vis4d.config.typing

Type definitions for configuration files.

Classes

DataConfig([initial_dictionary, type_safe, ...])

Configuration for a data set.

ExperimentConfig([initial_dictionary, ...])

Configuration for an experiment.

ExperimentParameters([initial_dictionary, ...])

Parameters for an experiment.

LrSchedulerConfig([initial_dictionary, ...])

Configuration for a learning rate scheduler.

OptimizerConfig([initial_dictionary, ...])

Configuration for an optimizer.

ParamGroupCfg

Parameter group config.

ParameterSweepConfig([initial_dictionary, ...])

Configuration for a parameter sweep.
class DataConfig(initial_dictionary=None, type_safe=True, convert_dict=True)[source]

Configuration for a data set.

This data object is used to configure the training and test data of an experiment. In particular, the train_dataloader and test_dataloader need to be config dicts that can be instantiated as a dataloader.

train_dataloader

Configuration for the training dataloader.

Type:

ConfigDict

test_dataloader

Configuration for the test dataloader.

Type:

ConfigDict

Example

>>> from vis4d.config.types import DataConfig
>>> from vis4d.zoo.base import class_config
>>> from my_package.data import MyDataLoader
>>> cfg = DataConfig()
>>> cfg.train_dataloader = class_config(MyDataLoader, ...)

Creates an instance of ConfigDict.

Warning: In most cases, this faithfully reproduces the reference structure of initial_dictionary, even if initial_dictionary is self-referencing. However, unexpected behavior occurs if self-references are contained within list, tuple, or custom types. For example:

d = {}
d['a'] = d
d['b'] = [d]
cd = ConfigDict(d)
cd.a    # refers to cd, type ConfigDict. Expected behavior.
cd.b    # refers to d, type dict. Unexpected behavior.

Warning: FieldReference values may be changed. If initial_dictionary contains a FieldReference with a value of type dict or FrozenConfigDict, that value is converted to ConfigDict.

Parameters:

  • initial_dictionary (Optional[Mapping[str, Any]]) –

    May be one of the following:

    1) dict. In this case, all values of initial_dictionary that are dictionaries are also be converted to ConfigDict. However, dictionaries within values of non-dict type are untouched.

    2) ConfigDict. In this case, all attributes are uncopied, and only the top-level object (self) is re-addressed. This is the same behavior as Python dict, list, and tuple.

    3) FrozenConfigDict. In this case, initial_dictionary is converted to a ConfigDict version of the initial dictionary for the FrozenConfigDict (reversing any mutability changes FrozenConfigDict made).

  • type_safe (bool) – If set to True, once an attribute value is assigned, its type cannot be overridden without .ignore_type() context manager (default: True).

  • convert_dict (bool) – If set to True, all dict used as value in the ConfigDict will automatically be converted to ConfigDict (default: True).

class ExperimentConfig(initial_dictionary=None, type_safe=True, convert_dict=True)[source]

Configuration for an experiment.

This data object is used to configure an experiment. It contains the minimal required configuration to run an experiment. In particular, the data, model, optimizers, and loss need to be config dicts that can be instantiated as a data set, model, optimizer, and loss function, respectively.

work_dir

The working directory for the experiment.

Type:

str | FieldReference

experiment_name

The name of the experiment.

Type:

str | FieldReference

timestamp

The timestamp of the experiment.

Type:

str | FieldReference

version

The version of the experiment.

Type:

str | FieldReference

output_dir

The output directory for the experiment.

Type:

str | FieldReference

seed

The random seed for the experiment.

Type:

int | FieldReference

log_every_n_steps

The number of steps after which the logs should be written.

Type:

int | FieldReference

use_tf32

Whether to use tf32.

Type:

bool | FieldReference

benchmark

Whether to enable benchmarking.

Type:

bool | FieldReference

params

Configuration for the experiment parameters.

Type:

ExperimentParameters

data

Configuration for the dataset.

Type:

DataConfig

model

Configuration for the model.

Type:

FieldConfigDictOrRef

loss

Configuration for the loss function.

Type:

FieldConfigDictOrRef

optimizers

Configuration for the optimizers.

Type:

list[OptimizerConfig]

data_connector

Configuration for the data connector.

Type:

FieldConfigDictOrRef

callbacks

Configuration for the callbacks which are used in the engine.

Type:

list[FieldConfigDictOrRef]

Creates an instance of FieldConfigDict.

Parameters:

  • initial_dictionary (Optional[Mapping[str, Any]]) –

    May be one of the following:

    1) dict. In this case, all values of initial_dictionary that are dictionaries are also be converted to ConfigDict. However, dictionaries within values of non-dict type are untouched.

    2) ConfigDict. In this case, all attributes are uncopied, and only the top-level object (self) is re-addressed. This is the same behavior as Python dict, list, and tuple.

    3) FrozenConfigDict. In this case, initial_dictionary is converted to a ConfigDict version of the initial dictionary for the FrozenConfigDict (reversing any mutability changes FrozenConfigDict made).

  • type_safe (bool) – If set to True, once an attribute value is assigned, its type cannot be overridden without .ignore_type() context manager.

  • convert_dict (bool) – If set to True, all dict used as value in the ConfigDict will automatically be converted to ConfigDict.

class ExperimentParameters(initial_dictionary=None, type_safe=True, convert_dict=True)[source]

Parameters for an experiment.

samples_per_gpu

Number of samples per GPU.

Type:

int

workers_per_gpu

Number of workers per GPU.

Type:

int

Creates an instance of FieldConfigDict.

Parameters:

  • initial_dictionary (Optional[Mapping[str, Any]]) –

    May be one of the following:

    1) dict. In this case, all values of initial_dictionary that are dictionaries are also be converted to ConfigDict. However, dictionaries within values of non-dict type are untouched.

    2) ConfigDict. In this case, all attributes are uncopied, and only the top-level object (self) is re-addressed. This is the same behavior as Python dict, list, and tuple.

    3) FrozenConfigDict. In this case, initial_dictionary is converted to a ConfigDict version of the initial dictionary for the FrozenConfigDict (reversing any mutability changes FrozenConfigDict made).

  • type_safe (bool) – If set to True, once an attribute value is assigned, its type cannot be overridden without .ignore_type() context manager.

  • convert_dict (bool) – If set to True, all dict used as value in the ConfigDict will automatically be converted to ConfigDict.

class LrSchedulerConfig(initial_dictionary=None, type_safe=True, convert_dict=True)[source]

Configuration for a learning rate scheduler.

scheduler

Configuration for the learning rate scheduler.

Type:

ConfigDict

begin

Begin epoch.

Type:

int

end

End epoch.

Type:

int

epoch_based

Whether the learning rate scheduler is epoch based or step based.

Type:

bool

convert_epochs_to_steps

Whether to convert the begin and end for a step based scheduler to steps automatically based on length of train dataloader. Enables users to set the iteration breakpoints as epochs. Defaults to False.

Type:

bool

convert_attributes

List of attributes in the scheduler that should be converted to steps. Defaults to None.

Type:

list[str] | None

Creates an instance of ConfigDict.

Warning: In most cases, this faithfully reproduces the reference structure of initial_dictionary, even if initial_dictionary is self-referencing. However, unexpected behavior occurs if self-references are contained within list, tuple, or custom types. For example:

d = {}
d['a'] = d
d['b'] = [d]
cd = ConfigDict(d)
cd.a    # refers to cd, type ConfigDict. Expected behavior.
cd.b    # refers to d, type dict. Unexpected behavior.

Warning: FieldReference values may be changed. If initial_dictionary contains a FieldReference with a value of type dict or FrozenConfigDict, that value is converted to ConfigDict.

Parameters:

  • initial_dictionary (Optional[Mapping[str, Any]]) –

    May be one of the following:

    1) dict. In this case, all values of initial_dictionary that are dictionaries are also be converted to ConfigDict. However, dictionaries within values of non-dict type are untouched.

    2) ConfigDict. In this case, all attributes are uncopied, and only the top-level object (self) is re-addressed. This is the same behavior as Python dict, list, and tuple.

    3) FrozenConfigDict. In this case, initial_dictionary is converted to a ConfigDict version of the initial dictionary for the FrozenConfigDict (reversing any mutability changes FrozenConfigDict made).

  • type_safe (bool) – If set to True, once an attribute value is assigned, its type cannot be overridden without .ignore_type() context manager (default: True).

  • convert_dict (bool) – If set to True, all dict used as value in the ConfigDict will automatically be converted to ConfigDict (default: True).

class OptimizerConfig(initial_dictionary=None, type_safe=True, convert_dict=True)[source]

Configuration for an optimizer.

optimizer

Configuration for the optimizer.

Type:

ConfigDict

lr_scheduler

Configuration for the learning rate scheduler.

Type:

list[LrSchedulerConfig] | None

param_groups

Configuration for the parameter groups.

Type:

list[ParamGroupCfg] | None

Creates an instance of ConfigDict.

Warning: In most cases, this faithfully reproduces the reference structure of initial_dictionary, even if initial_dictionary is self-referencing. However, unexpected behavior occurs if self-references are contained within list, tuple, or custom types. For example:

d = {}
d['a'] = d
d['b'] = [d]
cd = ConfigDict(d)
cd.a    # refers to cd, type ConfigDict. Expected behavior.
cd.b    # refers to d, type dict. Unexpected behavior.

Warning: FieldReference values may be changed. If initial_dictionary contains a FieldReference with a value of type dict or FrozenConfigDict, that value is converted to ConfigDict.

Parameters:

  • initial_dictionary (Optional[Mapping[str, Any]]) –

    May be one of the following:

    1) dict. In this case, all values of initial_dictionary that are dictionaries are also be converted to ConfigDict. However, dictionaries within values of non-dict type are untouched.

    2) ConfigDict. In this case, all attributes are uncopied, and only the top-level object (self) is re-addressed. This is the same behavior as Python dict, list, and tuple.

    3) FrozenConfigDict. In this case, initial_dictionary is converted to a ConfigDict version of the initial dictionary for the FrozenConfigDict (reversing any mutability changes FrozenConfigDict made).

  • type_safe (bool) – If set to True, once an attribute value is assigned, its type cannot be overridden without .ignore_type() context manager (default: True).

  • convert_dict (bool) – If set to True, all dict used as value in the ConfigDict will automatically be converted to ConfigDict (default: True).

class ParamGroupCfg[source]

Parameter group config.

custom_keys

List of custom keys.

Type:

list[str]

lr_mult

Learning rate multiplier.

Type:

NotRequired[float]

decay_mult

Weight Decay multiplier.

Type:

NotRequired[float]

class ParameterSweepConfig(initial_dictionary=None, type_safe=True, convert_dict=True)[source]

Configuration for a parameter sweep.

Confguration object for a parameter sweep. It contains the minimal required configuration to run a parameter sweep.

method

Sweep method that should be used (e.g. grid)

Type:

str

sampling_args

Arguments that should be passed to the sweep method. E.g. for grid, this would be a list of tuples of the form (parameter_name, parameter_values).

Type:

list[tuple[str, Any]]

suffix

Suffix that should be appended to the output directory. This will be interpreted as a string template and can contain references to the sampling_args. E.g. “lr_{lr:.2e}_bs_{batch_size}”.

Type:

str

Creates an instance of FieldConfigDict.

Parameters:

  • initial_dictionary (Optional[Mapping[str, Any]]) –

    May be one of the following:

    1) dict. In this case, all values of initial_dictionary that are dictionaries are also be converted to ConfigDict. However, dictionaries within values of non-dict type are untouched.

    2) ConfigDict. In this case, all attributes are uncopied, and only the top-level object (self) is re-addressed. This is the same behavior as Python dict, list, and tuple.

    3) FrozenConfigDict. In this case, initial_dictionary is converted to a ConfigDict version of the initial dictionary for the FrozenConfigDict (reversing any mutability changes FrozenConfigDict made).

  • type_safe (bool) – If set to True, once an attribute value is assigned, its type cannot be overridden without .ignore_type() context manager.

  • convert_dict (bool) – If set to True, all dict used as value in the ConfigDict will automatically be converted to ConfigDict.