator

static cls()

lists: Optional[List[str]]

list_ratios: Optional[List[float]] = None

__init__(lists: Optional[List[str]] = <factory>, list_ratios: Optional[List[float]] = None) → None

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

classmethod from_dict(kvs: Optional[Union[dict, list, str, int, float, bool]], *, infer_missing=False) → A

classmethod from_json(s: Union[str, bytes, bytearray], *, parse_float=None, parse_int=None, parse_constant=None, infer_missing=False, **kw) → A

classmethod schema(*, infer_missing: bool = False, only=None, exclude=(), many: bool = False, context=None, load_only=(), dump_only=(), partial: bool = False, unknown=None) → dataclasses_json.mm.SchemaF[A]

to_dict(encode_json=False) → Dict[str, Optional[Union[dict, list, str, int, float, bool]]]

to_json(*, skipkeys: bool = False, ensure_ascii: bool = True, check_circular: bool = True, allow_nan: bool = True, indent: Optional[Union[int, str]] = None, separators: Optional[Tuple[str, str]] = None, default: Optional[Callable] = None, sort_keys: bool = False, **kw) → str

class tfaip.scenario.listfile.params.ListFileTrainerPipelineParams(setup: tfaip.trainer.params.TrainerPipelines = <factory>, train: TDataGeneratorTrain = <factory>, val: TDataGeneratorVal = <factory>)

Bases: tfaip.trainer.params.TrainerPipelineParams[tfaip.scenario.listfile.params.ListsFileGeneratorParams, tfaip.scenario.listfile.params.ListsFileGeneratorParams]

Implemented TrainerPipelineParams that replaces the defaults of train and val to ListFileGeneratorParams.

__init__(setup: tfaip.trainer.params.TrainerPipelines = <factory>, train: TDataGeneratorTrain = <factory>, val: TDataGeneratorVal = <factory>) → None

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

classmethod from_dict(kvs: Optional[Union[dict, list, str, int, float, bool]], *, infer_missing=False) → A

classmethod from_json(s: Union[str, bytes, bytearray], *, parse_float=None, parse_int=None, parse_constant=None, infer_missing=False, **kw) → A

classmethod schema(*, infer_missing: bool = False, only=None, exclude=(), many: bool = False, context=None, load_only=(), dump_only=(), partial: bool = False, unknown=None) → dataclasses_json.mm.SchemaF[A]

to_dict(encode_json=False) → Dict[str, Optional[Union[dict, list, str, int, float, bool]]]

to_json(*, skipkeys: bool = False, ensure_ascii: bool = True, check_circular: bool = True, allow_nan: bool = True, indent: Optional[Union[int, str]] = None, separators: Optional[Tuple[str, str]] = None, default: Optional[Callable] = None, sort_keys: bool = False, **kw) → str

train: TDataGeneratorTrain

val: TDataGeneratorVal

Data Generator

Definition of the ListFileDataGenerator

class tfaip.scenario.listfile.datagenerator.ListsFileDataGenerator(mode: tfaip.data.pipeline.definitions.PipelineMode, params: T)

Bases: tfaip.data.pipeline.datagenerator.DataGenerator[tfaip.scenario.listfile.params.ListsFileGeneratorParams]

DataGenerator of ListFiles, a text file where each line is a sample.

On training, random samples (file names) will be chosen respecting the list rations (if multiple lists are passes), see ListMixDefinition.

On any other mode, the samples will be yielded in Order.

generate() → Iterable[tfaip.data.pipeline.definitions.Sample]